KR20120075515A - Personalized content recommendation system and method - Google Patents

Abstract

PURPOSE: A user preference content recommendation system and method thereof are provided to recommend preferred content to a user based on evaluation by evaluating the performance of content recommendation models. CONSTITUTION: A content recommendation model creation unit(2510) creates plural content recommendation models by using data for recommending content. A recommendation model performance evaluation unit(2540) evaluates content recommendation models of a content recommendation performance by using view log information for plural tests. A content recommendation unit(2550) selects a model from the created recommendation models based on the evaluated recommendation performance.

Description

User preferred content recommendation system and method {PERSONALIZED CONTENT RECOMMENDATION SYSTEM AND METHOD}

The present invention relates to a user preferred content recommendation system and method, and more particularly, a user who evaluates the performance of a plurality of content recommendation models using various types of test viewing logs and recommends preferred content to the user based on the same. A preferred content recommendation system and method.

The rapid development of video coding technology and internet and data communication has reached the level of real-time broadcasting of high-definition multimedia video through an IP (Internet Protocol) -based network. With the proliferation of the Internet, the demand for two-way communication, the demand for high-definition video, and the convergence of broadcasting and communication, IPTV was born as the related technologies reached maturity. In IPTV, high-definition multimedia video can be watched in real time through the IP network, and unlike the existing broadcast, the IP network uses the IP network, thereby enabling bidirectional communication. Therefore, the IPTV user can select and watch a video he / she wants at a desired time, instead of simply watching a video broadcasted by a broadcaster according to time.

However, since there are so many types of content to select from IPTV, it is difficult for many users to select a desired content from among numerous contents. In order to solve this problem, conventionally, the category to which the VOD belongs is classified into a major classification, a middle classification, a small classification, and the like, and a step-by-step search method is provided. However, there is a problem in that it is possible to provide a user with easy to understand the bundle, but the user cannot provide a list of recommended content.

In addition, a technology for analyzing a user's viewing propensity based on the viewing history of the user and recommending the user's most preferred content has appeared. However, even with such a technology, various recommendation items are diversified. It was difficult to accurately predict the user's preferred contents by verifying the recommended performance.

On the other hand, as the prior art, Korean Patent Publication No. 10-2009-0046353 discloses the invention "personalized multi-recommended learning method" is disclosed, using the personalized multi-recommended method learning to improve the recommendation performance As a method, a method of applying a recommendation method using similar preferences of similar users and a content-based recommendation method using user's personal history information and attribute information of digital content at the same time for each learning period and user, and then applying a method of high recommendation accuracy. Relates to a technology related to.

However, the above-mentioned prior art cannot evaluate the performance of recommendation for a plurality of recommendation criteria using a plurality of types of test viewing log information, and thus there is a problem in that it provides a limit for providing accurate recommendations to users.

Therefore, there is a strong demand for a content recommendation technology capable of verifying recommendation performance in various ways based on various recommendation criteria and providing accurate preferred content to a user.

Some embodiments of the present invention provide a user preferred content recommendation system and method that can provide recommended content to a user using a plurality of content recommendation models.

In addition, an embodiment of the present invention can provide a user preferred content recommendation system and method that can evaluate the performance of each content recommendation model in a variety of ways using a plurality of test viewing log information.

In addition, an embodiment of the present invention provides a user preferred content recommendation system and method for generating a content recommendation list by selecting a content recommendation model optimized for a specific user based on a viewing log of a specific user.

As a technical means for achieving the above-mentioned technical problem, the 1st aspect of this invention is a content recommendation model generation part which produces | generates a plurality of content recommendation models using the data which are based on content recommendation, and a plurality of types for testing. A recommendation model performance evaluation unit for evaluating recommendation performance of the generated plurality of content recommendation models, and selecting one of the generated recommendation models based on the evaluated recommendation performances using the viewing log, and selecting the selected recommendation model. The system may provide a content recommendation system for recommending preferred content to a user in an interactive broadcasting service including a content recommending unit recommending a user's preferred content to a user terminal.

In addition, the second aspect of the present invention, generating a plurality of content recommendation model using at least one of the metadata information, user profile information and the viewing log information of the user, the viewing viewing for the test of the viewing log information Extracting a log, generating a plurality of types of test viewing log information using the extracted test viewing log, and using the plurality of types of test viewing log information, respectively, the plurality of content recommendation models A method of generating a content recommendation model used to generate a user recommendation content list in an interactive broadcasting service may include evaluating recommendation performance of.

In addition, the third aspect of the present invention, receiving the viewing log information of the user viewing the content, based on the viewing log information of the user, one of a plurality of content recommendation model used for generating a user content recommendation list Selecting a list, generating a list of recommended content based on the selected content recommendation model, and providing the generated list of recommended content to the user terminal of the user, wherein the plurality of content recommendation models include content. In the recommendation server, the recommendation performance is evaluated by the plurality of test viewing log information, and in the selecting of one of the plurality of content recommendation models, selecting one of the plurality of content recommendation models based on the evaluated recommendation performance. In a two-way broadcast service to provide recommended content to users. Can be.

