Natural language processing for optimized extractive summarization

BACKGROUND

Various embodiments of the present invention address technical challenges related to performing extractive summarization, such as extractive summarization of multi-party interaction transcripts. Various embodiments of the present invention address the shortcomings of natural language summarization systems and disclose various techniques for efficiently and reliably performing extractive summarization, such as extractive summarization of multi-party interaction transcripts.

BRIEF SUMMARY

In general, embodiments of the present invention provide methods, apparatus, systems, computing devices, computing entities, and/or the like for performing extractive summarization, such as extractive summarization of multi-party interaction transcripts. Certain embodiments of the present invention utilize systems, methods, and computer program products that perform extractive summarization of multi-party interaction transcripts by using per-party utterance subsets, integer linear programming (ILP) operations configured to perform constrained optimization of an overall summary utility measure across eligible extractive summaries, and using information quality measures and/or linguistic quality measures of interaction utterances to determine utility measures for extractive summaries.

In accordance with one aspect, a method is provided. In one embodiment, the method comprises: identifying one or more per-party utterance subsets in the multi-party call transcript, wherein each per-party utterance subset of the one or more per-party utterances comprises a related subset of the plurality of interaction utterances that is associated with an interaction party of a plurality of interaction parties; generating a plurality of eligible extractive summaries, wherein: (i) each eligible extractive summary of the plurality of eligible comprises a covered subset of the plurality of interaction utterances that complies with one or more optimization constraints, and (ii) the one or more optimization constraints comprises a similarity-based optimization constraint requiring that, if the covered subset for a particular eligible extractive summary of the plurality of extractive summaries comprises a particular interaction utterance of the plurality of interaction utterances that is in a particular per-party utterance subset of the one or more per-party utterances subsets, then the covered subset for the particular eligible extractive summary should further comprise each other interaction utterance of the plurality of interaction utterances that is in any per-party utterance subset of the one or more per-party utterances subsets other than the particular per-party utterance subset and that has a threshold-satisfying utterance similarity measure with respect to the particular interaction utterance; for each eligible extractive summary of the plurality of eligible extractive summaries, determining an overall summary utility measure; generating the optimal extractive summary based at least in part on each overall summary utility measure for an eligible extractive summary of the plurality of eligible extractive summaries; and performing one or more summary-based actions based at least in part on the optimal extractive summary.

In accordance with another aspect, a computer program product is provided. The computer program product may comprise at least one computer-readable storage medium having computer-readable program code portions stored therein, the computer-readable program code portions comprising executable portions configured to: identify one or more per-party utterance subsets in the multi-party call transcript, wherein each per-party utterance subset of the one or more per-party utterances comprises a related subset of the plurality of interaction utterances that is associated with an interaction party of a plurality of interaction parties; generate a plurality of eligible extractive summaries, wherein: (i) each eligible extractive summary of the plurality of eligible comprises a covered subset of the plurality of interaction utterances that complies with one or more optimization constraints, and (ii) the one or more optimization constraints comprises a similarity-based optimization constraint requiring that, if the covered subset for a particular eligible extractive summary of the plurality of extractive summaries comprises a particular interaction utterance of the plurality of interaction utterances that is in a particular per-party utterance subset of the one or more per-party utterances subsets, then the covered subset for the particular eligible extractive summary should further comprise each other interaction utterance of the plurality of interaction utterances that is in any per-party utterance subset of the one or more per-party utterances subsets other than the particular per-party utterance subset and that has a threshold-satisfying utterance similarity measure with respect to the particular interaction utterance; for each eligible extractive summary of the plurality of eligible extractive summaries, determine an overall summary utility measure; generating the optimal extractive summary based at least in part on each overall summary utility measure for an eligible extractive summary of the plurality of eligible extractive summaries; and perform one or more summary-based actions based at least in part on the optimal extractive summary.

In accordance with yet another aspect, an apparatus comprising at least one processor and at least one memory including computer program code is provided. In one embodiment, the at least one memory and the computer program code may be configured to, with the processor, cause the apparatus to: identify one or more per-party utterance subsets in the multi-party call transcript, wherein each per-party utterance subset of the one or more per-party utterances comprises a related subset of the plurality of interaction utterances that is associated with an interaction party of a plurality of interaction parties; generate a plurality of eligible extractive summaries, wherein: (i) each eligible extractive summary of the plurality of eligible comprises a covered subset of the plurality of interaction utterances that complies with one or more optimization constraints, and (ii) the one or more optimization constraints comprises a similarity-based optimization constraint requiring that, if the covered subset for a particular eligible extractive summary of the plurality of extractive summaries comprises a particular interaction utterance of the plurality of interaction utterances that is in a particular per-party utterance subset of the one or more per-party utterances subsets, then the covered subset for the particular eligible extractive summary should further comprise each other interaction utterance of the plurality of interaction utterances that is in any per-party utterance subset of the one or more per-party utterances subsets other than the particular per-party utterance subset and that has a threshold-satisfying utterance similarity measure with respect to the particular interaction utterance; for each eligible extractive summary of the plurality of eligible extractive summaries, determine an overall summary utility measure; generating the optimal extractive summary based at least in part on each overall summary utility measure for an eligible extractive summary of the plurality of eligible extractive summaries; and perform one or more summary-based actions based at least in part on the optimal extractive summary.

BRIEF DESCRIPTION OF THE DRAWINGS

Having thus described the invention in general terms, reference will now be made to the accompanying drawings, which are not necessarily drawn to scale, and wherein:

<figref idref="DRAWINGS">FIG. 1</figref> provides an exemplary overview of an architecture that can be used to practice embodiments of the present invention.

<figref idref="DRAWINGS">FIG. 2</figref> provides an example predictive data analysis computing entity in accordance with some embodiments discussed herein.

<figref idref="DRAWINGS">FIG. 3</figref> provides an example external computing entity in accordance with some embodiments discussed herein.

<figref idref="DRAWINGS">FIG. 4</figref> is a flowchart diagram of an example process for generating an optimal extractive summary of a multi-party interaction transcript in accordance with some embodiments discussed herein.

<figref idref="DRAWINGS">FIG. 5</figref> provides an operational example of a multi-party interaction transcript in accordance with some embodiments discussed herein.

<figref idref="DRAWINGS">FIG. 6</figref> is a flowchart diagram of an example process for generating an optimal extractive summary of a multi-party interaction transcript in accordance with detected per-party utterances in accordance with some embodiments discussed herein.

<figref idref="DRAWINGS">FIG. 7</figref> is a flowchart diagram of an example process for performing interlinear programming (ILP) to generate an optimal extractive summary of a multi-party interaction transcript in accordance with some embodiments discussed herein.

<figref idref="DRAWINGS">FIG. 8</figref> provides an operational example of a summary output user interface in accordance with some embodiments discussed herein.

DETAILED DESCRIPTION

Various embodiments of the present invention now will be described more fully hereinafter with reference to the accompanying drawings, in which some, but not all, embodiments of the inventions are shown. Indeed, these inventions may be embodied in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will satisfy applicable legal requirements. The term “or” is used herein in both the alternative and conjunctive sense, unless otherwise indicated. The terms “illustrative” and “exemplary” are used to be examples with no indication of quality level. Like numbers refer to like elements throughout. Moreover, while certain embodiments of the present invention are described with reference to predictive data analysis, one of ordinary skill in the art will recognize that the disclosed concepts can be used to perform other types of data analysis.

I. Overview

Aspects of the present invention present an automatic extractive summarization method of multi-party (e.g., two-party) transcript data along with extraction of content information that utilizes unsupervised machine learning techniques. A proposed method uses an ILP formulation to select sentences from utterances of each party in a transcript to form an extractive summary. The proposed method also leverages the similarity in dialogues between two or more parties to select the sentences in the summary. Content words may also be detected by formulating a cost function as a part of a cost function that is included in the ILP task formulation. Manual and automatic evaluation results demonstrate the success of the proposed method in achieving high scores in terms of both readability and informativeness of extracted summaries. Some aspects of the present invention rely on extracting relevant information in the form of a summary and getting content words from a transcript to understand and improve customer experience, boost sales, reduce customer churn, and streamline the efficiency of customer service processes.

According to some aspects of the present invention, a proposed method weighs each utterance in a transcript by measuring the utterance's information quality and linguistic quality, and generates an extractive summarization for each party, where each party's utterances involved in a transcript, may in some embodiments be treated as a separate cluster. According to some embodiments, it is assumed that most of the information during a conversation is centered around a limited set of special content words and a dialogue exchanged between parties are more or less similar in context. Aspects of the proposed solution focus on the coverage of important content words across party utterances and computation of similarity among the utterances involved in the call transcript. The machine learning models utilized by aspects of the present invention to perform extractive summarization are fully unsupervised and do not rely on any annotations. The needed input for such models may simply consist of a list of utterances for two or more parties without any metadata. In doing so, the noted aspects of the present invention present extractive summarization solutions that are both reliable, computationally efficient, and can be trained without using annotated training data.

Exemplary inventive and technologically advantageous aspects of the present invention include the following aspects: (i) using an ILP-based summarization solution that treats utterances by different parties separately and is thus able to infer predictive insights based at least in part on cross-party interactions and cross-party relationships between utterances, (ii) selecting utterances to include in the summary by maximizing weight of content words as well as informativeness and linguistic quality of utterances, which in turn improves the accuracy and reliability of the generated summary, and (iii) utilizing an unsupervised approach that partitions the utterances in per-party utterance subsets and extracting summaries using informativeness and linguistic quality of the noted per-party utterance subsets.

II. Definitions

The term “multi-party interaction transcript” may refer to a data entity that is configured to describe a temporal flow of verbal interactions between two or more interaction parties. An example of a multi-party interaction transcript is a call transcript between two or more parties to the call, such as a call transcript for a call between a customer service agent and a customer. In the noted example, the call transcript may describe verbal interactions by the parties in a temporally sequential manner, where each verbal interaction by a party may include one or more utterances (e.g., on or more sentences). For example, with respect to the call transcript for a call between a customer service agent and a customer, the call transcript may describe that a first set of utterances by the customer service agent (e.g., “Hello, how is your day today and how can I help you?”) is temporally followed by a second set of utterances by the customer (e.g., “Thank you, doing well, I am trying to check my account balance.”), which may then be temporally followed by a third set of utterances by the customer service agent, and so on. Other examples of multi-party interaction transcripts include meeting transcripts, conference call transcripts, auction transcripts, and/or the like.

The term “interaction utterance” may refer to a data entity that is configured to describe a semantically coherent unit of words that is recorded by the multi-party interaction transcript. An example of an interaction utterance is a sentence. In some embodiments, to detect interaction utterances in a multi-party interaction transcript, a predictive data analysis computing entity utilizes one or more speech fragmentation algorithms, such as one or more sentence detection algorithms. Each interaction utterance in a multi-party interaction transcript is typically associated with an interaction party of the plurality of interaction parties that are in turn associated with the multi-party interaction transcript. Accordingly, in some embodiments, the interaction utterances in a multi-party interaction transcript may be divided into two or more subsets, where each subset includes the set of interaction utterances in a multi-party interaction transcript that is associated with a particular interaction party of the two or more interaction parties. The described subsets are referred to herein as per-party utterance subsets.

