KR101259604B1 - Contents recommendation system and method based on social network - Google Patents

Abstract

PURPOSE: A system for recommending content based on a social network and a content recommendation method thereof are provided to supply content which is selected by social network members corresponding to ranking. CONSTITUTION: A content recommendation system(10) includes a tagging unit(100), a database(200), and a content recommendation unit(300). The content recommendation system receives situation data of a user obtained through a sensor module(30) from a mobile device(20). When a social network member of the user writes content, the tagging unit assigns one or more situation tags to the content. Content writer information and a situation tag of the content are stored in the database. The content recommendation unit recommends content which similarity is high to the user. [Reference numerals] (100) Tagging unit; (20) Mobile client; (200) Database; (30) Sensor module; (300) Content recommendation unit; (40) Content repository; (50) SNS repository; (AA) Wireless internet

Description

Content recommendation system and method based on social network

The present invention relates to a social network-based content recommendation system and a content recommendation method, and more particularly, a content recommendation system and content recommendation for providing a user with content most suitable for a user's situation using context information and a social network of the user. It is about a method.

Situational awareness, one of the key technologies in the ubiquitous computing era, has become a core technology for providing more advanced forms of human-centered services. In particular, the remarkable development of mobile technology and sensor technology has led to the convergence of situational awareness technology and various information processing technologies, enabling new and more effective human-centered services. In recent years, many researches are being conducted to provide more effective personalization services by applying context awareness technology in the recommendation system field, which is a representative form of personalization services.

 In his research, recommendation refers to the process of using the opinions and experiences of a group similar to the target user to effectively identify the content or products that might be of interest from a large selectable set. Defined. However, recommendation in a ubiquitous computing environment is that the computing system can pinpoint the user's potential needs before the user himself actively directs the needs of his or her content or products, and then timely It should be redefined as a process that allows access to content or products that may be of interest.

The user's situation information is largely divided into static situation information such as profile information such as age, gender, and occupation, and dynamic situation information with relatively frequent changes such as temperature, weather, or user's personal biometric information. Can be. As the importance of situational awareness technology increases, recent researches on recommendation systems are also suggesting techniques for identifying user preferences and interests by using user contextual information. For example, based on existing user preferences. In order to overcome the limitations of collaborative filtering, hybrid filtering techniques such as item based filtering and user based filtering, or music recommendation techniques based on fuzzy bayesian network are provided. Proposed.

On the other hand, a technique that introduces a social tagging technique has also been proposed. As a representative example, a clustering technique is applied to a folksonomy that is formed while a user tags a content tag. Therefore, a recommendation system may be used to effectively extract contents that are likely to be of interest to the user, and in this case, the clustered result may be changed according to the change of situation in consideration of the situation information of the individual user.

In addition, various studies have been attempted to overcome the limitations of existing recommendation techniques, and research on personalized recommendation techniques for providing personalized recommendation services is being actively conducted. However, most of these studies considered static contextual information such as user profile information, history of user's past contents or products, and location and time information was considered very limited as dynamic contextual information. In addition, in forming similar groups of target users, relying only on filtering techniques, which are existing techniques, does not effectively compose similar user groups that are expected to make similar tastes or dispositions as well as similar choices in various situations. It has a limitation. Finally, there is a lack of research on the context-aware recommendation system that provides more personalized contents in the ubiquitous computing environment in which various types of contents are exploded and generated.

Ellison et al. (3) analyzed three types of social capital according to the degree of participation of SNS by analyzing the purpose and form of use of SNS for the users of Facebook, a representative social network service. And direct research and analysis showed that these social assets have a significant impact on human life, including consumer life, well-being, academics, and hobbies. In other words, in the online environment as well as offline, the opinions or members of members belonging to a personal social network formed online are more trusted than the contents or information provided through portals or search engines.

Therefore, it is necessary to propose a recommendation technique for acquiring the current situation information of the user and recommending the content most suitable for the current situation of the user by using the acquired situation information and the social network of the user.

The present invention solves the problems of the existing recommendation techniques and provides a more advanced form of recommendation system in the ubiquitous computing environment and a new form of context awareness recommendation method.