According to the problem solving means of the present invention described above, it is possible to effectively provide the recommended content to the user by using a plurality of content recommendation model.

In addition, according to the problem solving means of the present invention described above, it is possible to evaluate the performance of each content recommendation model in various ways using a plurality of test viewing log information.

In addition, according to one of the other problem solving means of the present invention, it is possible to generate a content recommendation list by selecting a content recommendation model optimized for the user based on the viewing log of the user requesting the content recommendation.

1 is a general schematic diagram of a general IPTV broadcasting system to which a user-customized content recommendation system according to an embodiment of the present invention is applied.
2 is a diagram illustrating a detailed configuration for providing a preferred content to a user in the IPTV service system 2000 according to an embodiment of the present invention.
3 illustrates an IPTV head-end server 2300, a data preprocessing server 2400, a content recommendation server 2500, and a web application providing server 2600 according to an embodiment of the present invention. FIG. 4 is a diagram illustrating a process of providing preferred content to the user terminal 4000.
4 is a detailed block diagram of a content recommendation server 2500 according to an exemplary embodiment of the present invention.
5 is a detailed flowchart of a method for generating and evaluating a content recommendation model by a content recommendation server 2500 according to an exemplary embodiment.
6 is a detailed flowchart of a method for recommending content by the content recommendation server 2500 according to an exemplary embodiment.

DETAILED DESCRIPTION Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art may easily implement the present invention. The present invention may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. In the drawings, parts irrelevant to the description are omitted in order to clearly describe the present invention, and like reference numerals designate like parts throughout the specification.

Throughout the specification, when a part is referred to as being "connected" to another part, it includes not only "directly connected" but also "electrically connected" with another part in between . In addition, when a part is said to "include" a certain component, which means that it may further include other components, except to exclude other components unless otherwise stated.

In addition, in the present specification, the content recommendation model refers to a model including information on how to recommend content to a user and used as a criterion for content recommendation, and includes a metadata-based content recommendation model, a profile-based content recommendation model, and a viewing log. Based on the content recommendation model can be generated.

Hereinafter, with reference to the accompanying drawings will be described in detail the present invention.

1 is a general schematic diagram of a general IPTV broadcasting system to which a user-customized content recommendation system according to an embodiment of the present invention is applied.

As shown in FIG. 1, the IPTV broadcasting system includes a broadcaster 1000, an IPTV service system 2000, a network network 3000, and a user terminal 4000.

In addition, the IPTV service system 2000 includes a baseband system 2010, a compression multiplexing system 2020, a conditional access system (CAS) 2030, a back office system 2040, a monitoring system 2050, A media management core (MOC) 2060, a subscriber management system 2070, a data broadcasting system 2080, an electronic program guide (EPG) system 2090, and a return path server system 2100.

The broadcaster 1000 serves to produce, edit, and change broadcast content and provide it to the IPTV service system 2000. The broadcaster 1000 may include a program provider (PP), a terrestrial wave, or a content provider (CP). In addition, the broadcast content provided by the broadcaster 1000 may include existing broadcast content and rich content on the Internet.

The IPTV service system 2000 receives and manages broadcast content from the broadcaster 1000 and distributes content to the user terminal 4000 to provide broadcast / recording / playback services. The received broadcast content may include related additional information and an EPG for this additional information.

The IPTV service system 2000 includes EPG information (PSIP / PSI /) including a subscriber concentrator, a broadcast video and audio signal, data broadcasting data, and a program recommendation service menu through a router supporting a multicast routing protocol. SI information) can be transmitted to IP set-top boxes of multiple subscribers by multicasting.

The baseband system 2010 receives an MPEG2 broadcast signal from an external program provider (PP) or an analog broadcast signal from terrestrial waves, converts the received source broadcast signal into a SDI (Serial Digital Interface) signal, and converts a frame ( Frames), and distributes video and audio signals from multiple broadcast channels (e.g. 100 channels) via a routine switcher, subtitle generator (CG) and automatic program controller (APC). At least one of an advertisement, a logo, or a subtitle is inserted into the broadcast video and audio signal (signal editing and processing) and transmitted to the compression multiplexing system 2020.

The compression multiplexing system 2020 uses an A / V encoder (A / V Encoder) for broadcasting video and audio signals (Video, Audio) received from the baseband system 2010 for each broadcasting channel (eg, 100 channels). ) To compress the SDI (Serial Digital Interface) video signal to H.264, and to compress the audio signal to MPEG-2 AAC to generate MPEG-2 TS (Transport Stream), and to compress compressed broadcast video and audio -2 after multiplexing data broadcasting data and EPG information (PSIP / PSI / SI information) generated by a data encoder and a PSI / SI generator together with the TS signal When the received MPEG-2 TS signal is used in the reception restriction technique, the TS-2 may be inputted to a scrambler, encrypted, and finally IP packetized to transmit an IP packetized TS (Transport Stream) broadcast signal.

Also, optionally, the reception restriction system 2030 may be used to prevent illegal viewing and illegal copying of program recommendation content.

The reception restriction system 2030 encrypts the real-time channel and performs pre-encryption of the VOD contents, and controls the viewing authority so that the channel and the content can be used only for the authenticated user. In order to prevent illegal copying of IPTV content, a digital rights management (DRM) scheme may be used instead of the reception restriction system 2030.