The term “per-party utterance subset” may refer to a data entity that is configured to describe a subset of interaction utterances in a multi-party interaction transcript that includes all of the interaction utterances that are associated with a particular interaction party of the two or more interaction parties that are associated with the particular multi-party interaction transcript. Accordingly, in some embodiments, the interaction utterances in a multi-party interaction transcript may be divided into two or more subsets, where each subset includes the set of interaction utterances in a multi-party interaction transcript that is associated with a particular interaction party of the two or more interaction parties. For example, with respect to the call transcript for a call between a customer service agent and a customer, the call transcript may be associated with two per-party utterance subsets: a first per-party utterance subsets that describes all of the interaction utterances associated with the customer service agent and a second per-party utterance subset that describes all of the interaction utterances associated with the customer.

The term “optimal extractive summary” may refer to a data entity that is configured to describe one or more segments of a multi-party interaction transcript that are deemed to optimally represent the contents of the multi-party interaction transcript. For example, the optimal extractive summary for a multi-party interaction transcript may include a covered subset of the interaction utterances associated with the multi-party interaction transcript that are deemed to optimally represent the contents of the multi-party interaction transcript. As another example, the optimal extractive summary for a multi-party interaction transcript may include one or more content words associated with the multi-party interaction transcript that are deemed to optimally represent the contents of the multi-party interaction transcript. As yet another example, the optimal extractive summary for a multi-party interaction transcript may include one or more word collections (e.g., one or more word collections) associated with the multi-party interaction transcript that are deemed to optimally represent the contents of the multi-party interaction transcript.

The term “content word” may refer to a data entity that is configured to describe a word that is deemed to likely have a substantial amount of information about the contents of at least one portion of a multi-party interaction transcript that includes the content word. Content words of a multi-party interaction transcript may be determined by using one or more semantically significant word detection techniques. For example, in accordance with one technique, words having particular grammatical/semantic roles (e.g., as inferred by PoS) may be deemed to be content words. Accordingly, in some embodiments, at least one of nouns, verbs, and adjectives (e.g., all three of nouns, verbs, and adjectives) may be deemed to be content words. As another example, in accordance with another technique, words that match entries in a dictionary of semantically significant words may be deemed to be content words. In some embodiments, a predictive data analysis computing entity assigns a weight to each content word using a word ranking algorithm, such as a core rank algorithm.

The term “information quality measure” may refer to a data entity that is configured to describe an estimated measure of the degree of information contained in the words of a corresponding interaction utterance. For example, the information quality measure for a particular sentence may describe the likelihood that the sentence contains significant information based at least in part on the words used in the sentence. In some embodiments, the information quality measure for an interaction utterance is determined by combining (e.g., aggregating) each per-word information quality measure for a word used in the interaction utterance. In some embodiments, the per-word information quality measure for a word in an interaction utterance is determined based at least in part on the Equation 1:

<maths id="MATH-US-00001" num="00001"><math overflow="scroll"><mtable><mtr><mtd><mrow><mrow><mi>S</mi><mo>⁡</mo><mrow><mo>(</mo><msub><mi>V</mi><mi>i</mi></msub><mo>)</mo></mrow></mrow><mo>=</mo><mrow><mrow><mo>(</mo><mrow><mn>1</mn><mo>-</mo><mi>d</mi></mrow><mo>)</mo></mrow><mo>+</mo><mrow><mi>d</mi><mo>×</mo><mrow><munderover><mo>∑</mo><mrow><msub><mi>V</mi><mi>j</mi></msub><mo>∈</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mrow><mi>adj</mi><mo>⁡</mo><mrow><mo>(</mo><msub><mi>V</mi><mi>i</mi></msub><mo>)</mo></mrow></mrow></mrow><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle></munderover><mo>⁢</mo><mrow><mfrac><msub><mi>w</mi><mi>ij</mi></msub><mrow><msub><mo>∑</mo><mrow><msub><mi>V</mi><mi>k</mi></msub><mo>∈</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mrow><mi>adj</mi><mo>⁡</mo><mrow><mo>(</mo><msub><mi>V</mi><mi>i</mi></msub><mo>)</mo></mrow></mrow></mrow></msub><mo>⁢</mo><msub><mi>w</mi><mi>jk</mi></msub></mrow></mfrac><mo>⁢</mo><mrow><mi>S</mi><mo>⁡</mo><mrow><mo>(</mo><msub><mi>V</mi><mi>i</mi></msub><mo>)</mo></mrow></mrow></mrow></mrow></mrow></mrow></mrow></mtd><mtd><mrow><mo>(</mo><mrow><mi>Equation</mi><mo>⁢</mo><mstyle><mspace width="0.8em" height="0.8ex" /></mstyle><mo>⁢</mo><mn>1</mn></mrow><mo>)</mo></mrow></mtd></mtr></mtable></math></maths><img file="US2022189484A1_D0001.tif" /><img file="US2022189484A1_D0002.tif" /><img file="US2022189484A1_D0003.tif" /><img file="US2022189484A1_D0004.tif" /><img file="US2022189484A1_D0005.tif" /><img file="US2022189484A1_D0006.tif" />

In Equation 1, V_irepresents the word, S(V_i) represents the per-word information quality measure for the word, adj(V_i) represents the neighboring words of V_ias determined based at least in part on a graph representation of the semantic relationship of the words of the interaction utterance, and dis a damping factor (e.g., 0.85).

The term “linguistic quality measure” may refer to a data entity that is configured to describe an estimated measure of the likelihood that an utterance represents a linguistically viable collection of words. In some embodiments, the linguistic quality measure is determined based at least in part on a language model, such as a trigram language model that assigns probabilities to sequences of words. For example, given an interaction utterance p_i^c^{<sup2>j </sup2>}that includes a sequence of words {w₁, w₂, . . . , w_q}, the linguistic quality score for the interaction utterance (denoted as LQ(p_i^c^{<sup2>j</sup2>})) may be determined using the equation

<maths id="MATH-US-00002" num="00002"><math overflow="scroll"><mrow><mrow><mrow><mi>LQ</mi><mo>⁡</mo><mrow><mo>(</mo><msubsup><mi>p</mi><mi>i</mi><msup><mi>c</mi><mi>j</mi></msup></msubsup><mo>)</mo></mrow></mrow><mo>=</mo><mfrac><mn>1</mn><mrow><mn>1</mn><mo>-</mo><mrow><mi>LL</mi><mo>⁡</mo><mrow><mo>(</mo><mrow><msub><mi>w</mi><mn>1</mn></msub><mo>,</mo><msub><mi>w</mi><mn>2</mn></msub><mo>,</mo><mi>…</mi><mo>⁢</mo><mstyle><mspace width="0.8em" height="0.8ex" /></mstyle><mo>,</mo><msub><mi>w</mi><mi>q</mi></msub></mrow><mo>)</mo></mrow></mrow></mrow></mfrac></mrow><mo>,</mo></mrow></math></maths><img file="US2022189484A1_D0007.tif" /><img file="US2022189484A1_D0008.tif" /><img file="US2022189484A1_D0009.tif" /><img file="US2022189484A1_D0010.tif" /><img file="US2022189484A1_D0011.tif" /><img file="US2022189484A1_D0012.tif" />

where LL(w₁, w₂, . . . , w_q) is a probability aggregation measure for the interaction utterance p_i^c^{<sup2>j </sup2>}that can be computed using the equation

<maths id="MATH-US-00003" num="00003"><math overflow="scroll"><mrow><mrow><mi>LL</mi><mo>⁡</mo><mrow><mo>(</mo><mrow><msub><mi>w</mi><mn>1</mn></msub><mo>,</mo><msub><mi>w</mi><mn>2</mn></msub><mo>,</mo><mi>…</mi><mo>⁢</mo><mstyle><mspace width="0.8em" height="0.8ex" /></mstyle><mo>,</mo><msub><mi>w</mi><mi>q</mi></msub></mrow><mo>)</mo></mrow></mrow><mo>=</mo><mrow><mfrac><mn>1</mn><mi>L</mi></mfrac><mo>⁢</mo><msub><mi>log</mi><mn>2</mn></msub><mo>⁢</mo><mrow><munderover><mo>∏</mo><mrow><mi>t</mi><mo>=</mo><mn>2</mn></mrow><mi>q</mi></munderover><mo>⁢</mo><mrow><mrow><mo>(</mo><mrow><mrow><msub><mi>w</mi><mi>t</mi></msub><mo>❘</mo><msub><mi>w</mi><mrow><mi>t</mi><mo>-</mo><mn>1</mn></mrow></msub></mrow><mo>,</mo><msub><mi>w</mi><mrow><mi>t</mi><mo>-</mo><mn>2</mn></mrow></msub></mrow><mo>)</mo></mrow><mo>.</mo></mrow></mrow></mrow></mrow></math></maths><img file="US2022189484A1_D0013.tif" /><img file="US2022189484A1_D0014.tif" /><img file="US2022189484A1_D0015.tif" /><img file="US2022189484A1_D0016.tif" /><img file="US2022189484A1_D0017.tif" /><img file="US2022189484A1_D0018.tif" />

The term “utterance coverage constraint” may refer to a data entity that is configured to describe an optimization constraint for determining an eligible extractive summarization of a multi-party interaction transcript, where the optimization constraint may be configured to require that, if the covered subset for a particular eligible extractive summary comprises a particular covered content word, then the covered subset for the particular eligible extractive summary should further comprise at least one interaction utterance of the plurality of interaction utterances that is associated with the particular covered content word. In other words, the utterance coverage may require that, for each content word that is included in an eligible extractive summary, at least one interaction utterance including the content word be included in the extractive summary. Thus, if an extractive summary includes at least one content word that is selected to be included in the extractive summary but includes no interaction utterances that comprise the selected content word, that extractive summary may not deemed an eligible extractive summary under the utterance coverage constraint. In some embodiments, the utterance coverage constraint is satisfied if the condition described by the equation Σ_j=1^zΣ_j∈T_{<sub2>j</sub2>}^jx_i^j≥y_kis satisfied, where j is an iteration variable that iterates over z interaction parties associated with the multi-party interaction transcript, T_j^kis the set of interaction utterances that are both associated with the jth interaction party and in which the kth content word appears, x_i^jan index variable that has the value of one if the ith interaction utterance of the jth interaction party is selected as part of an eligible extractive summary and has the value of zero otherwise, and y_kis an index variable that has the value of one if the kth content word is selected as part of an eligible extractive summary and has the value of zero otherwise.

