KR20080096887A - Information search ranking system reflecting user's interest and its method - Google Patents

Abstract

The present invention relates to an information retrieval ranking system and method that reflects user interest.

The information retrieval ranking system disclosed by the present invention comprises an UAL collection module for collecting a plurality of user behavior logs (UAL) from a user terminal or a network provider server connected to an information communication network, and the document DB based on the collected user behavior logs. For all users gathered in the individual documents accumulated in, the interest rank (AR) is calculated from the product of the interest value (AV) and the user's behavior influence (IV) for all actions shown by each user, and is calculated in the interest rank DB. A rank calculation module for updating the rank of interest and a rank value for calculating and ranking a rank value (RV) from the rank of interest of interest rank DB for each document retrieved from the document DB by receiving a query-based search request from a user terminal. It includes a calculation module.

According to the present invention, by converting the behavior shown in the individual document to the interest and then assigning it to the document and applying it to the ranking, it is possible to provide the user with improved search results.

Description

RANKING SYSTEM BASED ON USER'S ATTENTION AND THE METHOD THEREOF}

1 is an exemplary view showing an entire system to which the technical spirit of the present invention is applied;

2 is a block diagram of an information retrieval ranking system according to the present invention;

3 is a detailed configuration diagram of an interest rank calculation module according to the present invention;

4 is a flowchart illustrating an information retrieval ranking method according to the present invention.

** Description of symbols for the main parts of the drawing **

200: information search ranking system

210: document collection module 220: search engine

230: UAL collection module 240: interest rank calculation module

242: interest value calculator 244: behavior influence calculator

250: rank value calculation module 10: document DB

20: UAL DB 30: Rank of interest DB

The present invention relates to a technique of calculating a user's interest in a document based on a user action log (UAL) and reflecting the same in a ranking of search results.

As is well known, the information retrieval system indexes documents corresponding to a query input by a user and generates search results. The documents included in the search results are provided in a ranked list form (set of links pointing to the documents) through statistical techniques such as content analysis and link analysis.

Here, 'document' is used in a somewhat vague sense, but can be generally defined as a web page. Documents such as web pages are largely composed of content and metadata, which form a comprehensive meaning that includes text, voice, and video files. The metadata may include document language, document title, document size, document identifier (eg, URL information), document format, category, and various other attributes.

On the other hand, the ranking of information retrieval mainly uses the content and metadata of the document, and the relationship information (eg, links or categories) between the documents. However, such information is described from the point of view of a provider who wants to generate or distribute the content, and does not reflect the viewpoint of the consumer who ultimately consumes the content. For example, information on the user's point of view, such as content that is noticeable or currently popular, is excluded from the ranking decision factor, while elements of the provider's position such as a title or a backlink in the document are employed.

A good example of an information retrieval system is Google. In the case of Google, based on the 'PageRank' technique, the link information (hyperlink, hyperlink) pointing to the document is analyzed in addition to the information included in the document, and the page rank value is assigned to the document. The information and the previously given page rank value (range of 0 to 10) are collected to provide a ranked search result. This technique can be well realized in websites where link information is the main factor.

However, user creation contents (UCC) and mobile contents, such as video contents and blog contents, which are rapidly increasing in recent years, lack link information. Therefore, a page rank technique cannot expect desirable search results.

The present invention converts the behaviors shown in individual documents (contents) into interests and assigns them to the corresponding documents to apply them to ranking, thereby providing the user with improved search results.

The information retrieval ranking method of the present invention for achieving the technical problem is basically applied to a system including a search engine that searches for documents accumulated in the document DB according to a search request and provides ranked search results.

According to a feature of the present invention, a process of collecting and accumulating a plurality of documents in a document DB through an information communication network, a process of collecting and accumulating user behavior logs (UAL) from a user terminal or a network provider server, and collected user behaviors The interest rank DB is calculated by calculating the interest rank (AR) from the product of the interest value (AV) and the behavior influence (IV) of all actions shown by each user for all users gathered in the individual documents accumulated based on the logs. Process of accumulating the rank value (RV) from the interest rank of the interest rank DB for each document retrieved from the document DB by receiving a query-based search request from the user terminal and providing a search result ranked by the user terminal. Is done.

)와, 행위 경과시간(t)을 변수로 하는 시그모이드 함수의 곱으로 계산한다.The interest value AV of the present invention is a predetermined weight given to each action performed by a user gathered in an individual document.