The back office system 2040 is a provisioning system and provides a billing processing function for the use of the IPTV program service for each subscriber.

The monitoring system 2050 is a control system. The monitoring system 2050 may monitor a transmission failure of an A / V broadcast signal for IPTV broadcasting, a reception failure by monitoring a downlink of an IPTV headend system, and confirm subtitles.

The media management system 2060 is a system that manages various business process information (program organization information, location information, contract information, product information, etc.) for operating a broadcasting business. The media management system 2060 integrally manages the information flow through the organic combination with each system at the center of the broadcasting center. The media management system 2060 manages broadcast program organization information, content and media management information, contract information of a program provider (PP) and a content provider (CP), and product information, and manages each system in the center of a broadcasting center. Through organic integration, it can act as a coordinator to manage and manage the flow of information.

The subscriber management system 2070 provides membership registration and termination and member information management for IPTV services.

The data broadcasting system 2080 authors, verifies, organizes, and transmits the data broadcasting data. The data broadcasting system 2080 includes a data agent manager (PSI) for managing data encoding, a PSI / SI generator (PSI) for generating program specific information / service information (SI). / SI Generator), Data Server / Data Encoder for encoding data broadcasting data into broadcast video and audio signals, and Multiplexer Manager for managing multiplexing functions of data into the broadcast video and audio signals Manager) and a scheduler user interface (Scheduler UI).

The EPG system 2090 includes an EPG server and provides an electronic program guide (EPG) service to the user terminal 4000.

The return path server system 2100 processes bidirectional data by a data provider (DP), provides online bill transfer from the user terminal 4000, usage history / billing interworking processing of bidirectional data, and personalization authentication processing. , And receive bidirectional data for the program recommendation service and transmit the corresponding response data to the user terminal 4000 in unicasting.

The network 3000 receives the broadcast content from the IPTV service system 2000 and delivers the received broadcast content to the user terminal 4000. The network network 3000 includes a backbone network and an access network, and the access network includes Ethernet, xDSL (ADSL, VDSL), Hybrid Fiber Coaxial Ca (HFC), and Fiber To The Curb), and may be configured in any one topology of a fiber to the home (FTTH) structure.

The user terminal 4000 is an apparatus for using an Internet broadcasting service, and generally includes an IPTV, a set top box (STB), and a remote controller. The IPTV outputs the broadcast content received from the IPTV service system 2000 and transmits the user response signal received through the remote controller to the IPTV service system 2000 through the return channel of the set-top box. The user terminal may use any one of a TV with an IP STB or an IP set-top box (IP STB), a computer, a notebook, or a personal portable terminal connected to the user's TV.

Hereinafter, a detailed configuration for providing preferred content to a user in the IPTV service system 2000 according to an embodiment of the present invention will be described with reference to FIG. 2.

2 is a diagram illustrating a detailed configuration for providing a preferred content to a user in the IPTV service system 2000 according to an embodiment of the present invention.

As shown in FIG. 2, the IPTV service system 2000 according to an exemplary embodiment of the present invention includes an IPTV headend server 2300, a data preprocessing server 2400, a content recommendation server 2500, and a web application providing server ( 2600).

The IPTV headend server 2300 provides the data preprocessing server 2400 with data based on content recommendation. The IPTV headend server 2300 may provide the data preprocessing server 2400 with metadata information about the content, user profile information, and viewing log information about what content the IPTV viewer watched.

Here, the metadata information about the content is various detailed information related to the content, for example, the title of the content, copyright information, language used, genre, detailed description, actor, director, producer, country of origin, year of production, city hall Includes information about ratings, reviews, previews, related websites, related keywords, awards, ratings, showtimes, prices and more. In addition, the user profile information is personal information of the viewer who views the content, and includes, for example, information about the user's age, gender, address, family composition, service subscription date, and the like. In addition, the viewing log information includes information about the viewing history of the content of viewers viewing the content.

The data preprocessing server 2400 preprocesses the data based on the content recommendation. The data preprocessing server 2400 integrates content metadata information, user profile information, and viewing log information distributed in various detailed systems of the IPTV service system 2000, removes errors / mismatches of data, or removes missing values. Can be processed.

The content metadata information, the user profile information, and the viewing log information may be distributed and managed in various detailed systems included in the IPTV service system 2000. For example, some systems may have only metadata of content that is currently viewable by the user. And some systems may have metadata of content that has been provided in the past. In this case, the data preprocessing server 2400 may integrate the content metadata information, the user profile information, and the viewing log information distributed in various detailed systems of the IPTV service system 2000 into one, and extract necessary information therefrom. Can be.

The content metadata information, the user profile information, and the viewing log information may include various types of detailed information, and the data preprocessing server 2400 may remove errors or inconsistencies of the data through a certain processing process before the data is used. By compensating for missing values in the data, preprocessed data can be efficiently used for content recommendation.