The term “content word coverage constraint” may refer to a data entity that is configured to describe an optimization constraint for determining an eligible extractive summarization of a multi-party interaction transcript, where the optimization constraint may be configured to require that, if the covered subset for a particular eligible extractive summary comprises a particular interaction utterance having one or more per-utterance content words, then the covered subset for the particular eligible extractive summary should comprise all of the one or more per-utterance content words. In other words, the content coverage constraint may require that, for each interaction utterance that is included in an eligible extractive summary, all content words appearing in the interaction utterance also be included in the eligible extractive summary. Thus, if an extractive summary includes an interaction utterance but does not include all of the content words of the interaction utterance, that extractive summary may not deemed an eligible extractive summary under the content word coverage constraint. In some embodiments, the content word coverage constraint is satisfied if the condition described by the equation Σ_k∈C_{<sub2>i</sub2>}^jy_k≥|C_i^j|×x_i^jsatisfied, where C_i^jis the set of content words in the ith interaction utterance of the jth interaction party, k is an iteration variable that iterates over C_i^j, |C_i^j| is the cardinality of C_i^j, and x_i^jan index variable that has the value of one if the ith interaction utterance of the jth interaction party is selected as part of an eligible extractive summary and has the value of zero otherwise.

The term “party coverage constraint” may refer to a data entity that is configured to describe an optimization constraint for determining an eligible extractive summarization of a multi-party interaction transcript, where the optimization constraint may be configured to require that each covered subset for an eligible extractive summary comprises at least one interaction utterance from each per-party utterance subset of the one or more per-party utterance subsets. In other words, the party coverage constraint may require that at least one interaction utterance of each party to the multi-party interaction transcript is included in an eligible extractive summary. Thus, if an extractive summary has no interaction utterances from one or more interaction parties to the corresponding multi-party interaction transcript, then the extractive summary may be deemed to be ineligible in accordance with the party coverage constraint. In some embodiments, the party coverage constraint is satisfied if the condition described by the equation Σ_i=1^N^{<sub2>j</sub2>}x_i^j≥1 is satisfied, where i is an iteration variable that iterates over N_jutterances of the jth interaction party and x_i^jis an index variable that has the value of one if the ith interaction utterance of the jth interaction party is selected as part of an eligible extractive summary and has the value of zero otherwise.

The term “party size constraint” may refer to a data entity that is configured to describe an optimization constraint for determining an eligible extractive summarization of a multi-party interaction transcript, where the optimization constraint may be configured to require that each covered subset for an eligible extractive summary comprises a below-threshold count of interaction utterances from each per-party utterance subset of the one or more per-party utterance subsets. In other words, the party size constraint may ensure that no interaction party is over-represented in an eligible extractive summary, where determining whether an interaction party is over-represented may be determined in accordance with whether the number of interaction utterances of the interaction party that appear in the eligible extractive summary exceed a per-party utterance count threshold value. Thus, assuming an exemplary per-party utterance count threshold value of ten, if an extractive summary includes eleven interaction utterances from a particular interaction party, then the extractive summary may be deemed ineligible in accordance with the party size constraint. In some embodiments, the party size constraint may be deemed satisfied if the condition described by the equation Σ_j=1^zx_i^j≥L_jis satisfied, where j is an iteration variable that iterates over z interaction parties associated with the multi-party interaction transcript, x_i^jis an index variable that has the value of one if the ith interaction utterance of the jth interaction party is selected as part of an eligible extractive summary and has the value of zero otherwise, and L_jis the above-noted per-party utterance count threshold value (number of interaction utterances) for the jth party.

The term “similarity-based constraint” may refer to a data entity that is configured to describe an optimization constraint for determining an eligible extractive summarization of a multi-party interaction transcript, where the optimization constraint may be configured to require that if the covered subset for a particular eligible extractive summary comprises a particular interaction utterance that is in a particular per-party utterance, then the covered subset for the particular eligible extractive summary should further comprise each other interaction utterance that is in any per-party utterance subset other than the particular per-party utterance subset and that has a threshold-satisfying utterance similarity measure with respect to the particular interaction utterance. In other words, the similarity-based constraint may require that, if a first interaction utterance from a first interaction party is included in an eligible extractive summary, then each interaction utterance of any other interaction party whose similarity measure with respect to the first interaction utterance satisfies (e.g., exceeds, exceeds or is equal to, and/or the like) an utterance similarity measure threshold value should also be included in the eligible extractive summary. Thus, if a first interaction utterance that is deemed to have threshold-satisfying similarity with respect to a second interaction utterance is included in an extractive summary, and further if the extractive summary does not include the second interaction utterance, then extractive summary may be deemed ineligible under the similarity-based constraint. In some embodiments, the similarity-based constraint is deemed satisfied if the condition described by the equation x_p^a+x_q^b≥2 if sim(s_p^a, s_q^b)≥0.4 is satisfied, where s_p^ais the ath interaction utterance of the pth party, s_q^bis the bth interaction utterance of the qth party, x_p^ais an index variable that has the value of one if s_p^ais selected as part of an eligible extractive summary and has the value of zero otherwise, x_q^bis an index variable that has the value of one if s_q^bis selected as part of an eligible extractive summary and has the value of zero otherwise, sim(s_p^a, s_q^b) is the similarity measure for s_p^aand s_q^b, and 0.4 is an example of an utterance similarity measure threshold value.

The term “eligible extractive summary” may refer to a data entity that is configured to describe a defined segment of a multi-party interaction transcript that satisfies one or more required optimization constraints, where a segment of a multi-party interaction transcript may include at least one a subset of interaction utterances of the multi-party interaction transcript and a subset of content words of the multi-party interaction transcript. Examples of required optimization constraints that may be satisfied by eligible extractive summaries include as an utterance coverage constraint, a content word coverage constraint, a party coverage constraint, a party size constraint, and a similarity-based constraint. As described above: (i) the utterance coverage constraint may require that, if the covered subset for a particular eligible extractive summary comprises a particular covered content word, then the covered subset for the particular eligible extractive summary should further comprise at least one interaction utterance of the plurality of interaction utterances that is associated with the particular covered content words; (ii) the content word coverage constraint may require that, if the covered subset for a particular eligible extractive summary comprises a particular interaction utterance having one or more per-utterance content words, then the covered subset for the particular eligible extractive summary should comprise all of the one or more per-utterance content words; (iii) the party coverage constraint may require that each covered subset for an eligible extractive summary comprises at least one interaction utterance from each per-party utterance subset of the one or more per-party utterance subsets; (iv) the party size constraint may require that each covered subset for an eligible extractive summary comprises a below-threshold count of interaction utterances from each per-party utterance subset of the one or more per-party utterance subsets; and (iv) the similarity-based constraint may require that if the covered subset for a particular eligible extractive summary comprises a particular interaction utterance that is in a particular per-party utterance, then the covered subset for the particular eligible extractive summary should further comprise each other interaction utterance that is in any per-party utterance subset other than the particular per-party utterance subset and that has a threshold-satisfying utterance similarity measure with respect to the particular interaction utterance.

The term “overall summary utility measure” may refer to a data entity that is configured to describe an estimated measure of usefulness of the extractive summary that is determined based at least in part on each information quality measure for an interaction utterance of the plurality of interaction utterances that is in the covered subset for the eligible extractive summary and each linguistic quality measure for an interaction utterance of the plurality of interaction utterances that is in the covered subset for the eligible extractive summary. In some embodiments, the overall summary utility measure may be determined based at least in part on the output of the equation Σ_j=1^zΣ_i=1^N^{<sub2>j</sub2>}IQ(s_i^j)·x_i^j+Σ_k=1^Mscore(k)·y_k, where j is an iteration variable that iterates over z interaction parties associated with the multi-party interaction transcript, i is an iteration variable that iterates over N_jutterances of the jth interaction party, s_i^jis the ith interaction utterance of the jth interaction party, IQ(s_i^j) is the information quality measure of s_i^j, LQ(s_i^j) is the linguistic quality measure of s_i^j, x_i^jis an index variable that has the value of one if s_i^jis selected as part of an eligible extractive summary and has the value of zero otherwise, k is an iteration variable that iterates over M content words, score(k) is the weight of the kth content word, and y_kis an index variable that has the value of one if the kth content word is selected as part of an eligible extractive summary and has the value of zero otherwise.

III. Computer Program Products, Methods, and Computing Entities

Embodiments of the present invention may be implemented in various ways, including as computer program products that comprise articles of manufacture. Such computer program products may include one or more software components including, for example, software objects, methods, data structures, or the like. A software component may be coded in any of a variety of programming languages. An illustrative programming language may be a lower-level programming language such as an assembly language associated with a particular hardware architecture and/or operating system platform. A software component comprising assembly language instructions may require conversion into executable machine code by an assembler prior to execution by the hardware architecture and/or platform. Another example programming language may be a higher-level programming language that may be portable across multiple architectures. A software component comprising higher-level programming language instructions may require conversion to an intermediate representation by an interpreter or a compiler prior to execution.

Other examples of programming languages include, but are not limited to, a macro language, a shell or command language, a job control language, a script language, a database query or search language, and/or a report writing language. In one or more example embodiments, a software component comprising instructions in one of the foregoing examples of programming languages may be executed directly by an operating system or other software component without having to be first transformed into another form. A software component may be stored as a file or other data storage construct. Software components of a similar type or functionally related may be stored together such as, for example, in a particular directory, folder, or library. Software components may be static (e.g., pre-established or fixed) or dynamic (e.g., created or modified at the time of execution).

A computer program product may include a non-transitory computer-readable storage medium storing applications, programs, program modules, scripts, source code, program code, object code, byte code, compiled code, interpreted code, machine code, executable instructions, and/or the like (also referred to herein as executable instructions, instructions for execution, computer program products, program code, and/or similar terms used herein interchangeably). Such non-transitory computer-readable storage media include all computer-readable media (including volatile and non-volatile media).

In one embodiment, a non-volatile computer-readable storage medium may include a floppy disk, flexible disk, hard disk, solid-state storage (SSS) (e.g., a solid state drive (SSD), solid state card (SSC), solid state module (SSM), enterprise flash drive, magnetic tape, or any other non-transitory magnetic medium, and/or the like. A non-volatile computer-readable storage medium may also include a punch card, paper tape, optical mark sheet (or any other physical medium with patterns of holes or other optically recognizable indicia), compact disc read only memory (CD-ROM), compact disc-rewritable (CD-RW), digital versatile disc (DVD), Blu-ray disc (BD), any other non-transitory optical medium, and/or the like. Such a non-volatile computer-readable storage medium may also include read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), flash memory (e.g., Serial, NAND, NOR, and/or the like), multimedia memory cards (MMC), secure digital (SD) memory cards, SmartMedia cards, CompactFlash (CF) cards, Memory Sticks, and/or the like. Further, a non-volatile computer-readable storage medium may also include conductive-bridging random access memory (CBRAM), phase-change random access memory (PRAM), ferroelectric random-access memory (FeRAM), non-volatile random-access memory (NVRAM), magnetoresistive random-access memory (MRAM), resistive random-access memory (RRAM), Silicon-Oxide-Nitride-Oxide-Silicon memory (SONOS), floating junction gate random access memory (FJG RAM), Millipede memory, racetrack memory, and/or the like.