) Is calculated as the product of the sigmoid function whose behavior elapsed time t is a variable.

로 계산하되,

는 시그모이드 함수이고,

는 사용자 h가 행한 모든 행위의 총회수이며, m은

를 모든 사용자 수로 나눈 값이다.Also, the behavioral influence (IV)

Calculate as

Is the sigmoid function,

Is the total number of all actions performed by user h, and m is

Divided by the number of users.

Specific features and advantages of the present invention will become more apparent from the following detailed description based on the accompanying drawings. In the meantime, when it is determined that the detailed description of the known functions and configurations related to the present invention may unnecessarily obscure the subject matter of the present invention, it should be noted that the detailed description is omitted.

1 is a diagram illustrating an entire system to which the present invention is applied, and a user action log (UAL) for a document that is viewed and viewed by one or more contents (document) servers not shown linked to an information communication network. : Collects and indexes a plurality of documents through a document user terminal 100 generating a user action log) and an information communication network, and collects user activity logs from the document user terminal or from a network provider server (not shown). After calculating an Attention Value (AV) based on the collected user behavior log, an information retrieval ranking system 200 for assigning an Attention Rank (AR) to each collected document and a query based As a result, the information search ranking system 200 includes a document search user terminal 300 that receives a searched search result that reflects user interest by requesting a search.

In the present invention, the user terminal includes a mobile phone or computer capable of internet communication, and is divided into a 'document user terminal' and a 'document search user terminal' for convenience of description. 'Document User Terminal' is based on the aspect of reading documents (e.g. web surfing) from a number of content servers connected to the information communication network and generating a log of user actions (UAL) accordingly. 'Accumulates interest ranks for each document based on a plurality of user activity logs collected by the information retrieval ranking system 200, and then provides ranking results according to the query from the information retrieval ranking system 200. It is based on the receiving side. In practice, however, both may be understood to be the same 'user terminal'.

On the other hand, the user action log (UAL) is a log file that records the user actions (user actions) that occur when a document user terminal reads a document, 1. document identifier, 2. action identifier, 3. The type of action, 4. time of action, and 5. additional data.

In more detail, the document identifier is an identifier for a document that is the target of an action. The 'action identifier' is information for identifying the user who caused the action, for example, may be an IP address or a MAC address when the user uses a computer, and may be a mobile phone number or unique mobile phone information when using a mobile phone.

In addition, the 'type of behavior' refers to the act of reading (looking at the summary of documents, previews, etc.), playing (looking at videos, music, images, etc.), viewing details (seeing the full text of the document, not summary information). ), Archiving (keeping or storing documents), making a purchase (purchasing documents or content for sale), making recommendations (recommending documents or content to others), evaluating (individual Expressing opinions as digitized or regularized information), attaching additional information (giving additional information such as comments or tags to documents or contents), bookmarking (storing or storing addresses for later viewing of documents or contents) And the like's behavior. Meanwhile, the 'additional data' may include, for example, a user's location and an environment type (for example, moving) as environment information of an action performed by a user.

2 is a detailed configuration diagram of the information retrieval ranking system 200 of the present embodiment. As shown, the information retrieval ranking system 200 includes a document collection module 210, a search engine 220, and an UAL collection module ( 230, the interest value calculation module 240, the rank value calculation module 250, the document DB 10, the UAL DB 20, and the interest rank DB 30 are configured.

The document collection module 210 collects a plurality of documents through the information communication network and stores them in the document DB 10, and indexes them to correspond to a query-based search request (a search request from a document search user terminal). Save it.

The search engine 220 is an engine having functions necessary for a general search for searching the document DB 10 according to an input query word and providing a search result. According to a feature of the present invention, the search engine 220 provides a ranked search result, referring to the interest rank AR of the interest rank DB 30 associated with each of the documents constituting the search result, 'RV' reflecting 'is given and provided in the form of ranked list. This is described in detail below.

In the present invention, 'user interest' means a numerical value by applying a user's memory model (or memory model) as a main factor for document (content) evaluation. As is well known, humans recognize a fact or object and then gradually forget it over time. This is the human memory model represented by the sigmoid function. Conventionally, in order to quantify that the user's attention is focused on a document, the placenta is based on link information (Google's PageRank) or the total number of times the document is viewed. However, these indicators are merely cumulative indicators that do not reflect the temporal component of the memory model.