The content recommendation server 2500 recommends content to the user based on the preprocessed data. The content recommendation server 2500 receives preprocessed data for content recommendation from the data preprocessing server 2400, generates a content recommendation model based on the received data, and provides the content to the user based on the generated content recommendation model. I can recommend it. In addition, the content recommendation server 2500 may generate various kinds of content recommendation models based on the preprocessed data, and evaluate the content recommendation model using various kinds of test viewing log information. A detailed configuration of the content recommendation server 2500 will be described later with reference to FIG. 4.

In response to the content recommendation request from the user terminal 4000, the web application providing server 2600 requests the content recommendation server 2500 for content recommendation, and receives recommended content information from the content recommendation server 2500. Give 4000.

When the web application providing server 2600 receives a content recommendation request from the user terminal 4000, the web application providing server 2600 provides the content recommendation server 2500 with the ID of the IPTV user and the ID of the content currently being viewed, from the content recommendation server 2500. The user may receive a recommendation content list suitable for the corresponding user, and provide the user terminal 4000 with a recommendation content list by constructing a user interface.

Here, the web application providing server 2600 is a server containing an application running on the IPTV in general, and can be linked with a browser-based user terminal 4000 for immediately reflecting various changes frequently performed on the IPTV. have.

In addition, the web application providing server 2600 may check whether the content in the recommended content list received from the content recommendation server 2500 is currently playable on the user terminal 4000, and the user has already viewed the content in the recommended content list. After the deletion, metadata information such as a title, actor, plot, and poster for the remaining contents may be received from the IPTV service system 2000. In addition, the received metadata information may be generated in a similar HTML form and provided to the user terminal 4000.

Hereinafter, an IPTV headend server 2300, a data preprocessing server 2400, a content recommendation server 2500, and a web application providing server according to an embodiment of the present invention will be described with reference to FIG. 3. A process of providing preferred content to the user terminal 4000 through 2600 will be described.

3 illustrates an IPTV head-end server 2300, a data preprocessing server 2400, a content recommendation server 2500, and a web application providing server 2600 according to an embodiment of the present invention. FIG. 4 is a diagram illustrating a process of providing preferred content to the user terminal 4000.

As shown in FIG. 3, the IPTV headend server 2300 may provide the data preprocessing server 2400 with various information necessary for recommending content, for example, content metadata, user profile information, and viewing log information. The data preprocessing server 2400 may preprocess the received data and provide it to the content recommendation server 2500.

In addition, the content recommendation server 2500 may use a plurality of content recommendation models (metadata-based content recommendation models, profile-based content recommendation models, and viewing log-based content recommendation models) and a plurality of test viewing log information using preprocessed data. General performance test, metadata test, profile test), and the content recommendation performance of each content recommendation model can be evaluated using the plurality of test viewing log information. In addition, the content recommendation server 2500 receives a content recommendation request from the web application providing server 2600, and in response, selects a content recommendation model optimized for the user who requested the content recommendation, and generates a content recommendation list for the user. Create

The generated content recommendation list is provided to the user terminal 4000 through the web application providing server 2600, and the user terminal 4000 receives the recommended content from the IPTV headend server 2300 based on the received content recommendation list. Can be played back.

Hereinafter, a detailed configuration of the content recommendation server 2500 according to an exemplary embodiment of the present invention will be described with reference to FIG. 4.

4 is a detailed block diagram of a content recommendation server 2500 according to an exemplary embodiment of the present invention.

As shown in FIG. 4, the content recommendation server 2500 according to an embodiment of the present invention generates a metadata-based recommendation model generator 2510, a profile-based recommendation model generator 2520, and a viewing log-based recommendation model. The unit 2530 includes a recommendation model performance evaluation unit 2540 and a content recommendation unit 2550. The content recommendation server 2500 receives the preprocessed data from the data preprocessing server 2400, generates a metadata-based, profile-based, and viewing-log based user-specific content recommendation model, and provides a web application by integrating and using the content recommendation model. When there is a content recommendation request from the server 2600, a content recommendation list may be generated and delivered.

The metadata-based recommendation model generator 2510 generates a metadata-based content recommendation model by using the preprocessed content metadata and the preprocessed viewing log information.

The metadata-based content recommendation model may recommend content by using only content metadata and using content metadata and viewing log information. The method of using only the content metadata is a method of recommending other content having the same value using information such as genre, actor, director, and the like in the content metadata. For example, other VODs in the action genre are recommended to users watching the VOD of the action genre, or other movie VOD starred by Brad Pitt for the user watching the movie VOD starred by Brad Pitt.

In addition, the method of utilizing the content metadata and the viewing log information is a method of recommending the content by performing machine learning training of the viewing log using the genre, the actors, and the director of the content metadata as qualities. For example, you may want to recommend a VOD that is watched by other people who watch a lot of action genres to specific users who are watching the VOD of the action genre, or that Brad Pitt appears to a specific user who is watching a movie VOD starred by Brad Pitt. It's a way of recommending VOD that other people who watched a lot of movies prefer.

The advantage of this metadata-based content recommendation model is that it can provide effective recommendations to enthusiasts of specific genres, or enthusiasts of specific directors or actors. In addition, in case of new VOD, no one has watched the VOD yet, so even if there is no viewing history about the VOD, it can be recommended to users who like the genre, director, or actor according to the metadata information of the VOD. This is a feature of data-based content recommendation model.

