NL2012433B1

Data clustering.

Abstract

The present disclosure describes methods and systems by which memory-efficient clustered data structures of related data entities (or "clusters") may be generated. Generation of clusters may begin by automatic generation of an initial, seed data entity of interest. Clusters of related data entities are generated based on the seed and according to cluster generation strategies (or cluster strategies). Seeds and related data entities may be accessed from various databases and data sources. The data analysis system may bring together data from multiple data sources in order to build clusters.

NL2012433B1, drawing sheet 1 — Sheet 1 of 10

Term

7.5 yearsto projected expiry

Projected expiry 14 March 2034, counted from filing; an application has no term until it is granted.

19 claims: 10 independent, 9 dependent

1
CONCLUSIES 1. Op een computer geïmplementeerde werkwijze voor het voortbrengen van een cluster met gerelateerde gegevensentiteiten, waarbij de werkwijze omvat:communiceren met één of meer elektronische gegevensopslagen waarin een veelheid gegevensentiteiten en respectievelijke gegevensentiteitattributen worden opgeslagen, waarbij de één of meer elektronische gegevensopslagen in communicatie staan met één of meer hardware computerprocessoren, waarbij de één of meer hardware computerprocessoren zijn ingericht met specifieke op een computer uitvoerbare instructies;het iteratief voortbrengen, voor elk van een veelheid van kiemgegevensentiteiten gekozen uit de veelheid van gegevensentiteiten, en door de één of meer hardware computerprocessoren, van een gegevensentiteitcluster door ten minste: het automatisch aanwijzen van de kiemgegevensentiteit als een initiële gegevensentiteit van de gegevensentiteitcluster;het benaderen, op basis van een clusterstrategie, van één of meer zoekprotocollen;het uitvoeren van ten minste een eerste van de één of meer zoekprotocollen op de één of meer elektronische gegevensopslagen om één of meer met de kiemgegevensentiteit geassocieerde gegevensentiteiten te identificeren, waarbij het uitvoeren van de eerste van de één of meer zoekprotocollen omvat: identificeren, door de één of meer hardware computerprocessoren, van ten minste één gegevensentiteitattribuut geassocieerd met de kiemgegevensentiteit;evalueren, door de één of meer hardware computerprocessoren, van de veelheid van gegevensentiteiten om de één of meer gegevensentiteiten te bepalen die de ten minste één gegevensentiteitattribuut delen met de kiemgegevensentiteit;het toevoegen van de één of meer gegevensentiteiten aan het gegevensentiteitcluster;het uitvoeren van ten minste een tweede van de één of meer zoekprofocollen op de één of meer elektronische gegevensopslagen om één of meer extra met de één of meer toegevoegde gegevensentiteiten geassocieerde gegevensentiteiten te identificeren, waarbij het uitvoeren van de tweede van de één of meer zoekprofocollen omvat: het identificeren, door de één of meer hardware computerprocessoren, van ten minste één gegevensentiteitattribuut, die geassocieerd is met ten minste één van de één of meer eerder aan het gegevensentiteitcluster toegevoegde gegevensentiteiten;en het evalueren, door de één of meer hardware computerprocessoren, van de veelheid van gegevensentiteiten om de één of meer extra gegevensentiteiten te bepalen die de ten minste één gegevensentiteitattribuut delen met de ten minste één van de één of meer eerder aan het gegevensentiteitcluster toegevoegde gegevensentiteiten;en toevoegen van de één of meer extra gegevensentiteiten aan het gegevensentiteitcluster.
2
Werkwijze volgens conclusie 1, omvattende het automatisch selecteren van de kiemgegevensentiteit.
3
Werkwijze volgens conclusie 1, omvattende het terughalen van de kiemgegevensentiteit in reactie op de selectie van de kiemgegevensentiteit door een gebruiker.
4
Werkwijze volgens één van de voorgaande conclusie, bovendien omvattend:vergelijken, door de één of meer hardware computerprocessoren, van gegevensentiteiten geassocieerd met het gegevensentiteitcluster met gegevensentiteiten geassocieerd met een tweede gegevensentiteitcluster;en in reactie op het bepalen dat ten minste één gegevensentiteit geassocieerd met het gegevensentiteitcluster een attribuut deelt met en/of is gerelateerd aan ten minste één gegevensentiteit geassocieerd met het tweede gegevensentiteitcluster, samenvoegen van het gegevensentiteitcluster en het tweede gegevensentiteitcluster.
5
Werkwijze volgens één van de voorgaande conclusie, waarbij de clusterstrategie één of meerdere zoekprofocollen omvat, en waarbij het vaststellen van één of meer extra gegevensentiteiten gerelateerd aan de kiemgegevensentiteit omvat:het uitvoeren van ten minste één zoekprotocol op de één of meer elektronische gegevensopslagen teneinde toegang te krijgen tot bepaalde gegevensentiteiten;en het evalueren van de bepaalde gegevensentiteiten die zijn teruggegeven door het ten minste één zoekprotocol teneinde één of meer extra gegevensentiteiten te identificeren voor opname in het gegevensentiteitcluster, waarbij de één of meer extra gegevensentiteiten het ten minste één gegevensent iteitattribuut delen met de kiemgegevens-entiteit.
6
Werkwijze volgens conclusie 5, waarbij het ten minste één zoekprotocol de gegevensentiteitkiem gebruikt als een gegevensparameterinvoer voor het ten minste één zoekprotocol .
7
Werkwijze volgens conclusie 5 of conclusie 6, waarbij een eerste zoekprotocol zoekt naar bepaalde gegevensentiteiten in een eerste elektronische gegevensopslag en een tweede zoekprotocol zoekt naar bepaalde gegevensentiteiten in een tweede elektronische gegevensopslag.
8
Werkwijze volgens conclusie 7, waarbij het eerste zoekprotocol bepaalde gegevensentiteiten teruggeeft die worden gebruikt als gegevensparameterinvoeren voor het tweede zoekprotocol.
9
Werkwijze volgens één van de conclusies 5 tot 8, bovendien omvattend:het identificeren, door de één of meer hardware computerprocessoren, van ten minste één andere gegevensentiteitattribuut geassocieerd met de één of meer extra gegevensentiteiten;het vaststellen, door de één of meer hardware computerprocessoren en op basis van de clusterstrategie en het van ten minste één andere gegevensentiteitattribuut, van één of meer tertiaire gegevensentiteiten gerelateerd aan de één of meer extra gegevensentiteiten;en het toevoegen, door de één of meer hardware computerprocessoren en op basis van de clusterstrategie, van de één of meer tertiaire gegevensentiteiten aan het gegevensentiteitcluster.
10
Werkwijze volgens één van de conclusies 5 tot en met 9, bovendien omvattend:het toekennen van een rangschikscore aan het gegevensentiteitcluster, waarbij de rangschikscore wordt gebruikt om het gegevensentiteitcluster te ordenen ten opzichte van een veelheid van ander gegevensentiteitclusters voortgebracht vanuit betreffende gegevensentiteitkiemen volgens de clusterstrategie .
11
Werkwijze volgens één van de voorgaande conclusies, waarbij de veelheid gegevensentiteiten betrekking hebben op financiële gegevens en ten minste één omvatten van:een rekeninggegevensentiteit, een transactiegegevensentiteit, een klantgegevensentiteit of een telefoonnummergegevensentiteit, waarbij de clusterstrategie is ingericht om gerelateerde gegevensentiteiten te identificeren voor het detecteren van mogelijke frauduleuze financiële activiteit, en waarbij het gegevensentiteitcluster geschikt is voor gebruik door een analist om verscheidene gegevensentiteiten vast te stellen die aan elkaar en aan mogelijke frauduleuze financiële activiteit zijn gerelateerd vast te stellen.
12
Op een computer geïmplementeerde werkwijze voor het prioriteren van een veelheid gegevensentiteitclusters, waarbij de werkwijze omvat:communiceren met één of meer elektronische gegevensopslagen waarin een veelheid gegevensentiteiten en betreffende gegevensentiteitattributen worden opgeslagen, waarbij de één of meer elektronische gegevensopslagen in communicatie staan met één of meer hardware computerprocessoren, waarbij de één of meer hardware computerprocessoren zijn ingericht met specifieke door de computer uit te voeren instructies;voortbrengen, door de één of meer hardware computerprocessoren, van een veelheid gegevensentiteitclusters waarin in elk ervan een verzameling gerelateerde gegevensentiteiten wordt opgeslagen, waarbij elk van de gegevensentiteitclusters wordt voortgebracht door: het identificeren en toevoegen van een kiemgegevensentiteit aan het gegevensentiteitcluster;en bepalen en toevoegen aan het gegevensentiteitcluster, op basis van een clusterstrategie van één of meer extra gegevensentiteiten gerelateerd aan de kiemgegevensentiteit;identificeren, door de één of meer hardware computerprocessoren, van een scorestrategie voor het prioriteren van de veelheid gegevensentiteitclusters;voor elk gegevensentiteitcluster: evalueren, door de één of meer hardware computerprocessoren en op basis van de scorestrategie, van het gegevensentiteitcluster;en toewijzen, door de één of meer hardware computerprocessoren en op basis van de evaluatie, van een score aan het gegevensentiteitcluster;en rangschikken van de veelheid van gegevensentiteitclusters volgens de betreffende toegekende scores;waarbij het toewijzen van een score aan het gegevensentiteitcluster omvat: het vaststellen van een veelheid basisscores voor het gegevensentiteitcluster;het vaststellen, op basis van de veelheid basisscores, van een totaalscore voor het gegevensentiteitcluster;en het toekennen van de totaalscore aan het gegevensentiteitcluster.
13
Werkwijze volgens conclusie 12, waarbij de aan elk gegevensentiteitcluster toegekende score overeenstemt met een mate van risico.
14
Werkwijze volgens conclusies 12 of 13, bovendien omvattend:presenteren, op een door een gebruiker te bekijken elektronisch beeldscherm, van een opsomming en/of samenvatting van één of meer van de veelheid gegevensentiteitclusters volgens de rangschikking.
15
Werkwijze volgens conclusie 14, waarbij de opsomming en/of samenvatting de kiemgegevensentiteit geassocieerd aan elk van de één of meer van de veelheid clusters presenteert.
16
Werkwijze volgens één van de conclusies 12 tot en met 15, bovendien omvattend:identificeren van een tweede scorestrategie voor het prioriteren van de veelheid gegevensentiteitclusters;voor elk gegevensentiteitcluster: evalueren van één of meer attributen van de verzameling gerelateerde gegevensentiteiten opgeslagen door het gegevensentiteitcluster volgens de tweede scorestrategie;en toewijzen van een tweede score aan het gegevensentiteitcluster;het herrangschikken van de veelheid van gegevensentiteitclusters volgens de betreffende toegekende tweede scores;en presenteren, op een door een gebruiker te bekijken elektronisch beeldscherm, van een opsomming en/of samenvatting van één of meer van de veelheid gegevensentiteitclusters volgens de herrangschikking.
17
Werkwijze volgens één van de conclusies 12 tot en met 16, waarbij de veelheid gegevensentiteiten betrekking hebben op financiële gegevens en ten minste één omvatten van:een rekeninggegevensentiteit, een transactiegegevensentiteit, een klantgegevensentiteit of een telefoonnummergegevensentiteit, waarbij de clusterstrategie is ingericht om gerelateerde gegevensentiteiten te identificeren voor het detecteren van mogelijke frauduleuze financiële activiteit, 5 en waarbij het gegevensentiteitcluster geschikt is voor gebruik door een analist om verscheidene aan elkaar en aan mogelijke frauduleuze financiële activiteit gerelateerde gegevensent iteiten vast te stellen.
18
Computerprogramma omvattende door een machine 10 leesbare instructies die, wanneer deze door een computer worden uitgevoerd, deze besturen om de werkwijze volgens één van de voorgaande conclusies te realiseren.
19
Inrichting ingericht om de werkwijze volgens één van de conclusies 1 tot en met 17 uit te voeren. 1/10