Meanwhile, as mentioned above, the UAL collection module 230 collects a user behavior log (UAL) from the document user terminal 100 or the network provider server, and stores and manages the user behavior log (UAL) in the UAL DB 20. The user behavior log UAL may be collected from a logging tool installed in the document user terminal 100 or software stored in a web browser, or may be stored in a network provider server. Of course, both collection methods can be used interchangeably.

The interest rank calculation module 240 configures the interest value calculator 242 and the behavioral impact calculator 244 as illustrated in FIG. 3 to collect all the users gathered in the individual document (p in Equation 1 below). For, calculate an interest rank (AR) from the product of the Attention Value (AV) for all actions shown by each user and the Influence Value (IV) of the user for each action, This is updated in the interest rank DB 30. The rank of interest AR is defined by the following equation.

.................. [수학식 1]

.................. Equation 1

In Equation 1, the interest value AV calculated by the interest value calculator 242 is represented by the following equation.

............................... [수학식 2]

............................... [ Equation 2 ]

는 사용자 h가 문서 p에 대해 k라는 행위(action)를 보인 것을 의미한다.

는 k라는 행위의 종류에 대해 기 부여된 소정의 가중치(weight)이며, t는 상기 관심값(AV)을 산출하는 시점의 시간과 k라는 행위가 실제 일어난 시간의 차이, 즉 k라는 행위의 경과시간(이하, 행위 경과시간)을 의미한다. here,

Means that user h showed k action on document p.

Is a predetermined weight given to the type of action k, and t is the difference between the time at which the interest value AV is calculated and the time k is actually occurring, i.e., the progress of the action k. It means time (hereinafter, elapsed time of action).

는 그 행위 경과시간에 따라 감소한다.The weight has a value between 0 and 1 depending on the type of action. For example, 0.2 for 'read', 0.4 for 'play', 0.5 for 'detail',... For example, 'recommended' may be defined as 0.9. Equation 2 is a sigmoid function whose behavior elapsed time t is a variable as defined on the right side, and models the aforementioned human memory. Thus weight

Decreases with the elapsed time of the act.

In the interest value AV calculated by the interest value calculator 242, the behavior influence IV of the user who shows the behavior should be considered. For example, if there are users who showed 1,000 actions and users who showed 10 actions, the former is 100 times more involved than the latter. In other words, when the user's interest in the document is calculated, the user who shows 1,000 behaviors exerts an influence that is excessively biased than the rest of the users. Therefore, a correction should be made in which the behavior influence IV of the user calculated by the behavior influence calculator 244 is reflected in the interest value AV. The behavioral impact (IV) on the user is represented by the following equation.

............................... [수학식 3]

............................... [ Equation 3 ]

는 시그모이드 함수를 나타내며,

는 사용자 h가 보인 모든 행위의 총회수이며, m은 상기

를 모든 사용자 수로 나눈 값이다(

). 이러한 [수학식 3]은 '사용자의 행위 표준편차'로 요약될 수 있다.here,

Represents the sigmoid function,

Is the total number of all actions shown by user h, and m is the above

Divided by all users (

). Equation 3 may be summarized as 'a standard deviation of a user's behavior'.

Meanwhile, when the search engine 220 extracts documents corresponding to the query by referring to the document DB 10, the rank calculation module 250 may rank information based on the interest rank AR for each extracted document. RV: Ranking Value is calculated and ranked. Only the interest rank AR described above may be used as an element for calculating the rank value, but in some cases, document similarity with respect to the query may be added as shown in the following equation.

...................... [수학식 4]

.......... [Equation 4]

Here, DocRel means predetermined document similarity with respect to the query word. Since the calculation of the sentence similarity has been disclosed in a number of cases implemented by various algorithms, detailed description thereof will be omitted.

Hereinafter, the information retrieval ranking method according to the above-described preferred embodiment will be described with reference to FIG. 4. First, the document collection module 210 collects and accumulates a plurality of documents through the information communication network (S100). This process is performed by a predetermined cycle in a conventional information retrieval system.

Subsequently, the UAL collection module 230 collects and accumulates a user behavior log (UAL) for the document from the document user terminal 100 or the network provider server (S200).

The interest rank calculation module 240 is based on the user behavior logs (UAL) collected in the S200 process, for all the users who showed the behavior in each document (document accumulated in the document DB), and all the behaviors shown by each user. The interest rank AR is calculated by reflecting the interest value AV and the influence IV of the user on each action (S300).