The profile-based recommendation model generator 2520 generates a profile-based recommendation model using the preprocessed user profile and the preprocessed viewing log information.

The profile-based recommendation model generator 2520 generates a content recommendation model to recommend the content by performing machine learning training of the viewing log by using the user profile information as the qualities of the user's age and gender. For example, recommend a VOD, such as a sports broadcast, watched by many other 20s, to a specific user in their 20s, or recommend a VOD, such as a drama broadcast, watched by many other 40s, to a specific user in their 40s. To create a content recommendation model. The advantage of the profile-based content recommendation model is that it follows the general attributes of the group to which the user belongs, so that the user does not watch any content so that the user is not able to see the content or have a tendency to understand the propensity of the user. The profile group to which they belong can recommend their preferred content. This can solve the disadvantage that the content of recommendation is overly specialized by recommending only the program similar to the already watched program, which is the disadvantage of general content recommendation models.

The viewing log-based recommendation model generation unit 2530 generates a viewing log-based content recommendation model using preprocessed viewing log information. The viewing log-based recommendation model generator 2530 may generate a content recommendation model according to an item association analysis method and a user clustering method.

The item association analysis method analyzes the viewing log of IPTV users and recommends VODs that are more likely to be seen by a person viewing a particular VOD. For example, if 90% of people who watched the VOD called A in the viewing log information saw the VOD called B, the VOD called B is recommended to the person currently watching the VOD called A.

In addition, the user clustering method is a method of clustering users by using viewing logs of IPTV users, and recommending VODs not watched by a specific user among VODs watched by other users of a cluster belonging to a specific user. The difference from the item association analysis method is that when recommending, many people who have similar VOD viewing propensity through the VOD viewing history that the user has seen not only consider one VOD currently viewed by the user. It recommends the VODs you watched.

In addition, the metadata-based content recommendation model, the profile-based content recommendation model, and the viewing log-based content recommendation model may be generated with various settings by various methods and algorithms.

The recommendation model performance evaluator 2540 evaluates content recommendation performance of the content recommendation model generated by the content recommendation model generator 2510, 2520, and 2530 by using the plurality of test viewing log information.

The recommendation model performance evaluator 2540 integrates the content recommendation model and tests the performance by using the metadata-based content recommendation model, the profile-based content recommendation model, and the viewing log-based content recommendation model. Can be evaluated Here, the test viewing log information refers to viewing log information that is not used to generate the content recommendation model of the recommendation model generators 2510, 2520, and 2530, and is separate from the viewing log information used to generate the content recommendation model. It may be replaced from time to time for optimization of the generation unit (2510, 2520, 2530). The test viewing log information may include a user ID, a content ID, a viewing time, and the like.

The recommendation model performance evaluation unit 2540 may integrate three recommendation models to evaluate content recommendation performance of the recommendation model as follows.

The recommendation model performance evaluation unit 2540 may classify the test viewing log information into various cases, and evaluate the content recommendation performance of the content recommendation models in each case using the divided test viewing log information. In addition, the recommendation model performance evaluator 2540 may classify the test viewing log information into general performance test viewing log information, metadata performance test viewing log information, and profile performance test viewing log information. The viewing log information for general performance test is to use the viewing log information for test as it is, and the viewing log information for performance test for each metadata is generated by dividing the viewing viewing log information for test according to the value of the metadata. The viewing log information may be generated by dividing the test viewing log information according to the value of the profile.

The recommended model performance evaluator 2540 applies the three recommended models to the viewing log information for the general performance test using the entire viewing log information for the general test as the viewing log information for the general performance test, and applies the general performance of each model. You can get it.

In addition, the recommendation model performance evaluation unit 2540 may divide the viewing log information for testing according to the value of the metadata and extract the viewing log information for performance testing for each metadata, and to the viewing log information for performance testing for each metadata. Each of the three recommended models can be applied to obtain the recommended performance for each model's metadata.

In more detail, the recommendation model performance evaluator 2540 may obtain metadata information corresponding to each content ID by using the preprocessed content metadata, and a threshold value of a specific metadata value among test viewing log information is used using the preprocessed content metadata. By selecting the viewing viewer logs as described above, the viewing log information for the metadata performance test can be extracted. At this time, the threshold ratio can be set by the system operator. In addition, when there are M metadata and each metadata has N values, the viewing log information for metadata performance test may be divided into M × N items.

For example, certain metadata is called "genre", and "genre" metadata has 10 values such as "action", "melo", "SF", and the threshold ratio is set to "70%". In this case, the recommendation model performance evaluator 2540 searches metadata information in the test viewing log information, and selects only the viewing log information of users having a ratio of “70%” or more watching the VOD whose “genre” is “action”. It is possible to extract test-viewing log information for a test in which the genre “metadata is“ action. ”In this manner, the recommended model performance evaluation unit 2540 may use the“ genre ”metadata such as“ action ”,“ melo ”,“ SF ”, and the like. If you can extract 10 test viewing log information related to "M", like "Genre", "Actor", "Director", and so on, you have N metadata like "Genre" metadata for each metadata.Respectively, it is possible to extract the log information for viewing list. In addition, the recommendation model performance evaluator 2540 may obtain the recommendation performance of each model by applying the three recommendation models to the viewing log information for the performance test for each (MxN) metadata.