Independent claims19

182 paragraphs, as filed

Field of the Invention

This invention relates to data clustering.

Background to the Invention

In a fraud investigation an analyst may have to make decisions regarding selection of electronic data entities within an electronic collection of data. Such a collection of data may include a large number of data items (also referred to as data entities) that may or may not be related to one another, and which may be stored in an electronic data store or memory. For example, such a collection of data may include hundreds of thousands, millions, tens of millions, hundreds of millions, or even billions of data entities, and may consume significant storage and/or memory. Determination and selection of relevant data entities within such a collection of data may be extremely difficult for the analyst. Further, processing of such a large collection of data (for example, as an analyst uses a computer to sift and/or search through huge numbers of data entities) may be extremely inefficient and consume significant processing and/or memory resources.

Summary Of The Invention

Embodiments of the present disclosure relate to automatic generation of memory-efficient clustered data structures and, more specifically, to automatic selection of an initial data entity of interest, adding of the initial data entity to the memory-efficient clustered data structure (which may be referred to herein as a cluster), and determining and adding one or more related data entities to the cluster. In various embodiments, a generated cluster may include far fewer data entities than the collection of data described above, and the data entities included in the cluster may only include those data entities that are relevant to a particular investigation (for example, a fraud investigation).

Accordingly, in an embodiment, processing of the generated cluster may be highly efficient as compared to the collection of data described above. This may be because, for example, a given fraud investigation by an analyst (for example, as the analyst sifts and/or searches through data entities of a cluster) may only require storage in memory of a single cluster data structure. Further, a number of data entities in a cluster may be several orders of magnitude smaller than in the entire electronic collection of data described above because only data entities related to each other are included in the cluster.

According to various embodiments, the present disclosure describes methods and systems by which memoryefficient clustered data structures of related data entities (or clusters) may be generated. Generation of clusters may begin by automatic generation, determination, and/or selection of an initial data entity of interest, called a seed. A data entity may include any data, information or things, such as a person, a place, an organization, an account, a computer, an activity, and event, and/or the like. Seeds may be automatically selected/generated according to various seed determination strategies, and clusters of related data entities may be generated based on those seeds and according to cluster generation strategies (or cluster strategies). The system may further generate a score, multiple scores, and/or metascores for each generated cluster, and may optionally rank or prioritize the generated clusters based on the generated metascores. High priority clusters may be of greater interest to an analyst as they may contain related data entities that meet particular criteria related to the analyst's investigation. In an embodiment, the system may enable an analyst to advantageously start an investigation with a prioritized cluster including many related data entities rather than a single randomly selected data entity. Further, as described above, processing requirements of the analyst's investigation may thereafter be highly efficient as compared to processing of the huge collection of data described above.

As mentioned above, this is because, for example, a given investigation by an analyst may only require storage in memory of a single cluster, and further, a number of data entities in a cluster may be several orders of magnitude smaller than in the entire electronic collection of data described above because only data entities related to each other are included in the cluster.

One embodiment of the present invention includes a method for generating a cluster of related data from a seed. This method may generally include retrieving a seed and adding the seed to a first cluster. This method may further include retrieving a cluster strategy referencing one or more data bindings. Each data binding specifies a search protocol for retrieving data. For each of the one or more data bindings, data parameters input to the search protocol are identified, the search protocol is performed using the identified data parameters, and data returned by the search protocol is evaluated for inclusion in the first cluster.

In a particular embodiment, the search protocol for a first one of the data bindings uses the seed as the data parameters input to the search protocol. Further, the search protocol for a first one of the data bindings returns data used as the identified data parameters input to the search protocol for a second one of the data bindings. This process may iteratively continue until all data bindings in the search strategy have been executed for all available data obtained while growing the first cluster.

In a particular embodiment, this method may further include retrieving a second cluster from a plurality of clusters, comparing data from the first cluster to data from the second cluster, and determining whether to merge the first cluster and the second cluster based on the comparison.

Other embodiments include, without limitation, a computer-readable medium that includes instructions that enable a processing unit to implement one or more aspects of the disclosed methods as well as a system having a processor, memory, and application programs configured to implement one or more aspects of the disclosed methods.

Advantageously, according to various embodiments, the disclosed techniques provide a more effective starting point for an investigation of data entities of various types. An analyst may be able to start an investigation from a cluster of related data entities instead of an individual data entity, which may reduce the amount of time and effort reguired to perform the investigation. The disclosed techniques may also, according to various embodiments, provide a prioritization of multiple clusters. For example, the analyst may also able to start the investigation from a high priority cluster, which may allow the analyst to focus on the most important investigations. In each case, the processing and memory requirements of such an investigation may be significantly reduced due to the creation and use of highly efficient cluster data structures of related data entities.