According to an aspect of the present invention, a social network-based content recommendation method comprising: (a) when content is accessed by a member of the social network, at least one context tag describing a situation at the time of access is assigned to the content; And storing the accessed information of the content, the at least one context tag given and the information of the member who accessed the content, and (b) a member set (size N) consisting of at least some of the members of the social network; Calculating a situation similarity between a user belonging to the social network and each member included in the member set, (c) K persons having a high degree of similarity among the members constituting the member set ( Extracting a member of K <= N) into a set of similar members, (d) included in the set of similar members Calculating a score for each piece of content included in the set of content of the user for a set of content accessed by a member for a predetermined period of time; and (e) using the calculated score. There is provided a method for recommending content based on a social network, the method including outputting at least a portion of content included in the collection as recommended content.

According to an aspect of the present invention, a social network-based content recommendation system, and when a content is accessed by a member of the social network, a tagging unit for giving the content at least one context tag describing the situation at the time of access A member set (size N) consisting of a database storing information of the accessed content, the at least one context tag granted and information of a member accessing the content, and at least some of the members of the social network. A situation similarity calculating unit that calculates a situation similarity between one user who belongs to the social network and each member included in the member set, and K persons having high situation similarity among members constituting the member set (K <= N Pseudoset that extracts the members of An evaluation score calculator for calculating a score for each piece of content included in the set of content of the user for a set of content accessed by a member included in the set of similar members for a predetermined period; A social network-based content recommendation system is provided that includes a recommendation content output unit configured to output at least a portion of content included in the content set as recommended content using the evaluation score.

According to an embodiment of the present invention, in consideration of both static contextual information and dynamic contextual information, it is possible to more accurately grasp a user's situation and to effectively recognize a potential information need of the user in a ubiquitous environment to provide content more effectively. Can be. In addition, the social network of individual users can be used to improve the accuracy of the recommendation system and user satisfaction.

1 is a block diagram illustrating a social network-based content recommendation system according to an embodiment of the present invention.
2 is a block diagram illustrating a content recommendation unit included in a social network-based content recommendation system according to an exemplary embodiment.
3 is a diagram illustrating an example of a social network.
4 is a diagram for explaining a set of similar members.
5 is a flowchart illustrating a social network-based content recommendation method according to an embodiment of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS The present invention is capable of various modifications and various embodiments, and specific embodiments are illustrated in the drawings and described in detail in the detailed description. It is to be understood, however, that the invention is not to be limited to the specific embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

The terms first, second, etc. may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another.

The terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting of the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this application, the terms "comprise" or "have" are intended to indicate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, and one or more other features. It is to be understood that the present invention does not exclude the possibility of the presence or the addition of numbers, steps, operations, components, components, or a combination thereof.

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

1 is a block diagram illustrating a social network-based content recommendation system according to an embodiment of the present invention.

Referring to FIG. 1, the content recommendation system 10 according to an exemplary embodiment may include a tagging unit 100, a database 200, and a content recommendation unit 300.

When a member of the social network accesses arbitrary content, the tagging unit 100 may assign at least one context tag to the accessed content. The situation tag is a tag that describes the situation at the time a member of the social network accesses the content. Table 1 below categorizes examples of situation tags assigned to content into major and minor categories.

Main Category Small classification Situation tag Situation data
Acquisition method location - Bedroom, kitchen / dining room, bathroom, living room,
dress room Situation generator Vital signs Heart rate> normal excitement electrocardiogram
sensor Heart rate = normal Excitement weather - Sunny, cloudy, rain, snow Situation generator season December to February winter - March-May spring June to August summer September to November autumn gender - Men and women User profile age 5 to 11 child User profile 12 to 19 Youth 20 to 29 20's 30 to 39 30s 40 to 49 40 spaces 50 to 59 50s 60- old man

In order to utilize numerical data obtained through various sensors or user profiles in the recommendation system, it is necessary to conceptually define the context data as situation tags. For example, referring to Table 1, when the situation data over time is March to May, it may be expressed as a situation tag of 'spring'. In addition, the physical situation data may be expressed as 'excitement' when the reference value is lower than the reference value based on 60 to 100 times during one minute, which is the normal heart rate of an adult, and 'non-excited' when the reference value is higher than the reference value.