In addition, the recommendation model performance evaluator 2540 may divide the viewing log information for test according to the value of the profile and extract the viewing log information for performance test for each profile. We can apply each of them to get the recommended performance of each model.

Specifically, the recommendation model performance evaluator 2540 may obtain profile information corresponding to each user ID by using the preprocessed user profile, and use the user whose value of a specific profile is greater than or equal to a threshold ratio among the viewing log information for testing using the preprocessed user profile. You can extract the viewing log information for profile performance test by selecting the viewing log information of. At this time, the threshold ratio can be set by the system operator. In addition, when there are M profiles and each profile has N values, the viewing log information for profile performance test may be divided into M × N items.

For example, if a particular profile is called “age”, and the “age” profile has 7 values such as “10s”, “20s”, “30s”, the recommended model performance evaluation unit 2540 In the test viewing log information, the profile information of the user ID can be queried to select only the users of the "ten" "age" and extract the test viewing log information of the "ten" age profile. The recommended model performance evaluation unit 2540 may extract test viewing log information for seven values related to the “age” profile such as “10s”, “20s”, “30s”, and so on. Similarly, if there are M profiles such as "age", "gender", "family composition information", and so on, each of the N test viewing log information can be extracted for each profile similarly to the "age" profile. The recommended model performance evaluation unit 2540 may obtain the recommended performance of each model by applying the three recommended models to the viewing log information for the profile performance test for the MxN items thus made.

The content recommendation unit 2550 recommends content to the user based on the performance evaluated for each content recommendation model. The content recommendation unit 2550 selects a content recommendation model suitable for the user based on the general performance, the performance for each metadata, and the profile-specific performance evaluated for each content recommendation model, and uses the selected content recommendation model to provide content to the user. I can recommend it.

The content recommendation unit 2550 is most optimal for the user to use any of the content recommendation models based on various information (user information, user viewing log information, etc.) of the user and the performance of each content recommendation model. You can determine if you can see the recommended performance. In addition, when a content recommendation request is received from the web application providing server 2600, the content recommendation unit 2550 may generate a list of recommended content using the model showing the best performance and provide it to the web application providing server 2600. have.

For example, in the recommendation model performance evaluation unit 2540, when the “age” profile is “20s” and the “gender” profile is “male”, the performance of the profile-based recommendation model is applied when all three models are applied. Is rated as the highest, the content recommendation unit 2550 provides the user with a profile-based recommendation model when a content recommendation request is received from a user whose age is in their twenties and the gender in male. Create a content recommendation list to be done.

In addition, the content recommendation request from the web application providing server 2600 may include, for example, a user ID, a content ID currently being watched, and the content recommendation unit 2550 may use the user ID of the corresponding user. After importing the viewing log information, classify the viewing log information by the same criteria as when dividing the viewing viewing log information for test, and if there are several models selected, select the model having the highest performance among the models and select the model. You can create a content recommendation list using.

The content recommendation server 2500 may not necessarily be used only for recommending content, but may be applied to user-customized IPTV advertisements. In this case, similar to recommending other VODs by analyzing VOD viewing patterns of IPTV users, VOD viewing patterns, etc. of IPTV users may be analyzed to recommend suitable advertisement images.

Hereinafter, a method of generating and evaluating a content recommendation model by the content recommendation server 2500 according to an exemplary embodiment will be described with reference to FIG. 5.

5 is a detailed flowchart of a method for generating and evaluating a content recommendation model by a content recommendation server 2500 according to an exemplary embodiment.

Step S500 is a step of receiving preprocessed data from the data preprocessing server 2400. In operation S500, the content recommendation server 2500 may receive preprocessed content metadata information, preprocessed user profile information, and preprocessed viewing log information from the data preprocessing server 2400.

Step S502 is a step of generating a variety of content recommendation models. In step S502, the content recommendation server 2500 uses the preprocessed content metadata information, the preprocessed user profile information, and the preprocessed viewing log information, based on the metadata recommendation model, the profile based content recommendation model, and the viewing log. Create a content recommendation model of.

In detail, in operation S502, the content recommendation server 2500 may generate a metadata-based recommendation model using preprocessed content metadata and preprocessed viewing log information. The metadata-based content recommendation model may be classified into a method using only content metadata and a method using content metadata and viewing log information.

In addition, in step S502, the content recommendation server 2500 may generate a profile-based recommendation model using the preprocessed user profile and the preprocessed viewing log information. For example, a profile-based content recommendation model may be generated to recommend content by performing machine learning training of a viewing log.

In addition, in step S502, the content recommendation server 2500 may generate a content recommendation model based on the viewing log by using the preprocessed viewing log information. The content recommendation model based on the viewing log based on the item association analysis method and the user clustering method. Can be generated.