Brief Description Of The Drawings

So that the manner in which the above recited features of the present invention can be understood in detail, a more particular description of the invention, briefly summarized above, may be had by reference to embodiments, some of which are illustrated in the appended drawings. It is to be noted, however, that the appended drawings illustrate only typical embodiments of this invention and are therefore not to be considered limiting of its scope, for the invention may admit to other equally effective embodiments.

Figure 1 is a block diagram illustrating an example data analysis system, according to one embodiment.

Figure 2 illustrates the generation of clusters by the data analysis system, according to one embodiment.

Figures 3A-3C illustrate the growth of a cluster of related data entities, according to one embodiment.

Figure 4 illustrates the ranking of clusters by the data analysis system, according to one embodiment of the present invention.

Figure 5 illustrates an example cluster analysis user interface (ÜI), according to one embodiment.

Figure 6 is a flow diagram of method steps for generating clusters, according to one embodiment.

Figure 7 is a flow diagram of method steps for scoring clusters, according to one embodiment.

Figure 8 illustrates components of a server computing system, according to one embodiment.

Detailed Description

According to various embodiments, the present disclosure describes methods and systems by which memoryefficient clustered data structures of related data entities (or clusters) may be generated. Generation of clusters may begin by automatic generation, determination, and/or selection of an initial data entity of interest, called a seed. As mentioned above, a data entity may include any data, information or things, such as a person, a place, an organization, an account, a computer, an activity, and event, and/or the like. Seeds may be automatically selected/generated according to various seed determination strategies, and clusters of related data entities may be generated based on those seeds and according to cluster generation strategies (or cluster strategies). Seeds and related data entities may be accessed from various databases and data sources including, for example, databases maintained by financial institutions, government entities, private entities, public entities, and/or publicly available data sources. Such databases and data sources may include a variety of information and data, such as, for example, personal information, financial information, tax-related information, computer network-related data, and/or computer-related activity data, among others. Further, the databases and data sources may include various relationships that link and/or associate data entities with one another. Various data entities and relationships may be stored across different systems controlled by different entities and/or institutions. According to various embodiments, the data analysis system may bring together data from multiple data sources in order to build clusters.

In various embodiments, the methods and systems of the present disclosure may enable a user to efficiently perform analysis and investigations of various data entities and clusters of data entities. For example, the system may enable a user (also referred to herein as an analyst) to perform various financial and security investigations related to a seed (for example, an initial data entity or data object). As described above, the processing and memory requirements of such an investigation may be significantly reduced due to the creation and use of highly efficient cluster data structures of related data entities.To perform financial and security investigations related to the seed, an analyst may have to search several layers of related data entities. For example, the analyst could investigate data entities related to a seed credit card account, by discovering the customer identifiers associated with the credit card account, the phone numbers associated with those customer identifiers, the additional customer identifiers associated with those phone numbers, and finally the additional credit card accounts associated with those additional customer identifiers. If the seed credit card account were fraudulent, then the analyst could determine that the additional credit card accounts could also be fraudulent. In such an investigation, the analyst would discover the relationship between the additional credit card accounts and the seed credit card accounts through several layers of related data entities. This technique is particularly valuable for investigations where the relationship between data entities could include several layers and would be difficult to identify.

In one embodiment, the data analysis system automatically discovers data entities related to a seed and stores the resulting relationships and related data entities together in a cluster. A cluster generation strategy specifies what searches to perform at each step of the investigation process. The searches produce layers of related data entities to add to the cluster. Thus, the analyst starts an investigation with the resulting cluster, instead of the seed alone. Starting with the cluster, the analyst may form opinions regarding the related data entities, conduct further analysis of the related data entities, or may query for additional related data entities. Further, for numerous such seeds and associated investigations, the data analysis system may prioritize the clusters based upon an aggregation of characteristics of the related data entities within the clusters. The data analysis system then displays summaries of the clusters. The summaries may be displayed according to the prioritization. The prioritization may assist the analyst in selecting what clusters to investigate.

In the following description, numerous specific details are set forth to provide a more thorough understanding of the present invention. However, it will be apparent to one of skill in the art that the present invention may be practiced without one or more of these specific details.

Figure 1 is a block diagram illustrating an example data analysis system 100, according to one embodiment of the present invention. As shown, the data analysis system 100 includes an application server 115 running on a server computing system 110, a client 135 running on a client computer system 130, and at least one database 140. Further, the client 135, application server 115, and database 140 may communicate over a network 150, e.g., to access cluster data sources 160.

The application server 115 includes a cluster engine 120 and a workflow engine 125. The cluster engine 120 is configured to build one or more clusters of related data entities, according to a defined analysis strategy. The cluster engine 120 may read data from a variety cluster data sources 160 to generate clusters from seed data. Once created, the resulting clusters may be stored on the server computer 110 or on the database 140. The operations of the cluster engine 120 are discussed in detail below in conjunction with Figures 2 and 3.

The cluster engine 120 is configured to score the clusters, according to a defined scoring strategy. The score may indicate the importance of analyzing the cluster. For instance, the cluster engine 120 could execute a scoring strategy that aggregates the account balances of credit card accounts within the cluster. If the cluster included a larger total balance than other clusters, then the cluster could be a greater liability for the financial institution. Thus, the cluster would be more important to analyze and would receive a higher score. In one embodiment, the cluster engine 120 organizes and presents the clusters according to the assigned scores. The cluster engine 120 may present summaries of the clusters and/or interactive representations of the clusters within the cluster analysis UI. For example, the representations may provide visual graphs of the related data entities within the clusters. The cluster engine 120 may generate the cluster analysis UI as a web application or a dynamic web page displayed within the client 135. The cluster engine 120 also allows an analyst to create tasks associated with the clusters. The operations of the cluster engine 120 are discussed in detail below in conjunction with Figures 4 and 5. In one embodiment, the cluster engine 120 generates clusters automatically, for subsequent review by analysts. Analysts may also assign tasks to themselves via a workflow UI. The workflow engine 125 consumes scores generated by the cluster engine 120. For example, the workflow engine 125 may present an analyst with clusters generated, scored, and ordered by the cluster engine 120.

The client 135 represents one or more software applications configured to present data and translate input, from the analyst, into requests for data analyses by the application server 115. In one embodiment, the client 135 and the application server 115 are coupled together. However, several clients 135 may execute on the client computer 130 or several clients 135 on several client computers 130 may interact with the application server 115. In one embodiment, the client 135 may be a browser accessing a web service.

While the client 135 and application server 115 are shown running on distinct computing systems, the client 135 and application server 115 may run on the same computing system. Further, the cluster engine 120 and the workflow engine 125 may run on separate applications servers 115, on separate server computing systems, or some combination thereof. Additionally, a history service may store the results generated by an analyst relative to a given cluster

In one embodiment, the cluster data sources 160 provide data available to the cluster engine to create clusters from a set of seeds. Such data sources may include relational data sources, web services data, XML data, etc.

For example, the data sources may be related to customer account records stored by a financial institution. In such a case, the data sources may include a credit card account data, bank account data, customer data, and transaction data. The data may include data attributes such as account numbers, account balances, phone numbers, addresses, and transaction amounts, etc. Of course, cluster data sources 160 is included to be representative of a variety of data available to the server computer system 110 over network 150, as well as locally available data sources.

The database 140 may be a Relational Database Management System (RDBMS) that stores the data as rows in relational tables. While the database 140 is shown as a distinct computing system, the database 140 may operate on the same server computing system 110 as the application server 115.

Figure 2 illustrates the generation of clusters by data analysis system 200, according to one embodiment . As shown, the data analysis system 200 interacts with a seed list 210, a cluster list 250, and a cluster strategy store 230.

The seed list 210 includes seeds 212-1, 212-2 ... 212-S and the cluster list 250 includes clusters 252-1, 252-2... 252-C. The cluster engine 120 is configured as a software application or thread that generates the clusters 252-1, 252-2... 252-C from the seeds 212-1, 212-2 ... 212-S.

Seeds 212 are the starting point for generating a cluster 252. To generate a cluster, the cluster engine 120 retrieves a given seed 212 from the seed list 210. The seed 212 may be an arbitrary data entity within the database 140, such as a customer name, a customer social security number, an account number, or a customer telephone number.