For the query-based search request from the document search user terminal 300, the search engine 220 searches for documents corresponding to the query in the document DB 10 (S400). The rank value calculation module 250 calculates the rank value RV by referring to the interest rank AR for each retrieved document from the interest rank DB 30, and then ranks the retrieved documents (S500). In this process, the document similarity as described above may be reflected in the rank value RV calculation.

Subsequently, the search engine 220 provides search results ranked at the document search user terminal 300 requesting the search (S600). Through this series of processes, the queryer (user) reads the high quality search results reflecting 'user interest'.

According to the present invention as described above, by using the behavior shown in the individual document as the user interest based on the human memory model, it is possible to provide a high-quality ranked search results sensitive to the user's interest or preference for the query-based search request. Can be. In addition, it is possible to provide a high-quality ranking search results for the user's work (UCC) that has recently increased rapidly.

Although the embodiments of the present invention have been described in detail above, the scope of the present invention is not limited thereto, and various modifications and improvements of those skilled in the art using the basic concepts of the present invention defined in the following claims are also provided. It belongs to the scope of rights.

Claims (12)

A ranking system including a search engine that searches documents accumulated in a document DB according to a search request and provides ranked search results. An UAL collection module for collecting a plurality of user activity logs (UAL) from a user terminal or a network provider server connected to the information communication network; Based on the collected user behavior logs, for all users gathered in the individual documents accumulated in the document DB, the interest is derived from the product of the interest value (AV) for all the behaviors shown by each user and the user's behavior influence (IV) An interest rank calculation module that calculates a rank AR and updates the interest rank DB; And A rank value calculation module that receives a query-based search request from a user terminal and calculates and ranks a rank value (RV) from the rank of interest of the rank of interest DB for each document retrieved from the document DB; Information search ranking system reflecting the user's interest, characterized in that it comprises a. The method according to claim 1, A document collection module for collecting a plurality of documents through the information communication network and indexing and storing them in the document DB; Information search ranking system reflecting the user's interest, characterized in that it further comprises. The method according to claim 1, The interest rank calculation module, A predetermined weight given to each action performed by one user gathered in the individual documents (

A value of interest calculator which calculates the value of interest AV from a product of a sigmoid function whose behavior elapsed time t is a variable; Information search ranking system reflecting the user's interest, characterized in that it comprises a. The method according to claim 1 or 3, The weight (

) Is an information retrieval ranking system reflecting user interest, characterized in that it has a value between 0 and 1. The method according to claim 1, The interest rank calculation module, An action influence calculator configured to calculate the action influence IV; Including; The behavioral influence (IV) is

Calculate as remind

Is the sigmoid function,

Is the total number of all actions performed by user h, and m is the above

Information search ranking system reflecting the user's interest, characterized in that divided by the number of users. The method according to claim 1, The rank value calculation module, And calculating the rank value by reflecting a document similarity degree to the query word in the rank of interest. The method according to claim 1, The user behavior log (UAL), An information retrieval ranking system reflecting user interest, comprising a document identifier, an action identifier, a type of action, and a time of the action. An information search ranking method provided by ranked search results. A first step of collecting and accumulating a plurality of documents in a document DB through an information communication network; A second step of collecting and accumulating a user behavior log (UAL) from a user terminal or a network provider server; Based on the collected user behavior logs, for all users gathered in the accumulated individual documents, the interest rank (AR) from the product of the interest value (AV) for all behaviors shown by each user and the user's behavioral impact (IV) A third step of calculating c) and accumulating it in the interest rank DB; And A fourth step of receiving a query-based search request from the user terminal and providing a search result obtained by ranking a rank value (RV) from the rank of interest of the rank of interest DB for each document retrieved from the document DB; Information search ranking method reflecting the user's interest, characterized in that it comprises a. The method according to claim 8, The interest value AV of the third process is A predetermined weight given to each action performed by one user gathered in the individual documents (

) And a sigmoid function multiplied by the elapsed time t as a variable. The method according to claim 8, The behavioral influence IV is

Calculate as remind

Is the sigmoid function,

Is the total number of all actions performed by user h, and m is the above

Information search ranking method reflecting the user's interest, characterized in that divided by the number of users. The method according to claim 8, The rank value RV of the fourth process is Information search ranking method reflecting the user's interest, characterized in that calculated by reflecting the document similarity of the query to the interest rank. The method according to claim 8, The user behavior log (UAL), An information retrieval ranking method that reflects user interest, comprising a document identifier, an action identifier, a type of action, and a time of the action.

Images