In one embodiment, a volatile computer-readable storage medium may include random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), fast page mode dynamic random access memory (FPM DRAM), extended data-out dynamic random access memory (EDO DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), double data rate type two synchronous dynamic random access memory (DDR2 SDRAM), double data rate type three synchronous dynamic random access memory (DDR3 SDRAM), Rambus dynamic random access memory (RDRAM), Twin Transistor RAM (TTRAM), Thyristor RAM (T-RAM), Zero-capacitor (Z-RAM), Rambus in-line memory module (RIMM), dual in-line memory module (DIMM), single in-line memory module (SIMM), video random access memory (VRAM), cache memory (including various levels), flash memory, register memory, and/or the like. It will be appreciated that where embodiments are described to use a computer-readable storage medium, other types of computer-readable storage media may be substituted for or used in addition to the computer-readable storage media described above.

As should be appreciated, various embodiments of the present invention may also be implemented as methods, apparatus, systems, computing devices, computing entities, and/or the like. As such, embodiments of the present invention may take the form of an apparatus, system, computing device, computing entity, and/or the like executing instructions stored on a computer-readable storage medium to perform certain steps or operations. Thus, embodiments of the present invention may also take the form of an entirely hardware embodiment, an entirely computer program product embodiment, and/or an embodiment that comprises combination of computer program products and hardware performing certain steps or operations. Embodiments of the present invention are described below with reference to block diagrams and flowchart illustrations. Thus, it should be understood that each block of the block diagrams and flowchart illustrations may be implemented in the form of a computer program product, an entirely hardware embodiment, a combination of hardware and computer program products, and/or apparatus, systems, computing devices, computing entities, and/or the like carrying out instructions, operations, steps, and similar words used interchangeably (e.g., the executable instructions, instructions for execution, program code, and/or the like) on a computer-readable storage medium for execution. For example, retrieval, loading, and execution of code may be performed sequentially such that one instruction is retrieved, loaded, and executed at a time. In some exemplary embodiments, retrieval, loading, and/or execution may be performed in parallel such that multiple instructions are retrieved, loaded, and/or executed together. Thus, such embodiments can produce specifically-configured machines performing the steps or operations specified in the block diagrams and flowchart illustrations. Accordingly, the block diagrams and flowchart illustrations support various combinations of embodiments for performing the specified instructions, operations, or steps.

IV. Exemplary System Architecture

<figref idref="DRAWINGS">FIG. 1</figref> is a schematic diagram of an example architecture <b>100</b> for performing predictive data analysis. The architecture <b>100</b> includes a predictive data analysis system <b>101</b> configured to receive predictive data analysis requests from external computing entities <b>102</b>, process the predictive data analysis requests to generate predictions, provide the generated predictions to the external computing entities <b>102</b>, and automatically perform prediction-based actions based at least in part on the generated predictions. An example of a prediction that can be generated using the predictive data analysis system <b>101</b> is a prediction about summarization of a natural language document, such as summarization of a transcript (e.g., summarization of a multi-party interaction transcript, such as a two-party call transcript).

In some embodiments, predictive data analysis system <b>101</b> may communicate with at least one of the external computing entities <b>102</b> using one or more communication networks. Examples of communication networks include any wired or wireless communication network including, for example, a wired or wireless local area network (LAN), personal area network (PAN), metropolitan area network (MAN), wide area network (WAN), or the like, as well as any hardware, software and/or firmware required to implement it (such as, e.g., network routers, and/or the like).

The predictive data analysis system <b>101</b> may include a predictive data analysis computing entity <b>106</b> and a storage subsystem <b>108</b>. The predictive data analysis computing entity <b>106</b> may be configured to receive predictive data analysis requests from one or more external computing entities <b>102</b>, process the predictive data analysis requests to generate predictions corresponding to the predictive data analysis requests, provide the generated predictions to the external computing entities <b>102</b>, and automatically perform prediction-based actions based at least in part on the generated predictions.

The storage subsystem <b>108</b> may be configured to store input data used by the predictive data analysis computing entity <b>106</b> to perform predictive data analysis as well as model definition data used by the predictive data analysis computing entity <b>106</b> to perform various predictive data analysis tasks. The storage subsystem <b>108</b> may include one or more storage units, such as multiple distributed storage units that are connected through a computer network. Each storage unit in the storage subsystem <b>108</b> may store at least one of one or more data assets and/or one or more data about the computed properties of one or more data assets. Moreover, each storage unit in the storage subsystem <b>108</b> may include one or more non-volatile storage or memory media including, but not limited to, hard disks, ROM, PROM, EPROM, EEPROM, flash memory, MMCs, SD memory cards, Memory Sticks, CBRAM, PRAM, FeRAM, NVRAM, MRAM, RRAM, SONOS, FJG RAM, Millipede memory, racetrack memory, and/or the like.

Exemplary Predictive Data Analysis Computing Entity

<figref idref="DRAWINGS">FIG. 2</figref> provides a schematic of a predictive data analysis computing entity <b>106</b> according to one embodiment of the present invention. In general, the terms computing entity, computer, entity, device, system, and/or similar words used herein interchangeably may refer to, for example, one or more computers, computing entities, desktops, mobile phones, tablets, phablets, notebooks, laptops, distributed systems, kiosks, input terminals, servers or server networks, blades, gateways, switches, processing devices, processing entities, set-top boxes, relays, routers, network access points, base stations, the like, and/or any combination of devices or entities adapted to perform the functions, operations, and/or processes described herein. Such functions, operations, and/or processes may include, for example, transmitting, receiving, operating on, processing, displaying, storing, determining, creating/generating, monitoring, evaluating, comparing, and/or similar terms used herein interchangeably. In one embodiment, these functions, operations, and/or processes can be performed on data, content, information, and/or similar terms used herein interchangeably.

As indicated, in one embodiment, the predictive data analysis computing entity <b>106</b> may also include one or more communications interfaces <b>220</b> for communicating with various computing entities, such as by communicating data, content, information, and/or similar terms used herein interchangeably that can be transmitted, received, operated on, processed, displayed, stored, and/or the like.

As shown in <figref idref="DRAWINGS">FIG. 2</figref>, in one embodiment, the predictive data analysis computing entity <b>106</b> may include, or be in communication with, one or more processing elements <b>205</b> (also referred to as processors, processing circuitry, and/or similar terms used herein interchangeably) that communicate with other elements within the predictive data analysis computing entity <b>106</b> via a bus, for example. As will be understood, the processing element <b>205</b> may be embodied in a number of different ways.

For example, the processing element <b>205</b> may be embodied as one or more complex programmable logic devices (CPLDs), microprocessors, multi-core processors, coprocessing entities, application-specific instruction-set processors (ASIPs), microcontrollers, and/or controllers. Further, the processing element <b>205</b> may be embodied as one or more other processing devices or circuitry. The term circuitry may refer to an entirely hardware embodiment or a combination of hardware and computer program products. Thus, the processing element <b>205</b> may be embodied as integrated circuits, application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs), programmable logic arrays (PLAs), hardware accelerators, other circuitry, and/or the like.

As will therefore be understood, the processing element <b>205</b> may be configured for a particular use or configured to execute instructions stored in volatile or non-volatile media or otherwise accessible to the processing element <b>205</b>. As such, whether configured by hardware or computer program products, or by a combination thereof, the processing element <b>205</b> may be capable of performing steps or operations according to embodiments of the present invention when configured accordingly.

In one embodiment, the predictive data analysis computing entity <b>106</b> may further include, or be in communication with, non-volatile media (also referred to as non-volatile storage, memory, memory storage, memory circuitry and/or similar terms used herein interchangeably). In one embodiment, the non-volatile storage or memory may include one or more non-volatile storage or memory media <b>210</b>, including, but not limited to, hard disks, ROM, PROM, EPROM, EEPROM, flash memory, MMCs, SD memory cards, Memory Sticks, CBRAM, PRAM, FeRAM, NVRAM, MRAM, RRAM, SONOS, FJG RAM, Millipede memory, racetrack memory, and/or the like.

As will be recognized, the non-volatile storage or memory media may store databases, database instances, database management systems, data, applications, programs, program modules, scripts, source code, object code, byte code, compiled code, interpreted code, machine code, executable instructions, and/or the like. The term database, database instance, database management system, and/or similar terms used herein interchangeably may refer to a collection of records or data that is stored in a computer-readable storage medium using one or more database models, such as a hierarchical database model, network model, relational model, entity-relationship model, object model, document model, semantic model, graph model, and/or the like.

In one embodiment, the predictive data analysis computing entity <b>106</b> may further include, or be in communication with, volatile media (also referred to as volatile storage, memory, memory storage, memory circuitry and/or similar terms used herein interchangeably). In one embodiment, the volatile storage or memory may also include one or more volatile storage or memory media <b>215</b>, including, but not limited to, RAM, DRAM, SRAM, FPM DRAM, EDO DRAM, SDRAM, DDR SDRAM, DDR2 SDRAM, DDR3 SDRAM, RDRAM, TTRAM, T-RAM, Z-RAM, RIMM, DIMM, SIMM, VRAM, cache memory, register memory, and/or the like.

As will be recognized, the volatile storage or memory media may be used to store at least portions of the databases, database instances, database management systems, data, applications, programs, program modules, scripts, source code, object code, byte code, compiled code, interpreted code, machine code, executable instructions, and/or the like being executed by, for example, the processing element <b>205</b>. Thus, the databases, database instances, database management systems, data, applications, programs, program modules, scripts, source code, object code, byte code, compiled code, interpreted code, machine code, executable instructions, and/or the like may be used to control certain aspects of the operation of the predictive data analysis computing entity <b>106</b> with the assistance of the processing element <b>205</b> and operating system.

As indicated, in one embodiment, the predictive data analysis computing entity <b>106</b> may also include one or more communications interfaces <b>220</b> for communicating with various computing entities, such as by communicating data, content, information, and/or similar terms used herein interchangeably that can be transmitted, received, operated on, processed, displayed, stored, and/or the like. Such communication may be executed using a wired data transmission protocol, such as fiber distributed data interface (FDDI), digital subscriber line (DSL), Ethernet, asynchronous transfer mode (ATM), frame relay, data over cable service interface specification (DOCSIS), or any other wired transmission protocol. Similarly, the predictive data analysis computing entity <b>106</b> may be configured to communicate via wireless external communication networks using any of a variety of protocols, such as general packet radio service (GPRS), Universal Mobile Telecommunications System (UMTS), Code Division Multiple Access 2000 (CDMA2000), CDMA2000 1× (1×RTT), Wideband Code Division Multiple Access (WCDMA), Global System for Mobile Communications (GSM), Enhanced Data rates for GSM Evolution (EDGE), Time Division-Synchronous Code Division Multiple Access (TD-SCDMA), Long Term Evolution (LTE), Evolved Universal Terrestrial Radio Access Network (E-UTRAN), Evolution-Data Optimized (EVDO), High Speed Packet Access (HSPA), High-Speed Downlink Packet Access (HSDPA), IEEE 802.11 (Wi-Fi), Wi-Fi Direct, 802.16 (WiMAX), ultra-wideband (UWB), infrared (IR) protocols, near field communication (NFC) protocols, Wibree, Bluetooth protocols, wireless universal serial bus (USB) protocols, and/or any other wireless protocol.