The cluster engine 120 generates the cluster 252 from the seed 212. In one embodiment, the cluster engine 120 generates the cluster 252 as a collection of data entities and the relationships between the various data entities. As noted above, the cluster strategy executes data bindings in order to add each additional layer of objects to the cluster. For example, the cluster engine 120 could generate the cluster 252 from a seed credit card account. The cluster engine 120 first adds the credit card account to the cluster 252. The cluster engine 120 could then add customers related to the credit card account to the cluster 252. The cluster engine 120 could complete the cluster 252 by adding additional credit card accounts related to those customers. As the cluster engine 120 generates the cluster 252, the cluster engine 120 stores the cluster 252 within the cluster list 250. The cluster 252 may be stored as a graph data structure. The cluster list 250 may be a collection of tables in the database 140. In such a case, there may be a table for the data entities of the cluster 252, a table for the relationships between the various data entities, a table for the attributes of the data entities, and a table for a score of the cluster 252. The cluster list 250 may include clusters 252 from multiple investigations. Note that the cluster engine 120 may store portions of the cluster 252 in the cluster list 250 as the cluster engine 120 generates the cluster 252. Persons skilled in the art will recognize that many technically feasible techniques exist for creating and storing graph data structures .

The cluster strategy store 230 includes cluster strategies 232-1, 232-2... 232-N. Each cluster strategy may include references 235 to one or more data bindings 237. As noted, each data binding may be used to identify data that may grow a cluster (as determined by the given search strategy 232). The cluster engine 120 executes a cluster strategy 232 to generate the cluster 252. Specifically, the cluster engine 120 executes the cluster strategy 232 selected by an analyst. The analyst may submit a selection of the cluster strategy 232 to the cluster engine 120 through the client 135.

Each cluster strategy 232 is configured as to perform an investigation processes for generating the cluster 252. Again, e.g., the cluster strategy 232 may include references 235 to a collection of data bindings executed to add layer after layer of data to a cluster. The investigation process includes searches to retrieve data entities related to the seed 212. For example, the cluster strategy 232 could start with a possibly fraudulent credit card account as the seed 212. The cluster strategy 232 would search for customers related to the credit card account, and then additional credit card accounts related to those customers. A different cluster strategy 232 could search for customers related to the credit card account, phone numbers related to the customers, additional customers related to the phone numbers, and additional credit card accounts related to the additional customers .

In one embodiment, the cluster strategy 232 includes a reference to at least one data binding 237. The cluster engine 120 executes the search protocol of specified by the data binding 237 to retrieve data, and the data returned by a given data binding forms a layer within the cluster 252. For instance, the data binding 237 could retrieve sets of customers related to an account by an account owner attribute. The data binding 237 retrieves the set of related data entities from a data source. For instance, the data binding 237-1 could define specify a database query to perform against a database. Likewise, the data binding 237-2 could define a connection to a remote relational database system and the data binding 237-3 could define a connection and query against a third-party web service. Once retrieved, the cluster strategy 232 may evaluate whether the returned data should be added to a cluster being grown from a given seed 212. Multiple cluster strategies 232 may reference a given data binding 237. The analyst can update the data binding 237, but typically updates the data binding 237 only if the associated data source changes. A cluster strategy 232 may also include a given data binding 237 multiple times. For example, executing a data binding 237 using one seed 212 may generate additional seeds for that data binding 237 (or generate seeds for another data binding 237). More generally, different cluster strategies 232-1, 232-2 ... 232-N may include different arrangements of various data bindings 237 to generate different types of clusters 252.

The cluster strategy 232 may specify that the cluster engine 120 use an attribute from the related data entities retrieved with one data binding 237, as input to a subsequent data binding 237. The cluster engine 120 uses the subsequent data binding 237 to retrieve a subsequent layer of related date entities for the cluster 252. For instance, the cluster strategy 232 could specify that the cluster engine 120 retrieve a set of credit card account data entities with a first data binding 237-1. The cluster strategy 232 could also specify that the cluster engine 120 then use the account number attribute from credit card account data entities as input to a subsequent data binding 237-2. The cluster strategy 232 may also specify filters for the cluster engine 120 to apply to the attributes before performing the subsequent data binding 237. For instance, if the first data binding 237-1 were to retrieve a set of credit card account data entities that included both personal and business credit card accounts, then the cluster engine 120 could filter out the business credit card accounts before performing the subsequent data binding 237-2.

In operation, the cluster engine 120 generates a cluster 252-1 from a seed 212-1 by first retrieving a cluster strategy 232. Assuming that the analyst selected a cluster strategy 232-2, then the cluster engine 120 would retrieve the cluster strategy 232-2 from the cluster strategy store 230.

The cluster engine 120 could then retrieve the seed 212-1 as input to the cluster strategy 232-2. The cluster engine 120 would execute the cluster strategy 232-2 by retrieving sets of data by executing data bindings 237 referenced by the cluster strategy 232-2. For example, the cluster strategy could execute data bindings 237-1, 237-2, and 237-3. The cluster engine 120 evaluates data returned by each data binding 237 to determine whether to use that data to grow the cluster 252-1. The cluster engine 120 may then use elements of the returned data as input to the next data binding 237. Of course, a variety of execution paths are possible for the data bindings 237. For example, assume one data binding 237 returned a set of phone numbers. In such a case, another data binding 237 could evaluate each phone number individually. As another example, one data binding 237 might use input parameters obtained by executing multiple, other data bindings 237. More generally, the cluster engine 120 may retrieves data for each data binding referenced by the cluster strategy 232-2. The cluster engine 120 then stores the complete cluster 252-1 in the cluster list 250.

As the cluster engine 120 generates the clusters 2521, 252-2... 252-C from seeds 212-1, 212-2... 212-S, the cluster list 250 may include overlapping clusters 252. Two clusters 252-1 and 252-C overlap if both clusters 252-1 and 252-C include a common data entity. Oftentimes, a larger cluster 252 formed by merging two smaller clusters 252-1 and 252-C may be a better investigation starting point than the smaller clusters 252-1 and 252-C individually. The larger cluster 252 may provide additional insight or relationships, which may not be available if the two clusters 252-1 and 252-C remain separate.

In one embodiment, the cluster engine 120 includes a resolver 226 that is configured to detect and merge two overlapping clusters 252 together. The resolver 226 compares the data entities within a cluster 252-1 to the data entities within each one of the other clusters 252-2 through 252-C. If the resolver 226 finds the same data entity within the cluster 252-1 and a second cluster 252-C, then the resolver 226 may merge the two clusters 252-1 and 252-C into a single larger cluster 252. For example, the cluster 252-1 and cluster 252-C could both include the same customer. The resolver 226 would compare the data entities of cluster 252-1 to the data entities of cluster 252-C and detect the same customer in both clusters 252. Upon detecting the same customer in both clusters 252, the resolver 226 could merge the cluster 252-1 with cluster 252-C. The resolver 226 may test each pair of clusters 252 to identify overlapping clusters 252. Although the larger clusters 252 may be better investigation starting points, an analyst may want to understand how the resolver 226 formed the larger clusters 252. The resolver 226, stores a history of each merge.

After the cluster engine generates a group of clusters from a given collection of seeds (and after merging or resolving the cluster), the cluster engine 120 may score, rank, or otherwise order the clusters relative to a scoring strategy 442.

In one embodiment, the analysis system 100, and more specifically, the cluster engine 120 receives a list of seeds to generate a group of clusters, subsequently ranked, ordered, and presented to analysts. That is, the cluster engine 120 consumes seeds generated by other systems. Alternatively, in other embodiments, cluster engine 120 may generate the seeds 212-1, 212-2... 212-S. For instance, the cluster engine 120 may include a lead generation strategy that identifies data entities as potential seeds 212. The lead generation strategy may apply to a particular business type, such as credit cards, stock trading, or insurance claims and may be run against a cluster data source 160 or an external source of information.

Figures 3A - 3C illustrate the growth of a cluster 252 of related data entities, according to one embodiment. As shown in Figure 3A, a cluster 252 includes a seed data entity 302, links 303-1 and 303-2, and related data entities 305-1 and 305-2. The cluster 252 is based upon a seed 212. The cluster engine 120 builds the cluster 252 by executing a cluster strategy 232 with the following searches:

• Find seed owner • Find all phone numbers related to the seed owner • Find all customers related to the phone numbers • Find all accounts related to the customers • Find all new customers related to the new accounts

Assuming that the seed 212 were a fraudulent credit card account, then the cluster engine 120 would add the credit card account to the cluster 252 as the seed data entity 302.

