Method and apparatus for an adaptive stereotypical profile for recommending items representing a user's interests

Method and device for recommending adaptive prototype profile of items representing user interest

Technical field

The present invention relates to a method and device for recommending interesting items such as television programs, and more particularly to a technology for recommending interesting programs and other items.

Background technique

With the increase in the number of channels that TV viewers can watch and the variety of programs available on these channels, it has gradually become a problem for TV viewers to distinguish interesting programs. The Electronic Program Guide (EPG) identifies available TV programs by, for example, title, time, date, and channel, and by allowing the available TV programs to be searched or classified according to personalized preferences, it facilitates the identification of programs of interest.

A variety of recommendation tools have been suggested or proposed for TV programs and other items that are of interest to recommendation. For example, a TV program recommendation tool applies the viewer's preferences to the EPG to obtain a set of recommended programs that are of interest to a particular viewer. In general, TV program recommendation tools use implicit or explicit techniques, or some combination of the above techniques to obtain viewer preferences. The implicit TV program recommendation tool generates TV program recommendations based on information obtained from the viewer's viewing history in an unobtrusive manner. On the other hand, the explicit TV program recommendation tool explicitly asks viewers for their preferences for program attributes, such as title, genre, actors, channel, and date/time, to obtain viewer profiles and generate recommendations.

Although currently available recommendation tools help users identify items of interest, they are subject to many restrictions. If these restrictions are overcome, the convenience and performance of the above recommendation tools can be greatly improved. For example, in order to be more comprehensive, the initialization of the explicit recommendation tool is very lengthy, requiring each new user to respond to a very detailed survey, specifying their preferences at a rough level of granularity. Although the implicit TV program recommendation tool obtains the profile unobtrusively by observing the viewing behavior, it takes a long time to make it accurate. In addition, this implicit TV program recommendation tool requires at least a minimum amount of viewing history to start making recommendations. Therefore, the recommendation cannot be made when the implicit TV program recommendation tool is first obtained.

Therefore, there is a need for methods and devices that can unobtrusively recommend items such as television programs before obtaining a fully personalized viewing history. In addition, there is a need for methods and devices that accurately capture the user's viewing behavior.

In general, a method and device for recommending items of interest to users, such as TV program recommendations, are disclosed. According to the principle of the present invention, the initial recommendation generated before the user's viewing history or purchase history is obtained is modified or converted to use feedback processing to better capture the user's viewing behavior.

Initially, for example, a stereotype is generated based on the viewing history of a specific viewing area, which is used to build a prototype profile. The generated prototype profile reflects the typical pattern of the project selected by the representative audience. The prototype profile is used to calculate recommendations relative to ground truth data. The centroid of each prototype in the prototype profile is used to calculate the spacing between each performance in the field real data. If there is an inconsistency between the calculated recommendation and the original real data on the ground, additional feedback is requested from the user, which is used to create a meta-profile. The later profile includes the set of all weights (eg, positive/negative supplements) that the user has provided to the show he/she wishes to be recommended or abandoned. Finally, the recommendation is recalculated using the later profile relative to the prototype profile.

Description of the drawings

By referring to the following detailed description and drawings, a more comprehensive understanding of the present invention, as well as further features and advantages of the present invention will be obtained.

Fig. 1 is a schematic block diagram of a TV program recommender according to the present invention; Fig. 2 is a flowchart describing the adaptive prototype profile processing in Fig. 1 that implements the principle of the present invention.

detailed description

Fig. 1 illustrates a TV program recommender 100 according to the present invention. As shown in FIG. 1, an exemplary TV program recommender 100 evaluates programs in a program database 200 to identify programs that are of interest to a particular viewer. For example, a set-top terminal/television (not shown) using well-known on-screen display technology can present a set of recommended programs to the viewer. Although the present invention is exemplified by TV program recommendations, the present invention can be applied to recommendations that are automatically generated based on the evaluation of user behaviors such as viewing history or purchase history. In particular, devices like set-top boxes, TiVo (hard disk recorders, PVR, etc.). The present invention can also be used in any application that uses user profile clustering. In the case of World Wide Web-Profile, the invention is embedded in a web browser.