When a member of the social network accesses the content and the tagging unit 100 assigns the context tag to the content, the information about the accessed content, the at least one context tag and the information about the member who accessed the content May be stored in the database 200. In addition, the database may store the entire context tag list, the list of member information of the social network, and / or the list of content information.

The content recommendation unit 300 may provide a content list suitable for the user when a user of the social network wants to receive the content. To this end, the content recommendation unit 300 may include a situation similarity calculator 310, a similar member set configuration unit 320, an evaluation score calculator 330, and a recommended content output unit 340. Each component of the content recommendation unit 300 that performs the content recommendation method according to the present invention will be described in detail later with reference to FIG. 2.

Meanwhile, the content recommendation system 10 may receive the context data or the context tag from the mobile terminal 20 of the user. In this case, the user terminal () may obtain the situation data through the sensor module 30. In addition, the content recommendation system 10 may transmit a recommendation list to the user terminal. In addition, the content recommendation system 10 may communicate with the SNS repository 40 to obtain member information of the social network, and may communicate with the content repository 50 to obtain content information. That is, the application running on the user's mobile terminal 20 transmits the user's current situation data obtained from the sensor module 30 to the content recommendation system 10 through the wireless Internet, and the content recommendation system 10 is the present invention. By applying a content recommendation method according to an embodiment of the present invention, a list of contents corresponding to a current situation of a user among the contents stored in the content repository 50 may be extracted, and a recommendation list may be constructed. 10) may communicate with the SNS repository 40 to obtain member information of the social network.

2 is a block diagram illustrating a content recommendation unit 300 included in a social network-based content recommendation system according to an exemplary embodiment.

First, the context similarity calculating unit 310 may extract a member set composed of at least some of the members of the social network, and calculate the context similarity between the user who is to be recommended for content and each member included in the member set.

Prior to constructing a content recommendation list to be provided to a user, an item-based filtering technique, which is a typical recommendation technique, forms a group of similar users having similar inclinations or tastes as the user and based on this, A list of recommendations will be organized based on content or products. In this case, the most common method for judging users with similar tendencies or tastes is to directly assign rating points to content or products experienced by each member, and calculate similarity between users or items based on this. will be. In the content recommendation system or method according to the present invention, a similar preference user group needs to be configured. Unlike the existing methods of configuring the similar user group based on user preference, the content recommendation method according to the present invention is more differentiated. By calculating the similarity of situation between members of social network and users, we construct similar user group based on this.

To this end, the situation similarity calculator 310 calculates the situation importance for each member included in the member set and each situation tag included in the entire situation tag set, and based on the calculated situation importance, the user and the The situation similarity between each member included in the member set can be calculated. In this case, the situation importance of the situation tag B of the member A may be calculated based on the number of times that the situation tag B is assigned by the member A.

In more detail, the context similarity calculator 310 hyperlinks a content, a context tag, and a user in contact with the content as shown in Equation 1 below to calculate the context similarity between the social network member and the user. It is organized by link structure between objects such as That is, the link structure F is represented by a tuple consisting of U, T, C, and Y. In this case, U, T, and C represent members, context tags, and contents constituting a social network, respectively, and Y represents a ternary relationship between U, T, and C. FIG.

[Equation 1]

,

,

As a result, an undirected hypergraph may be constructed through the above modeling step, and based on the equation, the following equation 2 may be applied to calculate the similarity between the members of the social network.

&Quot; (2) &quot;

는 구성원A_i에 의해 태깅된 전체 상황 태그의 평균 태깅 횟수를 의미한다. In Equation 2, when a matrix S between a social network member and a situation tag is given, CF _ij denotes a situation similarity between the i th member and the j th member of the social network. S _ik denotes the situation importance of the k th situation tag C _k for the social network i th member A _i , which may be calculated as the number of times the situation tag C _k has been tagged by the member A _i .

Denotes the average number of tagging of all situation tags tagged by member A _i .