The cluster engine 120 would then use the account owner attribute of the credit card account as input to a data binding 237. The cluster engine 120 would execute the search protocol of data binding 237 to retrieve the customer data identifying the owner of the fraudulent credit card account. The cluster engine 120 would then add the customer data to the cluster 252 as the related data entity 305-1. The cluster engine 120 would also add the account owner attribute as the link 303-1 that relates the account number to the customer data of the owner. The cluster engine 120 would execute the next search of the cluster strategy 232 by inputting the customer identifier attribute of the customer data into a data binding 237 to retrieve a phone data. The cluster engine 120 would then add the phone data as the related data entity 305-2 and the customer identifier attribute as the link 303-2 between the customer data and the phone data. At this point in the investigation process, the cluster 252 would include the seed data entity 302, two links 303-1 and 303-2, and two related data entities 305-1 and 305-2. That is, the cluster 252 includes the fraudulent credit card account, the customer data of the owner of the credit card, and the phone number of the owner. By carrying the investigation process further, the cluster engine 120 could reveal further related information e.g., additional customers or potentially fraudulent credit card accounts.

Turning to Figure 3B, the cluster engine 120 would continue executing the cluster strategy 232 by searching for additional account data entities related to the phone number of the owner of the fraudulent credit card account. As discussed, the phone number would be stored as related data entity 305-2. The cluster engine 120 would input the phone owner attribute of the phone number to a data binding 237.

The cluster engine 120 would execute the search protocol of data binding 237 to retrieve the data of two additional customers, which the cluster engine 120 would store as related data entities 305-3 and 305-4. The cluster engine 120 would add the phone owner attribute as the links 303-3 and 304-4 between the additional customers and the phone number.

Figure 3C shows the cluster 252 after the cluster engine 120 performs the last step of the cluster strategy 232. For example, the cluster engine 120 would use the customer identifier attribute of the related data entity 305-3 and 3054 to retrieve and add additional account data entities as the related data entities 305-5 and 305-6. The cluster engine 120 would couple the related data entities 305-5 and 305-6 to the related data entities 305-3 and 305-4 with the customer identifier attributes stored as links 303-5 and 303-6. Thus, the cluster 252 would include six related data entities 305 related by six links 303, in addition to the seed data entity 302. The analyst could identify and determine whether the additional data account entities, stored as related data entities 305-3 and 305-4, represent fraudulent credit card accounts more efficiently, than if the analyst started an investigation with only the seed 212. As the foregoing illustrates, with the cluster engine 120 and cluster strategy 232, the analyst is advantageously able to start an investigation from a cluster 252 that already includes several related data entities 305.

Figure 4 illustrates the ranking of clusters 252 by the data analysis system 100 shown in Figure 1, according to one embodiment of the present invention. As shown, Figure 4 illustrates some of the same elements as shown in Figure 1 and Figure 2. In addition, Figure 4 illustrates a scoring strategy store 440, coupled to the workflow engine 125. The cluster engine 120 coupled to the cluster list 250. The scoring strategy store 440 includes scoring strategies 442-1, 442-2... 442-R.

The cluster engine 120 executes a scoring strategy 442 to score a cluster 252. For example, the cluster engine 120 may generate a cluster, via a cluster strategy/data bindings, and attempt to resolve it with existing clusters. Thereafter, the cluster engine 120 may score the resulting cluster with any scoring strategies associated with a given cluster generation strategy. In one embodiment, the score for a cluster may be a meta score generated as an aggregation of scores generated for different aspects, metrics, or data of a cluster. Ordering for a group of clusters, (according to a given scoring strategy) may be performed done on demand when requested by a client. Alternatively, the analyst may select of a scoring strategy 442 to the cluster engine 120 through the clientl35 or the analyst may include the selection within a script or configuration file. In other embodiments, the cluster engine 120 may execute several scoring strategies 442 to determine a combined score for the cluster 252.

The scoring strategy 442 specifying an approach for scoring a cluster 252. The score may indicate the relative importance or significance of a given cluster 252. For instance, the cluster engine 120 could execute a scoring strategy 442-1 to determine a score by counting the number of a particular data entity type within the cluster 252. Assume, e.g., a data entity corresponds to a credit account. In such a case, a cluster with a large number of accounts opened by a single individual (possibly within a short time) might correlate with a higher fraud risk. Of course, a cluster score may be related to a high risk of fraud based on the other data in the cluster, as approaite for a given case.

More generally, each scoring strategy 442 may be tailored based on the data in clusters created by a given cluster strategy 230 and the particular type of risk or fraud (or amounts at risk).

In operation, the cluster engine 120 scores a cluster 252-1 by first retrieving a scoring strategy 442. For example, assume a analyst selects scoring strategy 442-1. In response, the cluster engine 120 retrieves the scoring strategy 442-1. The cluster engine 120 also retrieves the cluster 252-1 from the cluster list 250. After determining the score of the cluster 252-1, the cluster engine 120 may store the score with the cluster 252-1 in the cluster list

250 .

The cluster engine 120 may score multiple clusters 252-1, 252-2... 252-C in the cluster list 250. The cluster engine 120 may also rank the clusters 252-1, 252-2... 252-C based upon the scores. For instance, the cluster engine 120 could rank the cluster 252-1, 252-2... 252-C from highest score to lowest score.

Figure 5 illustrates an example cluster analysis UI 500, according to one embodiment. As discussed, the workflow engine 125 is configured to present the cluster analysis UI 500. As shown, the cluster analysis UI 500 includes a lead box 510, a cluster strategy box 530, a cluster summary list 525, a cluster search box 520, and a cluster review window 515. The workflow engine 125 may generate the cluster analysis UI 500 as a web application or a dynamic web page displayed within the client 135.

The lead box 510 allows the analyst to select a seed list 210 or a suitable lead generation strategy. The lead generation strategy generates a seed list 210. The lead generation strategy may generate a seed list 210 from the database 140 or an external source of information (e.g., a cluster data source 160).

The cluster strategy box 530 displays the cluster strategies 232 that the cluster engine 120 ran against the seed list 210. The cluster engine 120 may execute multiple cluster strategies 232 against the seed list 210, so there may be multiple cluster strategies 232 listed in the cluster strategy box 530. The analyst may click on the name of a given cluster strategy 232 in the cluster strategy box 530 to review the clusters 252 that the cluster strategy 232 generated.

The workflow engine 125 displays summaries of the clusters 252 in the cluster summary list 525. For example, the summaries ,may include characteristics of the clusters 252, such as identifiers, the scores, or analysts assigned to analyze the clusters 252. The workflow engine 125 can select the clusters 252 for the display in the cluster summary list 525 according to those or other characteristics. For instance, the workflow engine 125 could display the summaries in the order of the scores of the clusters 252, where a summary of the highest scoring cluster 252 is displayed first.

The workflow engine 125 controls the order and selection of the summaries within the cluster summary list 525 based upon the input from the analyst. The cluster search box 520 includes a search text box coupled to a search button and a pull-down control. The analyst may enter a characteristic of a cluster 252 in the search text box and then instruct the workflow engine 125 to search for and display clusters 252 that include the characteristic by pressing the search button. For example, the analyst could search for clusters with a particular score. The pull-down control includes a list of different characteristics of the clusters 252, such as score, size, assigned analyst, or date created. The analyst may select one of the characteristics to instruct the workflow engine 125 to present the summaries of the clusters 252 arranged by that characteristic.

The workflow engine 125 is also configured to present details of a given cluster 252 within the cluster review window 515. The workflow engine 125 displays the details of the cluster 252, e.g., the score, or average account balances within a cluster, when the analyst clicks a mouse pointer on the associated summary within the cluster summary list 525.

The workflow engine 125 may present details of the cluster 252, such as the name of the analyst assigned to analyze the cluster 252, the score of the cluster 252, and statistics or graphs generated from the cluster 252. These details allow the analyst to determine whether to investigate the cluster 252 further. The cluster review window 515 also includes a button which may be clicked to investigate a cluster 252 within a graph and an assign button for assigning a cluster to an analyst.

The analyst can click a mouse pointer on the button to investigate the cluster 252 within an interactive graph.

The interactive representation is a visual graph of the cluster 252, where icons represent the entities of the cluster 252 and lines between the icons represent the links between entities of the cluster 252. For example, the workflow engine 125 could display the interactive graph of the cluster 252 similar to the representation of the cluster 252 in Figure 3C. The interactive representation allows the analyst to review the attributes of the related data entities or perform queries for additional related data entities.

An administrative user can click a mouse pointer on the assign button to assign the associated cluster 252 to an analyst. The workflow engine 125 also allows the administrative user to create tasks associated with the clusters 252, while the administrative user assigns the cluster 252. For example, the administrative user could create a task for searching within the three highest scoring clusters 252 for fraudulent credit card accounts. The workflow engine 125 may display the summaries in the cluster summary list 525 according to the names of the analysts assigned to the clusters 252. Likewise, the workflow engine 125 may only display summaries for the subset of the clusters 252 assigned to an analyst.