Before the user's viewing history 140 is available, for example, when the user obtains the TV program recommender 100 for the first time, the TV program recommender 100 generates a TV program recommendation. As shown in FIG. 1, the TV program recommender 100 uses the viewing history 130 from one or more third parties to recommend programs of interest to a specific user. Generally, the third-party viewing history 130 is based on the viewing habits of one or more sampled populations, and demographic statistics, for example, represent the age, income, gender, and education of the majority of the population.

As shown in FIG. 1, the third-party viewing history 130 includes a group of programs that are watched or not watched by a specific group of people. Obtain the group of programs watched by observing the programs actually watched by a specific group of people. For example, the unwatched group of programs is obtained by randomly sampling programs in the program database 200. In a further variation, according to the teachings of U.S. Patent Application Serial No. 09/819,286 filed on March 28, 2001, entitled "An Adaptive Sampling Technique for Selecting Negative Examples for Artificial Intelligence Applications" Group shows. This application has been assigned to the assignee of the present invention and is hereby incorporated by reference.

The TV program recommender 100 processes the third-party viewing history 130 to generate a prototype profile that reflects the typical pattern of TV programs watched by representative viewers. A prototype profile is a cluster of TV programs (data points) that are similar to each other in some respects. The prototype profile can be generated in any of a variety of ways. For example, for example, the US patent application serial number NO.xx/xxx,xxx filed on November 14, 2001 and titled "Method and Apparatus for Generating a Stereotypical Profile for Recommending Items of Interest Using Item-Based Clustering", and November 2001 It is described in the US patent application serial number NO.xx/xxx,xxx entitled "Method and Apparatus for Generating aStereotypical Profile for Recommending Items of InterestUsing Feature-Based Clustering" filed on March 13. Each application is hereby incorporated by reference.

The TV program recommender 100 may be implemented as any computing device, such as a personal computer or a workstation, and it includes a processor 115 such as a central processing unit (CPU), and a memory 120 such as RAM and/or ROM. For example, the TV program recommender 100 may also be implemented as an application specific integrated circuit (ASIC) in a set-top terminal or a display (not shown). In addition, the TV program recommender 100 can be implemented as any available TV program recommender, such as the commercial TivoTM system available from Tivo, Inc. of Sunyvale, California, or the "Method and Apparatus for Recommending Television" filed on December 17, 1999. The TV program recommender described in the "Programming Using Decision Trees" US Patent Application No. 09/466,406, and the US Patent Application No. 09/498,271 filed on February 4, 2000, entitled "Bayesian TV Show Recommender" The TV program recommender described in and the "Three-Way Media Recommendation Method and The TV program recommender described in US Patent Application No. 09/627,139 of "System", or any combination thereof, each application is incorporated herein by reference.

The TV program recommender 100 includes a program database 200 and server routines in the memory 120, such as a prototype profile processing 300, and (not shown) a cluster routine, an average calculation routine, a spacing calculation routine, and a cluster Performance estimation routine. Generally, the program database 200 may be implemented as a known electronic program guide and record the information of each program available in a given time interval. The adaptive prototype profile processing 300 (i) processes the third-party viewing history 130 to generate a prototype profile, which reflects the typical pattern of television programs watched by a representative audience; (ii) uses the selected prototype to generate relative For the recommendation of the so-called real situation on the ground, the centroid of each prototype in the prototype profile is used to calculate the distance between each performance in the real data on the ground (the real data on the ground is a set of performances for which the user has given specific information, The specific information is, for example, how much he/she likes the performance. For example, the user can indicate that he/she loves the performance "Seinfeld", and the love can be converted to between 0.85 and 1.0 or converted to other suitable score conversion methods); iii) If there is an inconsistency between the calculated recommendation and the original real-world data (for example, if the user indicates that he/she loves "Seinfeld", the score should be between 0.85 and 1.0, so we know that the score is less than when calculating the recommendation 0.85, there is an inconsistency). The user feedback 160 is then used to convert the recommendation from the additional feedback requested by the user; (iv) the user feedback is then used to create a meta-profile that includes the users wishes to him/her to be recommended or abandoned The set of all weights (such as positive/negative supplements) provided by the show. (v) The recommendation is recalculated using a later profile relative to the prototype profile.