In step S504, a test viewing log is extracted and changed. In step S504, the content recommendation server 2500 may extract the viewing log information not used for generating the content recommendation model in step S502 from the viewing log information received from the data preprocessing server 2400, and extracts the extracted viewing log information. It can be used as test log information. In addition, the content recommendation server 2500 may extract and change test viewing log information from time to time in order to optimize the content recommendation model.

Step S506 is a step of determining a viewing log for a general performance test. In operation S506, the content recommendation server 2500 may determine that the content recommendation server 2500 is to be used as the general performance test viewing log by using the extracted test viewing log as it is.

Step S508 is a step of determining the viewing log for the metadata performance test. In operation S508, the content recommendation server 2500 may divide the extracted test viewing log according to the value of the metadata included in the viewing log, and determine the viewing log for the metadata performance test from the divided viewing log.

For example, the content recommendation server 2500 may have a specific metadata of “genre” and detailed metadata of “genre” having 10 values such as “action”, “melo”, “SF”, and a threshold ratio. If this is set to "70%," select only users who have watched VODs with "70%" or more of "genre" as "action" in the test viewing log information and the metadata is "genre"-"action". In this way, the recommended model performance evaluation unit 2540 extracts ten test viewing log information related to “genre” metadata such as “action”, “melo”, and “SF”. Can be extracted. Similarly, if there are M metadata such as "genre", "actor", "director", etc., for each metadata, similar to the "genre" metadata, test viewing log information about N detailed values may be extracted. Can be.

Step S510 is a step of determining a viewing log for a profile performance test. In operation S508, the content recommendation server 2500 may divide the extracted test viewing log according to the value of the profile included in the viewing log, and determine the viewing log for the metadata performance test from the divided viewing log.

In addition, the content recommendation server 2500 may, for example, call a particular profile “age”, and the “age” profile may have seven values such as “teen”, “20”, and “30”. If there is, the recommended model performance evaluation unit 2540 searches the profile information of the user ID in the test viewing log information and selects only the users whose age is "10", and the test is performed for the test with the age profile of "10". In this way, the content recommendation server 2500 also extracts seven test viewing log information related to the “age” profile, such as “10”, “20”, “30” and so on. Similarly, if there are M profiles such as "age", "gender", "family composition information", and so on, each of the N test viewing log information can be extracted for each profile similarly to the "age" profile. .

Step S512 is a step of evaluating the performance of each recommendation model using each test viewing log. In step S512, the content recommendation server 2500 displays the viewing log information for the general performance test, the viewing log information for the metadata performance test, and the viewing log information for the profile performance test, respectively. By applying to the log-based content recommendation model, it is possible to evaluate the performance of each content recommendation model.

Hereinafter, a method of recommending content by the content recommendation server 2500 according to an exemplary embodiment of the present invention will be described with reference to FIG. 6.

6 is a detailed flowchart of a method for recommending content by the content recommendation server 2500 according to an exemplary embodiment.

Step S600 is a step of receiving information about a user who views content. In operation S600, a content recommendation request of the user terminal 4000 is received from the web application providing server 2600, and the received content recommendation request may include various information about the user of the user terminal 4000. The user information may include, for example, a user ID, a content ID currently being viewed, and the like.

Step S602 is a step of classifying viewing log information of a user using the received user information. In operation S602, after viewing the viewing log information of the corresponding user by using the received user ID, the viewing log information of the user may be distinguished based on the same criteria as that of the viewing viewing log information for testing. That is, in step S602, the viewing log information of the user may be divided into at least one of general use, metadata use, and profile use.

Step S604 is a step of determining a content recommendation model to be used for generating a user content recommendation list. In step S604, the content recommendation model may be selected based on the viewing log information of the user identified in step S602. For example, in the recommendation model performance evaluation unit 2540, when the “age” profile is “20s” and the “gender” profile is “male”, the performance of the profile-based recommendation model is applied when all three models are applied. Is rated as the highest, the content recommendation unit 2550 indicates that the user is "age 20" and "gender" is "male" from the user viewing log information divided when the content recommendation request is received from the user. You can view and select profile-based recommendation models.

In operation S604, when there are several types of viewing log information divided from the viewing log information of the user, the content recommendation model having the highest performance may be selected for the viewing viewing log information corresponding to each viewing log information. have.

Step S606 is a step of generating a recommended content list. In operation S606, a list of recommended content for the corresponding user may be generated using the selected content recommendation model.

Step S608 is a step of deleting part of the generated content recommendation list. In operation S608, the content recommendation server 2500 may delete content already viewed by the user from the generated content recommendation list.

In operation S610, the content recommendation list is provided to the user terminal 4000. In operation S610, the content recommendation server 2500 may provide the user terminal 4000 with a content recommendation list generated through the web application providing server 2600.

Accordingly, the content recommendation server 2500 may generate and provide a list of recommended content using a content recommendation model having the best performance in response to the content recommendation request.

One embodiment of the present invention may also be embodied in the form of a recording medium including instructions executable by a computer, such as program modules, being executed by a computer. Computer readable media can be any available media that can be accessed by a computer and includes both volatile and nonvolatile media, removable and non-removable media. In addition, computer readable media may include both computer storage media and communication media. Computer storage media includes both volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Communication media typically includes computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave, or other transmission mechanism, and includes any information delivery media.