Although not shown, the predictive data analysis computing entity <b>106</b> may include, or be in communication with, one or more input elements, such as a keyboard input, a mouse input, a touch screen/display input, motion input, movement input, audio input, pointing device input, joystick input, keypad input, and/or the like. The predictive data analysis computing entity <b>106</b> may also include, or be in communication with, one or more output elements (not shown), such as audio output, video output, screen/display output, motion output, movement output, and/or the like.

Exemplary External Computing Entity

<figref idref="DRAWINGS">FIG. 3</figref> provides an illustrative schematic representative of an external computing entity <b>102</b> that can be used in conjunction with embodiments of the present invention. In general, the terms device, system, computing entity, entity, and/or similar words used herein interchangeably may refer to, for example, one or more computers, computing entities, desktops, mobile phones, tablets, phablets, notebooks, laptops, distributed systems, kiosks, input terminals, servers or server networks, blades, gateways, switches, processing devices, processing entities, set-top boxes, relays, routers, network access points, base stations, the like, and/or any combination of devices or entities adapted to perform the functions, operations, and/or processes described herein. External computing entities <b>102</b> can be operated by various parties. As shown in <figref idref="DRAWINGS">FIG. 3</figref>, the external computing entity <b>102</b> can include an antenna <b>312</b>, a transmitter <b>304</b> (e.g., radio), a receiver <b>306</b> (e.g., radio), and a processing element <b>308</b> (e.g., CPLDs, microprocessors, multi-core processors, coprocessing entities, ASIPs, microcontrollers, and/or controllers) that provides signals to and receives signals from the transmitter <b>304</b> and receiver <b>306</b>, correspondingly.

The signals provided to and received from the transmitter <b>304</b> and the receiver <b>306</b>, correspondingly, may include signaling information/data in accordance with air interface standards of applicable wireless systems. In this regard, the external computing entity <b>102</b> may be capable of operating with one or more air interface standards, communication protocols, modulation types, and access types. More particularly, the external computing entity <b>102</b> may operate in accordance with any of a number of wireless communication standards and protocols, such as those described above with regard to the predictive data analysis computing entity <b>106</b>. In a particular embodiment, the external computing entity <b>102</b> may operate in accordance with multiple wireless communication standards and protocols, such as UMTS, CDMA2000, 1×RTT, WCDMA, GSM, EDGE, TD-SCDMA, LTE, E-UTRAN, EVDO, HSPA, HSDPA, Wi-Fi, Wi-Fi Direct, WiMAX, UWB, IR, NFC, Bluetooth, USB, and/or the like. Similarly, the external computing entity <b>102</b> may operate in accordance with multiple wired communication standards and protocols, such as those described above with regard to the predictive data analysis computing entity <b>106</b> via a network interface <b>320</b>.

Via these communication standards and protocols, the external computing entity <b>102</b> can communicate with various other entities using concepts such as Unstructured Supplementary Service Data (USSD), Short Message Service (SMS), Multimedia Messaging Service (MMS), Dual-Tone Multi-Frequency Signaling (DTMF), and/or Subscriber Identity Module Dialer (SIM dialer). The external computing entity <b>102</b> can also download changes, add-ons, and updates, for instance, to its firmware, software (e.g., including executable instructions, applications, program modules), and operating system.

According to one embodiment, the external computing entity <b>102</b> may include location determining aspects, devices, modules, functionalities, and/or similar words used herein interchangeably. For example, the external computing entity <b>102</b> may include outdoor positioning aspects, such as a location module adapted to acquire, for example, latitude, longitude, altitude, geocode, course, direction, heading, speed, universal time (UTC), date, and/or various other information/data. In one embodiment, the location module can acquire data, sometimes known as ephemeris data, by identifying the number of satellites in view and the relative positions of those satellites (e.g., using global positioning systems (GPS)). The satellites may be a variety of different satellites, including Low Earth Orbit (LEO) satellite systems, Department of Defense (DOD) satellite systems, the European Union Galileo positioning systems, the Chinese Compass navigation systems, Indian Regional Navigational satellite systems, and/or the like. This data can be collected using a variety of coordinate systems, such as the Decimal Degrees (DD); Degrees, Minutes, Seconds (DMS); Universal Transverse Mercator (UTM); Universal Polar Stereographic (UPS) coordinate systems; and/or the like. Alternatively, the location information/data can be determined by triangulating the external computing entity's <b>102</b> position in connection with a variety of other systems, including cellular towers, Wi-Fi access points, and/or the like. Similarly, the external computing entity <b>102</b> may include indoor positioning aspects, such as a location module adapted to acquire, for example, latitude, longitude, altitude, geocode, course, direction, heading, speed, time, date, and/or various other information/data. Some of the indoor systems may use various position or location technologies including RFID tags, indoor beacons or transmitters, Wi-Fi access points, cellular towers, nearby computing devices (e.g., smartphones, laptops) and/or the like. For instance, such technologies may include the iBeacons, Gimbal proximity beacons, Bluetooth Low Energy (BLE) transmitters, NFC transmitters, and/or the like. These indoor positioning aspects can be used in a variety of settings to determine the location of someone or something to within inches or centimeters.

The external computing entity <b>102</b> may also comprise a user interface (that can include a display <b>316</b> coupled to a processing element <b>308</b>) and/or a user input interface (coupled to a processing element <b>308</b>). For example, the user interface may be a user application, browser, user interface, and/or similar words used herein interchangeably executing on and/or accessible via the external computing entity <b>102</b> to interact with and/or cause display of information/data from the predictive data analysis computing entity <b>106</b>, as described herein. The user input interface can comprise any of a number of devices or interfaces allowing the external computing entity <b>102</b> to receive data, such as a keypad <b>318</b> (hard or soft), a touch display, voice/speech or motion interfaces, or other input device. In embodiments including a keypad <b>318</b>, the keypad <b>318</b> can include (or cause display of) the conventional numeric (0-9) and related keys (#, *), and other keys used for operating the external computing entity <b>102</b> and may include a full set of alphabetic keys or set of keys that may be activated to provide a full set of alphanumeric keys. In addition to providing input, the user input interface can be used, for example, to activate or deactivate certain functions, such as screen savers and/or sleep modes.

The external computing entity <b>102</b> can also include volatile storage or memory <b>322</b> and/or non-volatile storage or memory <b>324</b>, which can be embedded and/or may be removable. For example, the non-volatile memory may be ROM, PROM, EPROM, EEPROM, flash memory, MMCs, SD memory cards, Memory Sticks, CBRAM, PRAM, FeRAM, NVRAM, MRAM, RRAM, SONOS, FJG RAM, Millipede memory, racetrack memory, and/or the like. The volatile memory may be RAM, DRAM, SRAM, FPM DRAM, EDO DRAM, SDRAM, DDR SDRAM, DDR2 SDRAM, DDR3 SDRAM, RDRAM, TTRAM, T-RAM, Z-RAM, RIMM, DIMM, SIMM, VRAM, cache memory, register memory, and/or the like. The volatile and non-volatile storage or memory can store databases, database instances, database management systems, data, applications, programs, program modules, scripts, source code, object code, byte code, compiled code, interpreted code, machine code, executable instructions, and/or the like to implement the functions of the external computing entity <b>102</b>. As indicated, this may include a user application that is resident on the entity or accessible through a browser or other user interface for communicating with the predictive data analysis computing entity <b>106</b> and/or various other computing entities.

In another embodiment, the external computing entity <b>102</b> may include one or more components or functionality that are the same or similar to those of the predictive data analysis computing entity <b>106</b>, as described in greater detail above. As will be recognized, these architectures and descriptions are provided for exemplary purposes only and are not limiting to the various embodiments.

In various embodiments, the external computing entity <b>102</b> may be embodied as an artificial intelligence (AI) computing entity, such as an Amazon Echo, Amazon Echo Dot, Amazon Show, Google Home, and/or the like. Accordingly, the external computing entity <b>102</b> may be configured to provide and/or receive information/data from a user via an input/output mechanism, such as a display, a camera, a speaker, a voice-activated input, and/or the like. In certain embodiments, an AI computing entity may comprise one or more predefined and executable program algorithms stored within an onboard memory storage module, and/or accessible over a network. In various embodiments, the AI computing entity may be configured to retrieve and/or execute one or more of the predefined program algorithms upon the occurrence of a predefined trigger event.

V. Exemplary System Operations

<figref idref="DRAWINGS">FIG. 4</figref> is a flowchart diagram of an example process <b>400</b> for generating an optimal extractive summary of a multi-party interaction transcript. Via the various steps/operations of the process <b>400</b>, the predictive data analysis computing entity <b>106</b> can generate extractive summaries in an efficient and reliable manner by utilizing an unsupervised machine learning model that utilizes predictive inferences grounded in relationships of interaction utterances across interaction parties to generate reliable and representative extractive summaries for multi-party interactions.

The process <b>400</b> begins at step/operation <b>401</b> when the predictive data analysis computing entity <b>106</b> identifies the multi-party interaction transcript. The multi-party interaction transcript may describe a temporal flow of verbal interactions between two or more interaction parties. An example of a multi-party interaction transcript is a call transcript between two or more parties to the call, such as a call transcript for a call between a customer service agent and a customer. In the noted example, the call transcript may describe verbal interactions by the parties in a temporally sequential manner, where each verbal interaction by a party may include one or more utterances (e.g., on or more sentences). For example, with respect to the call transcript for a call between a customer service agent and a customer, the call transcript may describe that a first set of utterances by the customer service agent (e.g., “Hello, how is your day today and how can I help you?”) is temporally followed by a second set of utterances by the customer (e.g., “Thank you, doing well, I am trying to check my account balance.”), which may then be temporally followed by a third set of utterances by the customer service agent, and so on. Other examples of multi-party interaction transcripts include meeting transcripts, conference call transcripts, auction transcripts, and/or the like.

Accordingly, a multi-party interaction transcript is associated with two or more interaction parties, where each interaction party describes a verbal participant in the multi-party interaction that is associated with the multi-party interaction transcript. While various embodiments of the present invention are described with reference to two-party interaction transcripts and/or with reference to call transcripts, a person of ordinary skill in the relevant technology will recognize that the disclosed techniques may be utilize to generate optimal extractive summaries of interaction transcripts associated with any number of interaction parties as well as to generate optimal extractive summaries of interaction transcripts associated with any type of multi-party interaction events, such as to generate optimal extractive summaries of meeting transcripts, conference call transcripts, auction transcripts, and/or the like.

A multi-party interaction transcript is associated with one or more interaction utterances. An interaction utterance may be any semantically coherent unit of words that is recorded by the multi-party interaction transcript. An example of an interaction utterance is a sentence. In some embodiments, to detect interaction utterances in a multi-party interaction transcript, the predictive data analysis computing entity <b>106</b> utilizes one or more speech fragmentation algorithms, such as one or more sentence detection algorithms. Each interaction utterance in a multi-party interaction transcript is typically associated with an interaction party of the plurality of interaction parties that are in turn associated with the multi-party interaction transcript. Accordingly, in some embodiments, the interaction utterances in a multi-party interaction transcript may be divided into two or more subsets, where each subset includes the set of interaction utterances in a multi-party interaction transcript that is associated with a particular interaction party of the two or more interaction parties. The described subsets are referred to herein as per-party utterance subsets. For example, with respect to the call transcript for a call between a customer service agent and a customer, the call transcript may be associated with two per-party utterance subsets: a first per-party utterance subsets that describes all of the interaction utterances associated with the customer service agent and a second per-party utterance subset that describes all of the interaction utterances associated with the customer.