Next, the similar member set configuration unit 320 extracts members having similar high relative similarity values to the current situation of the user among all social network members according to the similarity of the situation calculated through Equation 2, and selects the similar member set. It consists of 3 to 4, a similar member set will be described.

3 is a diagram illustrating an example of a social network that can be applied to a content recommendation system according to an embodiment of the present invention. As shown in FIG. 3, if a social relationship is established between each member, each member may be connected by an edge. 4 is a diagram illustrating a similar member set of user A configured in the social network of FIG. 3. Referring to FIG. 4, when the situation similarity calculating unit 310 calculates the situation similarity between the user A and the remaining members, when the size of the similar member set is set to 4, the similar member set forming unit 320 may determine the situation similarity. A similar member set can be formed by extracting four members D, F, I, and J having high values.

Referring back to FIG. 2, the evaluation score calculator 330 extracts content accessed by a member included in the similar member set for a predetermined period of time to configure a content set. Thereafter, the evaluation score calculator 330 may calculate an evaluation score between each member included in the similar member set and each content included in the content set.

In order to apply the item-based filtering technique in the general recommendation system, the user's explicit evaluation score of the content or product should be used. In other words, the user is recommended contents that are predicted to be preferred by the user based on the content that the user has obtained a good score from the past. However, assigning a user's own evaluation score to the contents experienced by the user not only causes inconvenience to users but also prevents the accuracy and reliability of the recommendation list, thereby hindering the performance of the entire recommendation system. In addition, since the similarity between users is not considered, the accuracy of the predicted correlation between contents decreases if the user is based on the evaluation of other users whose preferences are not similar. Therefore, the evaluation score calculation unit 330 included in the content recommendation system according to the present invention performs an automatic evaluation of the content in consideration of the situation information at the time when the user purchased or encountered the product or the content, rather than an explicit evaluation from the user. And calculate the evaluation score.

To this end, the evaluation score calculator 330 may configure the content set by reading the content accessed for the predetermined period for each member included in the similar member set and adding the content to the set of the content. For each content included in the content set, the number of times a situation tag is assigned to each content may be calculated, and an evaluation score of the user for each content included in the content set may be calculated.

An example of a method of calculating the evaluation score is described in more detail: a group of k members constituting a set of similar members, a set S of m pieces of content that they have accessed over a period of time, and lastly a contributor to each content. The situation tag set C describing the situation at the time may be defined as in Equation 3.

&Quot; (3) &quot;

에서 구성원 u_i가 접한 콘텐츠 s_i에 대한 평가점수 CP_ij은 아래의 수학식 4를 통해 산출될 수 있다.
In this case, u _i and s _i mean the i th member and the j th content, respectively. Based on this

The evaluation score CP _ij for the content s _i contacted by the member u _i can be calculated through Equation 4 below.

&Quot; (4) &quot;

At this time, c _q represents the entire set of situation tags assigned to the corresponding content. cf _iq denotes the number of times situation q has been tagged in the content s _i , N _i denotes the number of all contents experienced by the participants u _i in the past, and n _q denotes the total number of contents tagged by the situation q.

Next, the recommended content output unit 340 may output at least some of the contents included in the content set as recommended content using the calculated evaluation score.

In one embodiment, the recommended content output unit 340 may output the recommended content by applying a general item-based filtering technique based on the calculated evaluation score.

In another embodiment, the recommended content output unit 340 calculates a rank value for each content included in the content set for the user, using the calculated situation similarity and the calculated score. Based on the rank value, at least a part of the content included in the content set may be extracted, and the extracted at least part of the content may be output as recommended content. In this case, the rank value for each content may be calculated by a page rank algorithm based on the calculated situation similarity and / or the calculated evaluation score.

On the other hand, the recommended content output unit 340 may arrange the respective contents according to the calculated rank value and then output the contents in the sorted order to provide the recommended contents.

An example of a method of calculating a rank value will be described in more detail. The rank value R _ai for a content S _i having a preference CP value equal to or greater than a threshold t (0 <t) according to the current situation of the user U _a is represented by the following equation. It can be calculated through Equation 5.