The interface shown in Figure 5 is included to illustrate one exemplary interface useful for navigating and reviewing clusters generated using the cluster engine 120 and the workflow engine 125. Of course, one of skill in the art will recognize that a broad variety of user interface constructs could be used to allow the analyst to select cluster strategies 232, scoring strategies 242, or seed generation strategies, initiate an investigation, or review and analyze the clusters 252. For example, the workflow engine 125 may display additional controls within the cluster analysis UI 500 for controlling the cluster generation process and selecting cluster strategies 232 or scoring strategies 242. Also, the workflow engine 125 may not display the lead box 510 or the options to select a lead generation strategy.

In addition, although the workflow engine 125 generates the cluster analysis UI 500, in different embodiments, the cluster analysis UI 500 is generated by a software application distinct from the workflow engine 125. Further, in different embodiments, the cluster review window 515 is configured to display a preview of the cluster 252 or additional statistics generated from the cluster 252. As such, an interactive representation of the cluster 252 may be presented in an additional UI or the cluster 252 may be exported to another software application for review by the analyst.

Figure 6 is a flow diagram of method steps for generating clusters, according to one embodiment. Although the method steps are described in conjunction with the systems of Figures 1 and 2, persons skilled in the art will understand that any system configured to perform the method steps, in any order, is within the scope of the present invention. Further, the method 600 may be performed in conjunction with method 700 for scoring a cluster, described below.

As shown, method 600 begins at step 605, where the cluster engine 120 retrieves a cluster strategy 232 and a seed 212. Once a cluster strategy is selected The cluster engine 120 identifies a list of seeds to build clusters using the selected cluster strategy. At step 610, the cluster engine 120 initializes a cluster 252 with one of the seeds in the list. The cluster 252 is stored as a graph data structure.

The cluster engine 120 initializes the graph data structure, and then adds the seed 212-1 to the graph data structure as the first data entity.

At step 615, the cluster engine 120 grows the cluster 252 by executing the search protocol of a data binding 237 from the cluster strategy 232-2. The cluster strategy 232-2 includes a series of data bindings 237 that the cluster engine 120 executes to retrieve related data entities. A given data binding 237 may include queries to execute against a cluster data source 160 using the seed as an input parameters. For example, if the seed 212-1 were an account number, then the data binding 237 might retrieve the data identifying the owner of the account with the account number. After retrieving this information, the cluster engine 120 would add the customer data entity to the cluster as a related data entity and the account owner attribute as the link between the seed 212-1 and the related data entity. After retrieving the related data entities, the cluster engine 120 adds them to the cluster 252.

At step 620, the cluster engine 120 determines if the cluster strategy 232-2 is fully executed. If so the method 600 returns to step 615 to execute additional data bindings for a given seed. Once the cluster strategy is executed for that seed, the cluster engine 120 may determine and assign a score to that cluster (relative to a specified scoring strategy). After generating clusters for a group of seeds, such clusters may be ordered or ranked based on the relative scores. Doing so allows an analyst to rapidly identify and evaluate clusters determined to represent a high risk of fraud (or having high amounts at risk).

At step 625, the cluster engine 120 stores the cluster 252 in cluster list 250. The cluster list 250 is a collection of tables within a relational database, where a table may include the seed and related data entities of the cluster 252 and another table may include links between the related data entities of the cluster 252. At step 630, the cluster engine 120 determines if there are more seeds 212 to analyze in the seed list 210. If so, the method 600 returns to step 605 to generate another cluster from the next seed. Otherwise, the method 600 ends. Note, while method 600 describes a single cluster being generated, one of skill in the art will recognize that the cluster generation process illustrated by method 600 may be performed in parallel.

Figure 7 is a flow diagram of method steps for scoring clusters, according to one embodiment. Although the method steps are described in conjunction with the systems of Figures 1 and 4, persons skilled in the art will understand that any system configured to perform the method steps, in any order, is within the scope of the present invention.

As shown, method 700 begins at step 705, where the cluster engine 120 retrieves a scoring strategy 442 and a cluster 252 (e.g., a cluster just created using the method 600 of Figure 6). In other cases, he cluster engine 120 may retrieve the scoring strategy 442 associated with a stored cluster. Other alternatives include an analyst selecting a scoring strategy 442 through the client 135, the cluster engine 120 via the cluster analysis UI 500, a script, or a configuration file. The cluster engine 120 retrieves the selected scoring strategy 442 from the scoring strategy store 440. The cluster engine 120 retrieves the cluster 252 from the cluster list 250.

At step 710, the cluster engine 120 executes the scoring strategy 442 against the cluster 252. The scoring strategy 442 specifies characteristics of the related data entities within the cluster 252 to aggregate. The cluster engine 120 executes the scoring strategy 442 by aggregated the specified characteristics together to determine a score. For instance, the cluster engine 120 could aggregate the account balances of related data entities that are account data entities. In such a case, the total amount of dollars included within the balances of the account data entities of the cluster 252 could be the score of the cluster 252.

At step 715, the cluster engine 120 stores the score with the cluster 252 in the cluster list 250. At step 720, the cluster engine 120 determines if there are more clusters 252 to score. For example, in one embodiment, a set of clusters may be re-scored using an updated scoring strategy.

In other cases, the cluster engine may score each cluster when it is created from a seed (based on a given cluster generation and corresponding scoring strategy). If more clusters remain to be scored (or re-scored), the method 700 returns to step 705.

At step 725, the cluster engine 125 ranks the clusters 252 according to the scores of the clusters 252. For example, after re-scoring a set of clusters (or after scoring a group of clusters generated from a set of seeds), the cluster engine 125 may rank the clusters 252 from highest score to lowest score. The ranking may be used to order a display of summaries of the clusters 252 presented to the analyst. The analyst may rely upon the ranking and scores to determine which clusters 252 to analyze first. The ranking and sorting may generally be performed on-demand when an analyst is looking for a cluster to investigate. Thus, the ranking need not happen at the same time as scoring. And further, the clusters may be scored (and later ranked) using different raking strategies.

Figure 8 illustrates components of a server computing system 110, according to one embodiment. As shown, the server computing system 110 includes, a central processing unit (CPU) 860, a network interface 850, a memory 820, and a storage 830, each connected to an interconnect (bus) 840. The server computing system 110 may also include an I/O device interface 870 connecting I/O devices 875 (e.g., keyboard, display and mouse devices) to the computing system 110. Further, in context of this disclosure, the computing elements shown in server computing system 110 may correspond to a physical computing system (e.g., a system in a data center) or may be a virtual computing instance executing within a computing cloud.

The CPU 860 retrieves and executes programming instructions stored in memory 820 as well as stores and retrieves application data residing in memory 820. The bus 840 is used to transmit programming instructions and application data between the CPU 860, I/O device interface 870, storage 830, network interface 850, and memory 820. Note that the CPU 860 is included to be representative of a single CPU, multiple CPUs, a single CPU having multiple processing cores, a CPU with an associate memory management unit, and the like. The memory 820 is generally included to be representative of a random access memory. The storage 830 may be a disk drive storage device. Although shown as a single unit, the storage 830 may be a combination of fixed and/or removable storage devices, such as fixed disc drives, removable memory cards, or optical storage, network attached storage (NAS), or a storage area-network (SAN).

Illustratively, the memory 820 includes a seed list 210, a cluster engine 120, a cluster list 250, and a workflow engine 125. The cluster engine 120 includes a cluster strategy 232-2. The particular cluster strategy 232-2 includes data bindings 237-1, 237-2, and 237-3, with which the cluster engine 120 accesses the cluster data source 160. The workflow engine 125 includes a scoring strategy 442-1.

Illustratively, the storage 830 includes a cluster strategy store 230, data bindings store 835, and a scoring strategy store 440. As discussed the cluster strategy store 230 may include a collection of different cluster strategies 232, such as cluster strategy 232-2. The cluster strategy store 230 may be a directory that includes the cluster strategies 232-1, 232-2... 232-N as distinct modules. The scoring strategy store 440 may include a collection of different scoring strategies 442, such as scoring strategy 442-2 and may also be a directory of distinct modules. The data binding store 835 includes data bindings 237-1, 237-2 ... 237-M, which may also be stored as distinct modules within a directory.