An operational example of a multi-party interaction transcript <b>500</b> is depicted in <figref idref="DRAWINGS">FIG. 5</figref>. As depicted in <figref idref="DRAWINGS">FIG. 5</figref>, the multi-party interaction transcript <b>500</b> is associated with a group of interaction utterances each followed by a numeric designator of zero or one, where the group of interaction utterances are listed in a temporally sequential order. As further depicted in <figref idref="DRAWINGS">FIG. 5</figref>, the multi-party interaction transcript <b>500</b> is associated with two interaction parties: a first interaction party associated with the numeric designator of zero and a second interaction party associated with the numeric designator of one. Accordingly, in the operational example of <figref idref="DRAWINGS">FIG. 500</figref>, the per-party utterance subsets for the multi-party interaction transcript <b>500</b> include: a first per-party utterance subset that includes each interaction utterance that is followed by the numeric designator of zero, and a second per-party utterance subset that includes each utterance that is followed by the numeric designator of one.

Returning to <figref idref="DRAWINGS">FIG. 4</figref>, at step/operation <b>402</b>, the predictive data analysis computing entity <b>106</b> updates the multi-party interaction transcript by performing preprocessing on the multi-party interaction transcript. Examples of preprocessing operations that may be performed as part of step/operation <b>402</b> may include stop word removal, non-relevant word filtering, named entity recognition (NER), and part of speech (PoS) tagging. One objective of the preprocessing operations performed at step/operation <b>402</b> may be to remove semantic inference complications resulting from presence of stop words and/or non-relevant words. Another objective of the preprocessing operations performed at step/operation <b>402</b> may be to highlight/record semantic relationships across words and/or other semantic units of the natural language data captured by the multi-party interaction transcript. In some embodiments, the former objective noted above is accomplished at least in part by performing stop word removal and non-relevant word filtering, while the latter objective noted above is accomplished at least in part by performing NER and PoS.

At step/operation <b>403</b>, the predictive data analysis computing entity <b>106</b> detects per-party utterance subsets in the multi-party interaction subset. As described above, a per-party utterance subset is a subset of interaction utterances in a multi-party interaction transcript that includes all of the interaction utterances that are associated with a particular interaction party of the two or more interaction parties that are associated with the particular multi-party interaction transcript. Accordingly, in some embodiments, the interaction utterances in a multi-party interaction transcript may be divided into two or more subsets, where each subset includes the set of interaction utterances in a multi-party interaction transcript that is associated with a particular interaction party of the two or more interaction parties. The described subsets are referred to herein as per-party utterance subsets. For example, with respect to the call transcript for a call between a customer service agent and a customer, the call transcript may be associated with two per-party utterance subsets: a first per-party utterance subsets that describes all of the interaction utterances associated with the customer service agent and a second per-party utterance subset that describes all of the interaction utterances associated with the customer.

At step/operation <b>404</b>, the predictive data analysis computing entity <b>106</b> processes the multi-party interaction transcript in accordance with the detected per-party utterance subsets in order to generate the optimal extractive summary. The optimal extractive summary for a multi-party interaction transcript describes one or more segments of the multi-party interaction transcript that are deemed to optimally represent the contents of the multi-party interaction transcript. For example, the optimal extractive summary for a multi-party interaction transcript may include a covered subset of the interaction utterances associated with the multi-party interaction transcript that are deemed to optimally represent the contents of the multi-party interaction transcript. As another example, the optimal extractive summary for a multi-party interaction transcript may include one or more content words associated with the multi-party interaction transcript that are deemed to optimally represent the contents of the multi-party interaction transcript. As yet another example, the optimal extractive summary for a multi-party interaction transcript may include one or more word collections (e.g., one or more word collections) associated with the multi-party interaction transcript that are deemed to optimally represent the contents of the multi-party interaction transcript.

In some embodiments, step/operation <b>404</b> may be performed in accordance with the process depicted in <figref idref="DRAWINGS">FIG. 6</figref>. The process depicted in <figref idref="DRAWINGS">FIG. 6</figref> begins at step/operation <b>601</b> when the predictive data analysis computing entity <b>106</b> detects content words in the multi-party interaction transcript. A content word may be a word that is deemed to likely have a substantial amount of information about the contents of at least one portion of a multi-party interaction transcript that includes the content word. Content words of a multi-party interaction transcript may be determined by using one or more semantically significant word detection techniques. For example, in accordance with one technique, words having particular grammatical/semantic roles (e.g., as inferred by PoS) may be deemed to be content words. Accordingly, in some embodiments, at least one of nouns, verbs, and adjectives (e.g., all three of nouns, verbs, and adjectives) may be deemed to be content words. As another example, in accordance with another technique, words that match entries in a dictionary of semantically significant words may be deemed to be content words. In some embodiments, the predictive data analysis computing entity <b>106</b> assigns a weight to each content word using a word ranking algorithm, such as a core rank algorithm.

At step/operation <b>602</b>, the predictive data analysis computing entity <b>106</b> determines an information quality measure and a linguistic quality measure for each interaction utterance in the multi-party interaction transcript. Example techniques for determining information quality measures and/or linguistic quality measures for interaction utterances are described below. However, a person of ordinary skill in the relevant technology will recognize that other techniques may be utilized to compute at least one of information quality measures and linguistic quality measures. In some embodiments, step/operation <b>602</b> can be performed concurrently with step/operation <b>601</b>.

An information quality measure may describe an estimated measure of the degree of information contained in the words of a corresponding interaction utterance. For example, the information quality measure for a particular sentence may describe the likelihood that the sentence contains significant information based at least in part on the words used in the sentence. In some embodiments, the information quality measure for an interaction utterance is determined by combining (e.g., aggregating) each per-word information quality measure for a word used in the interaction utterance. In some embodiments, the per-word information quality measure for a word in an interaction utterance is determined based at least in part on the equation

<maths id="MATH-US-00004" num="00004"><math overflow="scroll"><mtable><mtr><mtd><mrow><mrow><mi>S</mi><mo>⁡</mo><mrow><mo>(</mo><msub><mi>V</mi><mi>i</mi></msub><mo>)</mo></mrow></mrow><mo>=</mo><mrow><mrow><mo>(</mo><mrow><mn>1</mn><mo>-</mo><mi>d</mi></mrow><mo>)</mo></mrow><mo>+</mo><mrow><mi>d</mi><mo>×</mo><mrow><mover><msub><mo>∑</mo><mrow><mi>Vj</mi><mo>⁢</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>∈</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mrow><mi>adj</mi><mo>⁡</mo><mrow><mo>(</mo><msub><mi>V</mi><mi>i</mi></msub><mo>)</mo></mrow></mrow></mrow></msub><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle></mover><mo>⁢</mo><mrow><mfrac><msub><mi>w</mi><mi>ij</mi></msub><mrow><msub><mo>∑</mo><msub><mi>V</mi><mrow><mi>k</mi><mo>⁢</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>∈</mo><mstyle><mspace width="0.3em" height="0.3ex" /></mstyle><mo>⁢</mo><mrow><mi>adj</mi><mo>⁡</mo><mrow><mo>(</mo><msub><mi>V</mi><mi>j</mi></msub><mo>)</mo></mrow></mrow></mrow></msub></msub><mo>⁢</mo><msub><mi>w</mi><mi>jk</mi></msub></mrow></mfrac><mo>⁢</mo><mrow><mi>S</mi><mo>⁡</mo><mrow><mo>(</mo><msub><mi>V</mi><mi>i</mi></msub><mo>)</mo></mrow></mrow></mrow></mrow></mrow></mrow></mrow></mtd><mtd><mrow><mo>(</mo><mrow><mi>Equation</mi><mo>⁢</mo><mstyle><mspace width="0.8em" height="0.8ex" /></mstyle><mo>⁢</mo><mn>1</mn></mrow><mo>)</mo></mrow></mtd></mtr></mtable></math></maths><img file="US2022189484A1_D0019.tif" /><img file="US2022189484A1_D0020.tif" /><img file="US2022189484A1_D0021.tif" /><img file="US2022189484A1_D0022.tif" /><img file="US2022189484A1_D0023.tif" /><img file="US2022189484A1_D0024.tif" />

where V_irepresents the word, S(V_i) represents the per-word information quality measure for the word, adj(V_i) represents the neighboring words of V_ias determined based at least in part on a graph representation of the semantic relationship of the words of the interaction utterance, and dis a damping factor (e.g., <b>0</b>.<b>85</b>). Exemplary techniques for determining information quality measures for interaction utterances are described in Banerj ee et al., <i>Multi</i>-<i>Document Abstractive Summarization Using ILP based Multi</i>-<i>Sentence Compression</i>, arXiv:1609.07034v1, available online at https://arxiv.org/pdf/1609.07034.pdf (2016).

A linguistic quality measure may describe an estimated measure of the likelihood that an utterance represents a linguistically viable collection of words. In some embodiments, the linguistic quality measure is determined based at least in part on a language model, such as a trigram language model that assigns probabilities to sequences of words. For example, given an interaction utterance p_i^c^{<sup2>j </sup2>}that includes a sequence of words {w₁, w₂, . . . , w_q}, the linguistic quality score for the interaction utterance (denoted as LQ(p_i^c^{<sup2>j</sup2>})) may be determined using the equation

<maths id="MATH-US-00005" num="00005"><math overflow="scroll"><mrow><mrow><mrow><mi>L</mi><mo>⁢</mo><mrow><mi>Q</mi><mo>⁡</mo><mrow><mo>(</mo><msubsup><mi>p</mi><mi>i</mi><msup><mi>c</mi><mi>j</mi></msup></msubsup><mo>)</mo></mrow></mrow></mrow><mo>=</mo><mfrac><mn>1</mn><mrow><mn>1</mn><mo>-</mo><mrow><mi>LL</mi><mo>⁡</mo><mrow><mo>(</mo><mrow><msub><mi>w</mi><mn>1</mn></msub><mo>,</mo><msub><mi>w</mi><mn>2</mn></msub><mo>,</mo><mi>…</mi><mo>⁢</mo><mstyle><mspace width="0.8em" height="0.8ex" /></mstyle><mo>,</mo><msub><mi>w</mi><mi>q</mi></msub></mrow><mo>)</mo></mrow></mrow></mrow></mfrac></mrow><mo>,</mo></mrow></math></maths><img file="US2022189484A1_D0025.tif" /><img file="US2022189484A1_D0026.tif" /><img file="US2022189484A1_D0027.tif" /><img file="US2022189484A1_D0028.tif" /><img file="US2022189484A1_D0029.tif" /><img file="US2022189484A1_D0030.tif" />

where LL (w₁, w₂, . . . , w_q) is a probability aggregation measure for the interaction utterance p_i^c^{<sup2>j </sup2>}that can be computed using the equation