[Equation 5]

는 유사 사용자 집단 내 구성원 U_a가 경험한 모든 콘텐츠에 부여된 가중치 값의 평균값을 의미한다. 결과적으로 최종적으로 계산된 각 콘텐츠의 랭킹 값 R에 따라 상위 k개의 콘텐츠를 추천 항목으로 최종 구성하게 된다.
Where k is the total number of members in the group of similar users.

Denotes an average value of weights assigned to all contents experienced by member U _a in the similar user group. As a result, the top k contents are finally configured as recommended items according to the ranking value R of each content finally calculated.

5 is a flowchart illustrating a method for recommending content by a social network-based content recommendation system according to an embodiment of the present invention. An example of a social network-based content recommendation method according to the present invention will be described with reference to FIG. 2. Since the situation similarity, the evaluation score, and the rank value calculated while the content recommendation method according to the present embodiment is performed have been described above, a detailed description thereof will be omitted.

First, in step S100, the social network-based content recommendation system may calculate a situation similarity between the user who is to be recommended for the content list and the members of the social network.

In operation S110, the social network-based content recommendation system may configure a similar member set by extracting K members having high context similarity.

In operation S120, the social network-based content recommendation system may configure a content set by extracting content that members of the similar member set have accessed for a predetermined period of time in the past.

In operation S130, the social network-based content recommendation system may calculate a user's evaluation score for each content included in the set of content.

In operation S140, the social network-based content recommendation system may assign a ranking value to each content included in the content set using contextual similarity and / or evaluation score.

In operation S150, the social network-based content recommendation system may provide the user with the high ranking content as the recommended content.

Therefore, the social network-based content recommendation method according to an embodiment of the present invention does not simply recommend content with a large number of tagging times or content that has been given a high evaluation score by other members. Depending on the similarity of the situation between members, you can provide a more personalized and reliable list of contents.

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 detailed 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. .

Claims (16)

As a method for recommending content in a social network based content recommendation system,
(a) the social network-based content recommendation system, when the content is accessed by a member of the social network, gives the content at least one context tag describing the situation at the time of access, and the information of the accessed content Storing the at least one context tag and information of a member accessing the content;
(b) the social network-based content recommendation system targets a member set (size N) composed of at least some of the members of the social network, and includes one user belonging to the social network and each member included in the member set. Calculating a situation similarity between the two;
(c) extracting, by the social network-based content recommendation system, members of K persons (K <= N) into similar member sets in order of the highest similarity of situations among the members constituting the member set;
(d) the social network-based content recommendation system evaluates an evaluation score for each content included in the set of contents of the user, targeting a set of contents accessed by a member included in the set of similar members for a predetermined period of time; Calculating; And
(e) the social network-based content recommendation system outputting at least a portion of the contents included in the content set as recommended content using the evaluation score
Including but not limited to:
In the step (b), the social network-based content recommendation system,
(b-1) calculating the situation importance for each member included in the member set and each situation tag included in the entire situation tag set, wherein the situation importance for the situation tag B of the member A is the situation tag B; Is calculated based on the number of times assigned by member A; And
(b-2) calculating situation similarity between the user and each member included in the member set based on the calculated situation importance;
Content recommendation method in a social network-based content recommendation system comprising a.
delete The method of claim 1,
The situation similarity is

(Where CF _ij is the situation similarity between the i th member and the j th member of the social network, S _ik is the situation importance for the k th situation tag of the i th member of the social network,

Is the average situation importance for the entire situation tag of the i-th member).
The method of claim 1,
The step (d)
(d-1) reading the content accessed during the predetermined period for each member included in the similar member set and adding the same to the set of contents;
(d-2) calculating an evaluation score for each content included in the content set of the user based on the content included in the content set and the context tag assigned to each content included in the content set
Content recommendation method in a social network-based content recommendation system comprising a.
5. The method of claim 4,
The evaluation score,