Although shown in memory 820, the seed list 210, cluster engine 120, cluster list 250, and workflow engine 125, may be stored in memory 820, storage 830, or split between memory 820 and storage 830. Likewise, copies of the cluster strategy 232-2, data binding 237-1, 237-2, and 237-3, and scoring strategy 442-2 may be stored in memory 820, storage 830, or split between memory 820 and storage 830.

Note, while financial fraud using credit card accounts is used as a primary reference example in the discussion above, one of ordinary skill in the art will recognize that the techniques described herein may be adapted for use with a variety of data sets. For example, information from data logs of online systems could be evaluated as seeds to improve cyber security. In such a case, a seed could be a suspicious IP address, a compromised user account, etc. From the seeds, log data, DHCP logs, IP blacklists packet captures, webapp logs, and other server and database logs could be used to create clusters of activity related to the suspicions seeds. Other examples include data quality analysis used to cluster transactions processed through a computer system (whether financial or otherwise).

Embodiments of the present disclosure have been described that relate to automatic generation of memoryefficient clustered data structures and, more specifically, to automatic selection of an initial data entity of interest, adding of the initial data entity to the memory-efficient clustered data structure, and determining and adding one or more related data entities to the cluster. As described above, in various embodiments, a generated cluster may include far fewer data entities as compared to a huge collection of data items that may or may not be related to one another. This may be because, for example, data entities included in a cluster may only include those data entities that are related to one another and which may be relevant to a particular investigation. Accordingly, in various embodiments, processing of generated clusters may be highly efficient because, for example, a given fraud investigation by an analyst may only require storage in memory of a single cluster data structure. Further, a number of data entities in a cluster may be several orders of magnitude smaller than in the huge collection of data items that may or may not be related to one another because only data entities related to each other are included in the cluster.

While the foregoing is directed to embodiments of the present invention, other and further embodiments of the invention may be devised without departing from the basic scope thereof. For example, aspects of the present invention may be implemented in hardware or software or in a combination of hardware and software. One embodiment of the invention may be implemented as a program product for use with a computer system. The program(s) of the program product define functions of the embodiments (including the methods described herein) and can be contained on a variety of computer-readable storage media. Illustrative computer-readable storage media include, but are not limited to: (i) non-writable storage media (e.g., read-only memory devices within a computer such as CD-ROM disks readable by a CD-ROM drive, flash memory, ROM chips or any type of solid-state non-volatile semiconductor memory) on which information is permanently stored; and (ii) writable storage media (e.g., hard-disk drive or any type of solid-state random-access semiconductor memory) on which alterable information is stored.

The invention has been described above with reference to specific embodiments. Persons of ordinary skill in the art, however, will understand that various modifications and changes may be made thereto without departing from the broader scope of the invention as set forth in the appended claims.

The foregoing description and drawings are, accordingly, to be regarded in an illustrative rather than a restrictive sense. Therefore, the scope of the present invention is determined by the claims that follow.

Various example embodiments of the disclosure can be described with respect to the following clauses:

Clause l.A computer-implemented method for generating a cluster of related data, the method comprising:

retrieving, by a cluster engine executing on a processor, a seed;

adding the seed to a first cluster; retrieving, by a cluster engine executing on a processor, a cluster strategy referencing one or more data bindings, wherein each data binding specifies a search protocol for retrieving data; and for each of the one or more data bindings: identifying data parameters input to the search protocol, performing the search protocol using the identified data parameters, and growing the cluster using data returned by the search protocol.

identifying data parameters input to the search protocol, performing the search protocol using the identified data parameters, and growing the cluster using data returned by the search protocol.

Clause 2. The method of clause 1, wherein growing the cluster using data returned by the search protocol comprises evaluating the data returned by the search protocol for inclusion in the first cluster.

Clause 3. The method of clause 1, wherein the search protocol for a first one of the data bindings uses the seed as the data parameters input to the search protocol.

Clause 4. The method of clause 1, wherein the search protocol for a first one of the data bindings searches for data in a first database and the search protocol for a second one of the data bindings searches for data in a second database .

Clause 5. The method of clause 1, wherein the search protocol for a first one of the data bindings returns data used as the identified data parameters input to the search protocol for a second one of the data bindings.

Clause 6. The method of clause 1, further comprising: retrieving a second cluster from a plurality of clusters;

comparing data from the first cluster to data from the second cluster; and determining whether to merge the first cluster and the second cluster based on the comparison .

Clause 7. The method of clause 1, further comprising, assigning a ranking score to the first cluster, wherein the ranking score is used to order the first cluster, relative to a plurality of other clusters generated from a respective seed according to the search strategy.

Clause 8. A computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform an operation for generating a cluster of related data, the operation comprising:

retrieving a seed;

adding the seed to a first cluster;

retrieving a cluster strategy referencing one or more data bindings, wherein each data binding specifies a search protocol for retrieving data; and for each of the one or more data bindings:

identifying data parameters input to the search protocol, performing the search protocol using the identified data parameters, and growing the cluster using data returned by the search protocol.

Clause 9. The computer-readable storage medium of clause 8, wherein growing the cluster using data returned by the search protocol comprises evaluating the data returned by the search otocol for inclusion in the first cluster.

Clause 10. The computer-readable storage medium of clause 8, wherein the search protocol for a first one of the data bindings uses the seed as the data parameters input to the search protocol.

Clause 11. The computer-readable storage medium of clause 8, wherein the search protocol for a first one of the data bindings searches for data in a first database and the search protocol for a second one of the data bindings searches for data in a second database.

Clause 12. The computer-readable storage medium of clause 8, wherein the search protocol for a first one of the data bindings returns data used as the identified data parameters input to the search protocol for a second one of the data bindings.

Clause 13. The computer-readable storage medium of clause 8, wherein the operation further comprises:

retrieving a second cluster from a plurality of clusters;

comparing data from the first cluster to data from the second cluster; and determining whether to merge the first cluster and the second cluster based on the comparison.

Clause 14. The computer-readable storage medium of clause 10, further comprising, assigning a ranking score to the first cluster, wherein the ranking score is used to order the first cluster, relative to a plurality of other clusters generated from a respective seed according to the search strategy.

Clause 15. A system, comprising:

a processor; and a memory storing one or more programs, which, when executed on the processor perform an operation for generating a cluster of related data, the operation comprising:

retrieving a seed, adding the seed to a first cluster, retrieving a cluster strategy referencing one or more data bindings, wherein each data binding specifies a search protocol for retrieving data, and for each of the one or more data bindings:

identifying data parameters input to the search protocol;

performing the search protocol using the identified data parameters; and growing the cluster using data returned by the search protocol.

Clause 16. The system of clause 15, wherein growing the cluster using data returned by the search protocol comprises evaluating the data returned by the search protocol for inclusion in the first cluster.

Clause 17. The system of clause 15, wherein the search protocol for a first one of the data bindings uses the seed as the data parameters input to the search protocol.

Clause 18. The system of clause 15, wherein the search protocol for a first one of the data bindings searches for data in a first database and the search protocol for a second one of the data bindings searches for data in a second database .

Clause 19. The system of clause 15, wherein the search protocol for a first one of the data bindings returns data used as the identified data parameters input to the search protocol for a second one of the data bindings.

Clause 20. The system of clause 19, wherein the operation further comprises:

retrieving a second cluster from a plurality of clusters;

comparing data from the first cluster to data from the second cluster; and determining whether to merge the first cluster and the second cluster based on the comparison.

Clause 21. The system of clause 19, further comprising, assigning a ranking score to the first cluster, wherein the ranking score is used to order the first cluster, relative to a plurality of other clusters generated from a respective seed according to the search strategy.

Clause 22. A computer-implemented method for prioritizing a plurality of clusters, the method comprising:

identifying a scoring strategy for prioritizing the plurality of clusters, wherein each cluster is generated from a seed and stores a collection of data retrieved using the seed;

for each cluster:

evaluating, by operation of a cluster engine executing on a processor, elements of the collection of data stored by the cluster according to the scoring strategy, and assigning a score to the cluster based on the evaluation; and ranking the clusters according to the respective scores assigned to the plurality of clusters.

Clause 23. The method of clause 22, wherein the collection of data stored by each cluster includes financial data evaluated by the scoring strategy for a risk of fraud.

Clause 24. The method of clause 22, wherein the score assigned to each cluster corresponds to an amount at risk.

Clause 25. The method of clause 22, wherein assigning a respective score to the cluster based on the evaluation comprises :

determining a plurality of base scores; determining, as the score to assign to the cluster, an aggregate score for the cluster from the plurality of base scores .