<maths id="MATH-US-00006" num="00006"><math overflow="scroll"><mrow><mrow><mrow><mi>LL</mi><mo>⁡</mo><mrow><mo>(</mo><mrow><msub><mi>w</mi><mn>1</mn></msub><mo>,</mo><msub><mi>w</mi><mn>2</mn></msub><mo>,</mo><mi>…</mi><mo>⁢</mo><mstyle><mspace width="0.8em" height="0.8ex" /></mstyle><mo>,</mo><msub><mi>w</mi><mi>q</mi></msub></mrow><mo>)</mo></mrow></mrow><mo>=</mo><mrow><mfrac><mn>1</mn><mi>L</mi></mfrac><mo>⁢</mo><msub><mi>log</mi><mn>2</mn></msub><mo>⁢</mo><mrow><munderover><mo>∏</mo><mrow><mi>t</mi><mo>=</mo><mn>2</mn></mrow><mi>q</mi></munderover><mo>⁢</mo><mrow><mo>(</mo><mrow><mrow><msub><mi>w</mi><mi>t</mi></msub><mo>❘</mo><msub><mi>w</mi><mrow><mi>t</mi><mo>-</mo><mn>1</mn></mrow></msub></mrow><mo>,</mo><msub><mi>w</mi><mrow><mi>t</mi><mo>-</mo><mn>2</mn></mrow></msub></mrow><mo>)</mo></mrow></mrow></mrow></mrow><mo>,</mo></mrow></math></maths><img file="US2022189484A1_D0031.tif" /><img file="US2022189484A1_D0032.tif" /><img file="US2022189484A1_D0033.tif" /><img file="US2022189484A1_D0034.tif" /><img file="US2022189484A1_D0035.tif" /><img file="US2022189484A1_D0036.tif" />

where L is the length of the interaction utterance p_i^c^{<sup2>j </sup2>}Exemplary techniques for determining linguistic quality measures for interaction utterances are described in Banerj ee et al., <i>Multi</i>-<i>Document Abstractive Summarization Using ILP based Multi</i>-<i>Sentence Compression</i>, arXiv:1609.07034v1, available online at https://arxiv.org/pdf/1609.07034.pdf (2016).

At step/operation <b>603</b>, the predictive data analysis computing entity <b>106</b> performs an ILP analysis based at least in part on each information quality measure for an interaction utterance and each linguistic quality measure for an interaction utterance to generate the optimal extractive summary. In some embodiments, the ILP operation is defined based an overall summary utility measure as well as a group of optimization constraints defined below. In some embodiments, to perform the ILP operation, the predictive data analysis computing entity <b>106</b> first generates a plurality of eligible extractive summaries, where each eligible extractive summary of the plurality of eligible comprises a covered subset of the plurality of interaction utterances that complies with one or more required optimization constraints. Afterward, the predictive data analysis computing entity <b>106</b> generates an overall summary utility measure for each eligible extractive summary based at least in part on each information quality measure for an interaction utterance of the plurality of interaction utterances that is in the covered subset for the eligible extractive summary and each linguistic quality measure for an interaction utterance of the plurality of interaction utterances that is in the covered subset for the eligible extractive summary. Thereafter, the predictive data analysis computing entity <b>106</b> generates the optimal extractive based at least in part on each overall summary utility measure for an eligible extractive summary of the plurality of eligible extractive summaries. In some embodiments, step/operation <b>603</b> can be performed concurrently with step/operations <b>601</b>-<b>602</b>.

In some embodiments, the ILP operation described above may be performed in accordance with the maximization problem max(Σ_j=1^zΣ_i=1^N^{<sub2>j</sub2>}IQ(s_i^j)LQ (s_i^j)·x_i^j+Σ_k=1^Mscore(k)·y_k) where j is an iteration variable that iterates over z interaction parties associated with the multi-party interaction transcript, i is an iteration variable that iterates over N_jutterances of the jth interaction party, s_i^jis the ith interaction utterance of the jth interaction party, IQ(s_i^j) is the information quality measure of s_i^j, LQ(s_i^j) is the linguistic quality measure of s_i^j, x_i^jis an index variable that has the value of one if s_i^jis selected as part of an eligible extractive summary and has the value of zero otherwise, k is an iteration variable that iterates over M content words, score(k) is the weight of the kth content word, and y_kis an index variable that has the value of one if the kth content word is selected as part of an eligible extractive summary and has the value of zero otherwise. In some embodiments, the noted optimization problem may be characterized by (i.e., may be framed as subject to) a group of required optimization constraints, such as at least one of (e.g., all of) an utterance coverage constraint, a content word coverage constraint, a party coverage constraint, a party size constraint, and a similarity-based constraint.

The utterance coverage constraint may require that, if the covered subset for a particular eligible extractive summary comprises a particular covered content word, then the covered subset for the particular eligible extractive summary should further comprise at least one interaction utterance of the plurality of interaction utterances that is associated with the particular covered content word. In other words, the utterance coverage may require that, for each content word that is included in an eligible extractive summary, at least one interaction utterance including the content word be included in the extractive summary. Thus, if an extractive summary includes at least one content word that is selected to be included in the extractive summary but includes no interaction utterances that comprise the selected content word, that extractive summary may not deemed an eligible extractive summary under the utterance coverage constraint. In some embodiments, the utterance coverage constraint is satisfied if the condition described by the equation Σ_j=1^zΣ_j∈T_{<sub2>j</sub2>}^kx_i^j≥y_kis satisfied, where j is an iteration variable that iterates over z interaction parties associated with the multi-party interaction transcript, T_j^kis the set of interaction utterances that are both associated with the jth interaction party and in which the kth content word appears, x_i^jis an index variable that has the value of one if the ith interaction utterance of the jth interaction party is selected as part of an eligible extractive summary and has the value of zero otherwise, and y_kis an index variable that has the value of one if the kth content word is selected as part of an eligible extractive summary and has the value of zero otherwise.

The content word coverage constraint may require that, if the covered subset for a particular eligible extractive summary comprises a particular interaction utterance having one or more per-utterance content words, then the covered subset for the particular eligible extractive summary should comprise all of the one or more per-utterance content words. In other words, the content coverage constraint may require that, for each interaction utterance that is included in an eligible extractive summary, all content words appearing in the interaction utterance also be included in the eligible extractive summary. Thus, if an extractive summary includes an interaction utterance but does not include all of the content words of the interaction utterance, that extractive summary may not deemed an eligible extractive summary under the content word coverage constraint. In some embodiments, the content word coverage constraint is satisfied if the condition described by the equation Σ_k∈C_{<sub2>i</sub2>}^jy_k≥|C_i^j|×x_i^jis satisfied, where C_i^jis the set of content words in the ith interaction utterance that contain the jth content word, k is an iteration variable that iterates over C_i^j, |C_i^j| is the cardinality of C_i^j, and x_i^jan index variable that has the value of one if the ith interaction utterance of the jth interaction party is selected as part of an eligible extractive summary and has the value of zero otherwise.

The party coverage constraint may require that each covered subset for an eligible extractive summary comprises at least one interaction utterance from each per-party utterance subset of the one or more per-party utterance subsets. In other words, the party coverage constraint may require that at least one interaction utterance of each party to the multi-party interaction transcript is included in an eligible extractive summary. Thus, if an extractive summary has no interaction utterances from one or more interaction parties to the corresponding multi-party interaction transcript, then the extractive summary may be deemed to be ineligible in accordance with the party coverage constraint. In some embodiments, the party coverage constraint is satisfied if the condition described by the equation Σ_i=1^N^{<sub2>j</sub2>}x_i^j≥1 is satisfied, where i is an iteration variable that iterates over N_jutterances of the jth interaction party and x_i^jis an index variable that has the value of one if the ith interaction utterance of the jth interaction party is selected as part of an eligible extractive summary and has the value of zero otherwise.

The party size constraint may require that each covered subset for an eligible extractive summary comprises a below-threshold count of interaction utterances from each per-party utterance subset of the one or more per-party utterance subsets. In other words, the party size constraint may ensure that no interaction party is over-represented in an eligible extractive summary, where determining whether an interaction party is over-represented may be determined in accordance with whether the number of interaction utterances of the interaction party that appear in the eligible extractive summary exceed a per-party utterance count threshold value. Thus, assuming an exemplary per-party utterance count threshold value of ten, if an extractive summary includes eleven interaction utterances from a particular interaction party, then the extractive summary may be deemed ineligible in accordance with the party size constraint. In some embodiments, the party size constraint may be deemed satisfied if the condition described by the equation Σ_j=1^zx_i^j≥L_j; is satisfied, where j is an iteration variable that iterates over z interaction parties associated with the multi-party interaction transcript, x_i^jis an index variable that has the value of one if the ith interaction utterance of the jth interaction party is selected as part of an eligible extractive summary and has the value of zero otherwise, and L_jis the per-party utterance count threshold value for the jth party.

The similarity-based constraint may require that if the covered subset for a particular eligible extractive summary comprises a particular interaction utterance that is in a particular per-party utterance, then the covered subset for the particular eligible extractive summary should further comprise each other interaction utterance that is in any per-party utterance subset other than the particular per-party utterance subset and that has a threshold-satisfying utterance similarity measure with respect to the particular interaction utterance. In other words, the similarity-based constraint may require that, if a first interaction utterance from a first interaction party is included in an eligible extractive summary, then each interaction utterance of any other interaction party whose similarity measure with respect to the first interaction utterance satisfies (e.g., exceeds, exceeds or is equal to, and/or the like) an utterance similarity measure threshold value should also be included in the eligible extractive summary. Thus, if a first interaction utterance that is deemed to have threshold-satisfying similarity with respect to a second interaction utterance is included in an extractive summary, and further if the extractive summary does not include the second interaction utterance, then extractive summary may be deemed ineligible under the similarity-based constraint. In some embodiments, the similarity-based constraint is deemed satisfied if the condition described by the equation x_p^a+x_q^b≥2 if sim(s_p^a, s_q^b)≥0.4 is satisfied, where s_p^ais the ath interaction utterance of the pth party, s_q^bis the bth interaction utterance of the qth party, x_p^ais an index variable that has the value of one if s_p^ais selected as part of an eligible extractive summary and has the value of zero otherwise, x_q^bis an index variable that has the value of one if s_b^qis selected as part of an eligible extractive summary and has the value of zero otherwise, sim(s_p^a, s_q^b) is the similarity measure for s_p^aand s_q^b, and 0.4 is an example of an utterance similarity measure threshold value.