Where CP _ij is the evaluation score for the j th content of the i th member, c _q is the set of situation tags assigned to the j th content, cf _jq is the number of times the q th situation tag is assigned to the j th content , N _i is the number of all contents accessed by the i th member during the predetermined period, and n _q is the total number of contents assigned to the q th situation tag). Way.
The method of claim 1,
In step (e),
Calculating a rank value for each content included in the content set for the user using the calculated situation similarity and the calculated score; And
Extracting at least some of the contents included in the content set based on the calculated rank value, and outputting the extracted at least some contents as recommended contents
Content recommendation method in a social network-based content recommendation system comprising a.
The method according to claim 6,
The rank value for each content is
A content recommendation method in a social network-based content recommendation system, which is calculated by a page rank algorithm with a weight based on at least one of the calculated situation similarity and the calculated evaluation score.
The method of claim 3,
In step (e),

(R _aj is a rank value for the j-th content of the a-th member of a (a is a natural number greater than or equal to K)

Is the average rank value for the entire contents accessed by the a-th member during the period, and N _i is the number of all contents accessed by the i-th member during the period; Calculating a value; And
Extracting at least some of the contents included in the content set based on the calculated rank value, and outputting the extracted at least some contents as recommended contents
Content recommendation method in a social network-based content recommendation system comprising a.
As a social network based content recommendation system,
A tagging unit, when the content is accessed by a member of the social network, assigning the content with at least one context tag describing a situation at the time of access;
A database for storing information of the accessed content, the at least one context tag assigned and information of a member accessing the content;
A situation similarity calculating unit configured to calculate a situation similarity between a user belonging to the social network and each member included in the member set, targeting a member set (size N) composed of at least some of the members of the social network;
A similar member set constituting unit extracting K members (K <= N) as similar member sets in the order of the highest similarity of situations among the members constituting the member set;
An evaluation score calculator configured to calculate an evaluation score for each content included in the set of content of the user, based on a set of contents accessed by a member included in the set of similar members for a predetermined period of time; And
A recommendation content output unit configured to output at least some of contents included in the content set as recommended content using the calculated evaluation scores;
The situation similarity calculation unit,
The situation importance is calculated for each member included in the member set and each situation tag included in the entire situation tag set. Here, the situation importance of the situation tag B of the member A is assigned by the member A. Is calculated based on the number of times
And a situation similarity between the user and each member included in the member set based on the calculated situation importance.
delete 10. The method of claim 9,
The situation similarity is

(Where CF _ij is the situation similarity between the i th member and the j th member of the social network, S _ik is the situation importance for the k th situation tag of the i th member of the social network,

Is an average situation importance for the overall situation tag of the i-th member).
10. The method of claim 9,
The evaluation score calculation unit,
For each member included in the similar member set, the content accessed during the predetermined period of time is read and added to the set of contents.
Social network-based content recommendation for calculating an evaluation score for each content included in the set of the content of the user based on the content included in the content set and the context tag assigned to each content included in the content set system.
The method of claim 12,
The evaluation score,

Where CP _ij is the evaluation score for the j th content of the i th member, c _q is the set of situation tags assigned to the j th content, cf _jq is the number of times the q th situation tag is assigned to the j th content N _i is the number of all contents accessed by the i-th member during the predetermined period, and n _q is the total number of contents to which the q-th situation tag is assigned.
10. The method of claim 9,
The recommended content output unit,
The rank value for each content included in the content set for the user is calculated using the calculated situation similarity and the calculated score.
The social network-based content recommendation system extracts at least a part of the contents included in the content set based on the calculated rank value, and outputs the extracted at least a part of the content as recommended content.
15. The method of claim 14,
The rank value for each content is
A social network based content recommendation system calculated by a page rank algorithm with a weight based on at least one of the calculated situation similarity and the calculated evaluation score. 10. The method of claim 9,
The recommended content output unit,

(R _aj is a rank value for the j-th content of the a-th member, where a (a is a natural number equal to or greater than 1)

Is the average rank value for the entire contents accessed by the a-th member during the period, and N _i is the number of all the contents accessed by the i-th member during the period; Calculate the value,
The social network-based content recommendation system extracts at least a part of the contents included in the content set based on the calculated rank value, and outputs the extracted at least a part of the content as recommended content.

Images