Clause 26. The method of clause 22, further comprising, presenting a summary for one or more of the plurality of clusters according to the ranking.

Clause 27. The method of clause 27, wherein the summary presents the seed associated with each of the clusters .

Clause 28. The method of clause 22, further comprising:

identifying a second scoring strategy for prioritizing the plurality of clusters;

for each cluster:

evaluating one or more attributes of the collection of data stored by the cluster according to the second scoring strategy, and assigning a second score to the cluster; re-ranking the clusters according to the respective second ranking scores assigned to the plurality of clusters; and presenting a second summary of the plurality of clusters according to the re-ranking.

Clause 29. A computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform an operation for prioritizing a plurality of clusters, the operation comprising:

identifying a scoring strategy for prioritizing the plurality of clusters, wherein each cluster is generated from a seed and stores a collection of data retrieved using the seed;

for each cluster:

evaluating elements of the collection of data stored by the cluster according to the scoring strategy, and assigning a score to the cluster based on the evaluation; and ranking the clusters according to the respective scores assigned to the plurality of clusters.

Clause 30. The computer-readable storage medium of clause 29, wherein the collection of data stored by each cluster includes financial data evaluated by the scoring strategy for a risk of fraud.

Clause 31. The computer-readable storage medium of clause 29, wherein the score assigned to each cluster corresponds to an amount at risk.

Clause 32. The computer-readable storage medium of clause 29, wherein assigning a respective score to the cluster based on the evaluation comprises:

determining a plurality of base scores; determining, as the score to assign to the cluster, an aggregate score for the cluster from the plurality of base scores.

Clause 33. The computer-readable storage medium of clause 29, wherein the operation further comprises, presenting a summary for one or more of the plurality of clusters according to the ranking.

Clause 34. The computer-readable storage medium of clause 33, wherein the summary presents the seed associated with each of the clusters.

Clause 35. The computer-readable storage medium of clause 29, wherein the operation further comprises:

identifying a second scoring strategy for prioritizing the plurality of clusters;

for each cluster:

Clause 36. A computer system, comprising:

a processor; and a memory storing one or more programs, which, when executed on the processor, perform an operation for prioritizing a plurality of clusters, the operation comprising :

identifying a scoring strategy for prioritizing the plurality of clusters, wherein each cluster is generated from a seed and stores a collection of data retrieved using the seed, for each cluster:

evaluating elements of the collection of data stored by the cluster according to the scoring strategy; and assigning a score to the cluster based on the evaluation, and ranking the clusters according to the respective scores assigned to the plurality of clusters.

Clause 37. The system of clause 36, wherein the collection of data stored by each cluster includes financial data evaluated by the scoring strategy for a risk of fraud.

Clause 38. The system of clause 36, wherein the score assigned to each cluster corresponds to an amount at risk.

Clause 39. The system of clause 36, wherein assigning a respective score to the cluster based on the evaluation comprises :

determining a plurality of base scores; determining, as the score to assign to the cluster, an aggregate score for the cluster from the plurality of base scores .

Clause 40. The system of clause 36, further comprising, presenting a summary for one or more of the plurality of clusters according to the ranking.

<td> Clause</td><td> 41.</td><td> The</td><td> system of</td><td> clause</td><td> 41, wherein</td><td> the</td>

<td> summary presents</td><td> the</td><td colspan="3"> seed associated with</td><td> each of the</td><td></td>

<td> clusters .</td><td></td><td></td><td></td><td></td><td></td><td></td>

<td> Clause</td><td> 42.</td><td> The</td><td> system of</td><td> clause</td><td> 36, wherein</td><td> the</td>

operation further comprises:

identifying a second scoring strategy for prioritizing the plurality of clusters; for each cluster:

evaluating one or more attributes of the collection of data stored by the cluster according to the second scoring strategy, and assigning a second score to the cluster;

re-ranking the clusters according to the respective second ranking scores assigned to the plurality of clusters; and presenting a second summary of the plurality of 5 clusters according to the re-ranking.

Citations
Document	Relation	Office	Cited during
US10721268B2	Cited by	United States of America	Applicant
US10834123B2	Cited by	United States of America	Applicant

59 members in 10 offices

This recordMember, by publication datePriority claim

Priority claims12

Priority claims
Document	Office	Kind	Date
201361800887	United States of America	P	2013-03-15
201361800887	United States of America	P	2013-03-15
201313968213	United States of America	A	2013-08-15
201313968213	United States of America	A	2013-08-15
201313968265	United States of America	A	2013-08-15
201313968265	United States of America	A	2013-08-15
13968213	–	–	–
13968265	–	–	–
61800887	–	–	–
US201313968213	–	–	–
US201313968265	–	–	–
US201361800887P	–	–	–

Members59

Family members, oldest first
Document	Office	Kind
GB201404457D0	United Kingdom	D0	2014-04-30
US8788405B1	United States of America	B1	2014-07-22
US8788407B1	United States of America	B1	2014-07-22
US8818892B1	United States of America	B1	2014-08-26
CA2845995A1	Canada	A1	2014-09-15
NL2012433A	Netherlands (Kingdom of the)	A	2014-09-16
EP2778983A1	European Patent Office (EPO)	A1	2014-09-17
DE102014204842A1	Germany	A1	2014-09-18
AU2014201505A1	Australia	A1	2014-10-02
US2014310282A1	United States of America	A1	2014-10-16
GB2513247A	United Kingdom	A	2014-10-22
US9135658B2	United States of America	B2	2015-09-15
US9165299B1	United States of America	B1	2015-10-20
US9171334B1	United States of America	B1	2015-10-27
US9177344B1	United States of America	B1	2015-11-03
US9202249B1	United States of America	B1	2015-12-01
US9230280B1	United States of America	B1	2016-01-05
EP2963577A1	European Patent Office (EPO)	A1	2016-01-06
US2016006749A1	United States of America	A1	2016-01-07
US2016034470A1	United States of America	A1	2016-02-04
US9344447B2	United States of America	B2	2016-05-17
NL2012433B1This record	Netherlands (Kingdom of the)	B1	2016-07-22
US2016344758A1	United States of America	A1	2016-11-24
US2016366164A1	United States of America	A1	2016-12-15
US9965937B2	United States of America	B2	2018-05-08
US9998485B2	United States of America	B2	2018-06-12
US2018270264A1	United States of America	A1	2018-09-20
US2019052648A1	United States of America	A1	2019-02-14
US10216801B2	United States of America	B2	2019-02-26
US10264014B2	United States of America	B2	2019-04-16
US10275778B1	United States of America	B1	2019-04-30
US2019158509A1	United States of America	A1	2019-05-23
US2019166135A1	United States of America	A1	2019-05-30
US2019205897A1	United States of America	A1	2019-07-04
EP2963577B1	European Patent Office (EPO)	B1	2019-11-27
EP2778983B1	European Patent Office (EPO)	B1	2019-12-25
EP2963577B8	European Patent Office (EPO)	B8	2020-01-01
DK2778983T3	Denmark	T3	2020-01-27
EP3627345A1	European Patent Office (EPO)	A1	2020-03-25
PL2778983T3	Poland	T3	2020-05-18
ES2766903T3	Spain	T3	2020-06-15
US10721268B2	United States of America	B2	2020-07-21
US2020304522A1	United States of America	A1	2020-09-24
US10798116B2	United States of America	B2	2020-10-06
EP3731166A1	European Patent Office (EPO)	A1	2020-10-28
US10834123B2	United States of America	B2	2020-11-10
EP3742694A1	European Patent Office (EPO)	A1	2020-11-25
US2020396237A1	United States of America	A1	2020-12-17
US10937034B2	United States of America	B2	2021-03-02
EP3627345B1	European Patent Office (EPO)	B1	2021-08-18
US11336681B2	United States of America	B2	2022-05-17
US2022239672A1	United States of America	A1	2022-07-28
US11546364B2	United States of America	B2	2023-01-03
US2023096596A1	United States of America	A1	2023-03-30
US11848760B2	United States of America	B2	2023-12-19
US11895137B2	United States of America	B2	2024-02-06
US2024146761A1	United States of America	A1	2024-05-02
EP3731166B1	European Patent Office (EPO)	B1	2024-08-28
US12238136B2	United States of America	B2	2025-02-25

Data clustering.

Abstract

Term

Priority claims12

Members59

Events

Numbers

Titles

Classification