In some embodiments, performing the ILP operation includes performing the following constrained optimization problem, which can be interpreted in light of the below-provided Table 1:

<tables id="TABLE-US-00001" num="00001"><table frame="none" colsep="0" rowsep="0"><tgroup align="left" colsep="0" rowsep="0" cols="1"><colspec colname="1" colwidth="217pt" align="left" /><thead><row><entry namest="1" nameend="1" align="center" rowsep="1" /></row></thead><tbody valign="top"><row><entry>max(Σ_(j=1)²Σ_(i=1)^NjIQ(s_i^j) LQ(s_i^j). x_i^j+ Σ_(k=1)^MScore(k), y_k)</entry></row><row><entry>subject to</entry></row><row><entry>1) Σ_(j=1)²Σ_{(j ∈ Tkj)}x_i^j≥ y_k(if the content word k is selected to be</entry></row><row><entry>included in the summary, then at least one utterance in which this content</entry></row><row><entry>word is present, is to be selected (in each party utterance summary))</entry></row><row><entry>2) Σ_{(k ∈} C_i^j₎y_k≥ |C_i^j| × x_i^j, j = 1, 2 (if a particular utterance is selected to</entry></row><row><entry>be included in the summary, then all the content words in that utterance</entry></row><row><entry>are also included in the summary (in each party utterance summary))</entry></row><row><entry>3) )</entry></row><row><entry>4) Σ_(i=1)^Njx_i^j≥ 1, j = 1, 2 (Each party summary is non non-empty)</entry></row><row><entry>5) Σ_(i=1)x_i^j≤ L_jj = 1, 2 (Each party utterance summary cannot exceed a</entry></row><row><entry>certain length)</entry></row><row><entry>6) x_p¹+ x_q²≥ 2, if sim(s_p¹, s_q²) ≥ 0.4, for all p = 1, . . . , N₁, q =</entry></row><row><entry>1, . . . , N₂(a pair of sentences from each party dialogue is selected if</entry></row><row><entry>similarity of the pair exceeds a threshold)</entry></row><row><entry>x_i^j∈ {0, 1}, j = 1, 2</entry></row><row><entry>y_k∈ {0, 1}, k = 1, 2, . . . , M</entry></row><row><entry namest="1" nameend="1" align="center" rowsep="1" /></row></tbody></tgroup></table></tables>

<tables id="TABLE-US-00002" num="00002"><table frame="none" colsep="0" rowsep="0"><tgroup align="left" colsep="0" rowsep="0" cols="2"><colspec colname="1" colwidth="42pt" align="left" /><colspec colname="2" colwidth="175pt" align="left" /><thead><row><entry namest="1" nameend="2" rowsep="1">TABLE 1</entry></row><row><entry namest="1" nameend="2" align="center" rowsep="1" /></row><row><entry>Notations</entry><entry>Meaning</entry></row><row><entry namest="1" nameend="2" align="center" rowsep="1" /></row></thead><tbody valign="top"><row><entry>M</entry><entry>Total no of content words</entry></row><row><entry>sim(s_p¹, s_q²)</entry><entry>Similarity of sentence s_p¹and sentence s_q²</entry></row><row><entry>N_j</entry><entry>No of sentences considered in the utterance group</entry></row><row>&lt;entry /&gt;<entry>for the jth party</entry></row><row><entry>S_i^j</entry><entry>ith sentence in the j-th party</entry></row><row><entry>IQ(s_i^j)</entry><entry>The informativeness of ith sentence unit in j-th party</entry></row><row><entry>LQ(s_i^j)</entry><entry>The linguistic quality of ith sentence unit in j-th party</entry></row><row><entry>Score(k)</entry><entry>The CoreRank score of kth content word.</entry></row><row><entry>x_i^j</entry><entry>Indicator variable for the i-th sentence for the j-th party</entry></row><row><entry>y_k</entry><entry>Indicator variable for the kth content word.</entry></row><row><entry>C_i^j</entry><entry>Set of content words in ith sentence in j-th party dialogues</entry></row><row><entry>S_k</entry><entry>Set of sentences containing the content word k.</entry></row><row><entry>T_k^j</entry><entry>set of sentences in the j-th party where content word</entry></row><row>&lt;entry /&gt;<entry>k is present</entry></row><row><entry>Imp(s_i^j)</entry><entry>The importance of ith sentence unit in j-th party</entry></row><row><entry namest="1" nameend="2" align="center" rowsep="1" /></row></tbody></tgroup></table></tables>

In some embodiments, step/operation <b>603</b> may be performed in accordance with the process depicted in <figref idref="DRAWINGS">FIG. 7</figref>. The process that is depicted in <figref idref="DRAWINGS">FIG. 7</figref> begins at step/operation <b>701</b> when the predictive data analysis computing entity <b>106</b> generates eligible extractive summaries. An eligible extractive summary describes a segment of a multi-party interaction transcript that satisfies one or more required optimization constraints, where a segment of a multi-party interaction transcript may include at least one a subset of interaction utterances of the multi-party interaction transcript and a subset of content words of the multi-party interaction transcript. Examples of required optimization constraints that may be satisfied by eligible extractive summaries include as an utterance coverage constraint, a content word coverage constraint, a party coverage constraint, a party size constraint, and a similarity-based constraint. As described above: (i) the utterance coverage constraint may require that, if the covered subset for a particular eligible extractive summary comprises a particular covered content word, then the covered subset for the particular eligible extractive summary should further comprise at least one interaction utterance of the plurality of interaction utterances that is associated with the particular covered content word; (ii) the content word coverage constraint may require that, if the covered subset for a particular eligible extractive summary comprises a particular interaction utterance having one or more per-utterance content words, then the covered subset for the particular eligible extractive summary should comprise all of the one or more per-utterance content words; (iii) the party coverage constraint may require that each covered subset for an eligible extractive summary comprises at least one interaction utterance from each per-party utterance subset of the one or more per-party utterance subsets; (iv) the party size constraint may require that each covered subset for an eligible extractive summary comprises a below-threshold count of interaction utterances from each per-party utterance subset of the one or more per-party utterance subsets; and (iv) the similarity-based constraint may require that if the covered subset for a particular eligible extractive summary comprises a particular interaction utterance that is in a particular per-party utterance, then the covered subset for the particular eligible extractive summary should further comprise each other interaction utterance that is in any per-party utterance subset other than the particular per-party utterance subset and that has a threshold-satisfying utterance similarity measure with respect to the particular interaction utterance.

At step/operation <b>702</b>, the predictive data analysis computing entity <b>106</b> determines an overall summary utility measure for each eligible extractive summary. The overall summary utility measure for an extractive summary may be a measure of usefulness of the extractive summary that is determined based at least in part on each information quality measure for an interaction utterance of the plurality of interaction utterances that is in the covered subset for the eligible extractive summary and each linguistic quality measure for an interaction utterance of the plurality of interaction utterances that is in the covered subset for the eligible extractive summary. As described above, the overall summary utility measure may be determined based at least in part on the output of the equation Σ_j=1^zΣ_i=1^N^{<sub2>j</sub2>}IQ(s_i^j)·x_i^j+Σ_k=1^Mscore(k)·y_k, where j is an iteration variable that iterates over z interaction parties associated with the multi-party interaction transcript, i is an iteration variable that iterates over N_jutterances of the jth interaction party, s_i^jis the ith interaction utterance of the jth interaction party, IQ(s_i^j) is the information quality measure of s_i^j, LQ(s_i^j) is the linguistic quality measure of s_i^j, x_i^jis an index variable that has the value of one if s_i^jis selected as part of an eligible extractive summary and has the value of zero otherwise, k is an iteration variable that iterates over M content words, score(k) is the weight of the kth content word, and y_kis an index variable that has the value of one if the kth content word is selected as part of an eligible extractive summary and has the value of zero otherwise.

At step/operation <b>703</b>, the predictive data analysis computing entity <b>106</b> generates the optimal extractive summary based at least in part on each overall summary utility measure for an eligible extractive summary. In some embodiments, the predictive data analysis computing entity <b>106</b> selects the eligible extractive summary that has the highest optimal extractive summary as the optimal extractive summary. In some embodiments, the optimal extractive summary includes at least one of one or more selected interaction utterances, one or more selected content words, and one or more selected constructed phrases that are constructed based at least in part on the one or more selected content words.

In some embodiments, subsequent to generating the optimal extractive summary, the predictive data analysis computing entity <b>106</b> generates per-party optimal summaries by dividing the interaction utterances in the optimal extractive summary into two segments based at least in part on associations of the interaction utterances. For example, if an optimal extractive summary includes five interaction utterances from party A and six interaction utterances from party B, then the updated optimal extractive summary may include a first per-party optimal summary that describes the five interaction utterances associated with the party A and a second per-party optimal summary that describes the six interaction utterances associated with the party B.

Returning to <figref idref="DRAWINGS">FIG. 4</figref>, at step/operation <b>405</b>, the predictive data analysis computing entity <b>106</b> updates the optimal extractive summary by performing one or more post-processing operations. In some embodiments, subsequent to generating the optimal extractive summary, the predictive data analysis computing entity <b>106</b> ranks the interaction utterances in the optimal extractive summary based at least in part on per-utterance weights determined for each interaction utterance in the optimal extractive summary, where the per-utterance weight for a particular interaction utterance in the optimal extractive summary may be determined based at least in part on at least one of (e.g., based at least in part on a combination/product of) the information quality measure for the particular interaction utterance and the linguistic quality measure for the particular interaction utterance.

In some embodiments, subsequent to generating the optimal extractive summary, the predictive data analysis computing entity <b>106</b> generates per-party optimal summaries by dividing the interaction utterances in the optimal extractive summary into two segments based at least in part on associations of the interaction utterances. For example, if an optimal extractive summary includes five interaction utterances from party A and six interaction utterances from party B, then the updated optimal extractive summary may include a first per-party optimal summary that describes the five interaction utterances associated with the party A and a second per-party optimal summary that describes the six interaction utterances associated with the party B. In some embodiments, subsequent to generating a per-party optimal summary, the predictive data analysis computing entity <b>106</b> ranks the interaction utterances in the per-party optimal summary based at least in part on per-utterance weights determined for each interaction utterance in the per-party optimal summary, where the per-utterance weight for a particular interaction utterance in the per-party optimal summary may be determined based at least in part on at least one of (e.g., based at least in part on a combination/product of) the information quality measure for the particular interaction utterance and the linguistic quality measure for the particular interaction utterance.

At step/operation <b>406</b>, the predictive data analysis computing entity <b>106</b> performs one or more summary-based actions based at least in part on the optimal extractive summary. For example, the predictive data analysis computing entity <b>106</b> may display the optimal extractive summary using a summary display user interface, such as the summary display user interface <b>800</b> of <figref idref="DRAWINGS">FIG. 8</figref>. As depicted in <figref idref="DRAWINGS">FIG. 8</figref>, the summary display user interface <b>800</b> displays the per-party optimal summary <b>801</b> for a first interaction party, the per-party optimal summary <b>802</b> for a second interaction party, the top content words <b>803</b> of the corresponding multi-party interaction transcript, and the constructed/inferred phrase <b>804</b> of the corresponding multi-party interaction transcript.

VI. Conclusion

Many modifications and other embodiments will come to mind to one skilled in the art to which this disclosure pertains having the benefit of the teachings presented in the foregoing descriptions and the associated drawings. Therefore, it is to be understood that the disclosure is not to be limited to the specific embodiments disclosed and that modifications and other embodiments are intended to be included within the scope of the appended claims. Although specific terms are employed herein, they are used in a generic and descriptive sense only and not for purposes of limitation.