In particular, in an exemplary embodiment, the cluster routine may be invoked by the adaptive prototype profile processing 300 to divide the third-party viewing history 130 (data collection) into clusters so that those points in a cluster ( TV show) is closer to the mean (centroid) of that cluster than any other cluster. The cluster routine calls the mean calculation routine to calculate the symbolic mean of the cluster. The interval calculation routine is called by the cluster routine to estimate the proximity of the TV program to each cluster based on the distance between the specific TV program and the mean value of the specific cluster. Then the cluster routine calls the cluster performance estimation routine to determine when the stopping criteria for generating the cluster is met, such as the submission on November 13, 2001 entitled "Method and Apparatus for Generating a stereotypical profile for recommending items of interest" It is further described in US Patent Application No. 10/014,189 of "using feature-based clustering", which is incorporated herein by reference.

Figure 2 is a flow chart describing an exemplary execution of the adaptive prototype profile processing 300 with features of the present invention. As previously indicated, the adaptive prototype profile processing 300 processes the third-party viewing history 130 in step 310 to generate a prototype profile based on a prototype reflecting a typical pattern of television programs watched by a representative viewer. In step 320, the selected prototype is used to generate a recommendation relative to the actual data on the ground. The recommendation is calculated by using the following equation to calculate the spacing between each performance in the field real data using the centroid of each prototype in the prototype profile.

D(S1-S2)=Σ1Nδ(S1i-S2i)]]> where S1 and S2 correspond to two performances, and N corresponds to the number of features that constitute the performance record. Please note that the distance D is normalized to be between 0 and 1.

Thereafter, in steps 330-350, the calculated recommendation is compared with the original on-site real data, and if there is an inconsistency between them, the user is prompted for additional feedback on the recommendation. Can get feedback from users by any conventional method. This feedback is then used to form the weighting factor. As an example, if the user indicates that he likes all Clint Eastwood movies, then there is Clint All ratings for Eastwood performances have been increased, and vice versa. In addition, the weighting factor is used at the program level and the feature level. For example, at the entire performance level or the features such as actors and genres that make up the performance. In step 360, the feedback is used to create a subsequent profile that includes a set of all weights (such as positive/negative supplements) that the user has provided for performances that he/she wishes to be recommended or abandoned. Finally, in step 370, the recommendation is recalculated by using the later profile relative to the prototype profile: D(S1-S2)=(1-W)Σ1Nδ(S1i-S2i)]]> It should be noted that Because the performance in the profile itself is centroid, the weight of the prototype profile is often set to 1. Intuitively, when the user gives feedback, he/she hopes that the evaluation of the performance will be closer to or farther away from the centroid. It should be noted that the measures given above give a distance. Ideally, when the performance has zero spacing, this means that the performance is closer to the centroid. To get a score; subtract from 1. As an example, if the user has given the following feedback for a particular performance-don't care, like, and love, this corresponds to 0, 0.7, and 1, respectively. In addition, let us assume that the actual calculated distance between the show and the prototype profile is 0.2. The following table shows the values calculated using the equation shown above.

Weighted Spacing Probability...

0 0.2 0.80.7 0.06 0.941 0 1.0 It should be noted that special boundary conditions need to be established when the user does not like it at all. For example, if the user says (-1), this performance is not recommended at all. In the case where the spacing exceeds 1, the spacing should be renormalized so that the score can be calculated.

It should be understood that the embodiments and modifications shown and described herein are only examples of the principle of the present invention, and various modifications can be made by those skilled in the art without departing from the spirit and scope of the present invention.