The foregoing description of the present invention is intended for illustration, and it will be understood by those skilled in the art that the present invention may be easily modified in other specific forms without changing the technical spirit or essential features of the present invention. will be. It is therefore to be understood that the above-described embodiments are illustrative in all aspects and not restrictive. For example, each component described as a single entity may be distributed and implemented, and components described as being distributed may also be implemented in a combined form.

The scope of the present invention is shown by the following claims rather than the above description, and all changes or modifications derived from the meaning and scope of the claims and their equivalents should be construed as being included in the scope of the present invention. do.

1000: broadcaster 2000: IPTV service system
2300: IPTV headend server 2400: data preprocessing server
2500: content recommendation server
2510: metadata-based recommendation model generator
2520: profile-based recommendation model generator
2530: viewer log-based recommendation model generator
2540: recommended model performance evaluation unit 2550: content recommendation unit
2600: web application providing server 3000: network network
4000: user terminal

Claims (17)

In a content recommendation system for recommending preferred content to a user in an interactive broadcasting service,
A content recommendation model generator that generates a plurality of content recommendation models using data based on the content recommendation,
A recommendation model performance evaluation unit for evaluating content recommendation performance of the generated plurality of content recommendation models using a plurality of types of test viewing log information; and
A content recommender which selects one of the generated recommendation models based on the evaluated recommendation performance and provides the user terminal with a content recommendation list generated based on the selected recommendation model;
Content recommendation system comprising a.
The method of claim 1,
And the recommendation model performance evaluation unit evaluates content recommendation performance by applying each of the plurality of types of test viewing logs to each of the generated plurality of content recommendation models.
The method of claim 2,
The content recommendation model generation unit generates a metadata-based content recommendation model, a profile-based content recommendation model, and a viewing log-based content recommendation model.
The method of claim 2,
The recommendation model performance evaluation unit evaluates content recommendation performances of the plurality of content recommendation models based on at least one of viewing log information for general performance testing, viewing log information for metadata testing, and viewing log information for profile testing. , Content recommendation system.
The method of claim 1,
The recommendation model performance evaluation unit selects the viewing log information not used to generate the content recommendation model from the viewing log information included in the data based on the content recommendation, and uses the viewing viewing log information for the test. .
The method of claim 5, wherein
The content recommendation unit selects a content recommendation model having the highest recommendation performance corresponding to the viewing log information of the user based on the viewing log information of the user who made the content recommendation request, and uses the selected content recommendation model to list the content recommendation list. To generate a content recommendation system.
The method according to claim 6,
The content recommendation unit is to delete the list of content watched by the user from the generated content recommendation list, content recommendation system.
The method of claim 1,
And the data based on the content recommendation includes at least one of metadata information about the content, user profile information, and viewing log information of the user.
The method of claim 8,
And the data based on the content recommendation is an error of the data removed from a data preprocessing server and a missing value is compensated for.
A method for generating a content recommendation model used for generating a user recommended content list in an interactive broadcasting service,
Generating a plurality of content recommendation models using at least one of metadata information about the content, user profile information, and viewing log information of the user;
Extracting test viewing log information from the viewing log information;
Extracting a plurality of types of test viewing log information from the extracted test viewing log information; and
Evaluating recommendation performance of the plurality of content recommendation models using the plurality of types of test viewing log information;
How to create a content recommendation model comprising a.
11. The method of claim 10,
The generating of the plurality of content recommendation models may include generating a metadata based content recommendation model, a profile based content recommendation model, and a viewing log based content recommendation model.
The method of claim 11,
The generating of the plurality of content recommendation models may include generating the metadata based content recommendation model using the content metadata and the viewing log information.
The method of claim 11,
The generating of the plurality of content recommendation models may include generating the profile based recommendation model using the user profile and the viewing log information.
The method of claim 11,
The generating of the plurality of content recommendation models may include generating the viewing log-based content recommendation model using the viewing log information.
11. The method of claim 10,
The plurality of types of test viewing log information may include viewing log information for general performance testing, viewing log information for metadata testing, and viewing log information for profile testing,
Evaluating the performance of the content recommendation model is based on at least one of the viewing log information for the general performance test, the viewing log information for the metadata test, and the viewing log information for the profile test, respectively, each of the plurality of content recommendation models. Evaluating a content recommendation performance of the content recommendation model.
11. The method of claim 10,
The extracting the test viewing log information comprises selecting viewing log information not used for generating the content recommendation model from the viewing log information of the user as test viewing log information.
In a method for providing recommended content to a user in an interactive broadcasting service,
Receiving user viewing log information from a user terminal,
Selecting one of a plurality of content recommendation models used to generate a content recommendation list based on the user viewing log information;
Generating a recommended content list based on the selected content recommendation model and the user viewing log information, and
Providing the generated recommended content list to the user terminal;
Including;
In the plurality of content recommendation models, content recommendation performance is evaluated by a plurality of types of test viewing log information in a content recommendation server.
Selecting one of the plurality of content recommendation models is selecting one of the plurality of content recommendation models based on the evaluated content recommendation performance.

Images