JP5182509B2 - Information recommendation device, information recommendation method, and information recommendation program - Google Patents

Description

  The present invention relates to an information recommendation device, an information recommendation method, and an information recommendation program, and more particularly, an information recommendation device, an information recommendation method, and an information recommendation method capable of recommending information in a field unexpected for a user based on characteristics of a community in the user's social network. Involved in information recommendation programs.

  The amount of information content such as product content and news existing on the Internet has become enormous, and it has become difficult for each user to find information appropriate to him without leaking. For sites that sell books, CDs, etc., even if you own a product that you are likely to like, if the user does not notice its presence due to the huge number of contents that the sales site has, the opportunity to sell the product There is a risk of losing.

  In order to solve this problem, it is recommended to recommend information content that the user is likely to like.

  As one of the methods, there is known a technique for estimating information that is likely to be a user's preference from data of other users whose data such as past behavior / preference of the user are similar.

  For example, in Patent Document 1, information contents registered in a community are compared with each other, and when there are more common contents than a preset threshold, it is determined that the contents are similar. And a method for recommending content that is not in the other.

  Patent Document 2 describes a method for analyzing an interest from each user's blog entry by ontology, and recommending an entry that is in one but not in the other when the ontology between users is similar. Yes.

  As another method, there is known a technique for estimating information that a user seems to like from data such as past behavior and preferences of a user's friend based on a recorded friend relationship represented by a social networking service (SNS) or the like. It has been.

  For example, Patent Document 3 describes a technique for determining a recommendation source of information based on the distance of friend relationships such as “friend” and “friend of friend” in SNS.

  Patent Document 4 describes a technique in which users belonging to the same community are recommended sources of information in SNS.

  As another method, information content is divided into categories in advance, and feature vectors for the content accessed by the user are created for each category. The feature vector of the user in the same category and the feature vector of the content not yet accessed by the user Comparing similar content (order matching) and recommending unexpected content by comparing user feature vectors and content feature vectors between different categories (cross matching) The technique to do is described in Patent Document 5.

JP2008-102846 JP2007-241754 JP 2006-309660 A JP 2006-059257 A JP2001-265808

  However, the techniques described in Patent Documents 1 and 2 for estimating information that seems to be a user's preference from data of other users whose data such as the user's past behavior / preference are similar are: There is a tendency for others similar to the user to recommend information about fields that were originally interested. For this reason, it is pointed out that a field unknown to the user is not effective, and information only in a field known to the user is recommended, and the diversity of information is lost.

  In addition, based on friendships described in Patent Documents 3 and 4, the technology for estimating user-preferred information from data such as past actions / preferences of user's friends is based on user preferences and friendships. Preference does not always match. Therefore, it is pointed out that the friendship between users does not guarantee the quality of information recommendation.

  In the technique described in Patent Document 5, cross-matching that compares the feature vector of a category known to the user with the feature vector of the content of an unknown category that the user has never accessed is theoretically possible. . However, regarding whether or not the user accepts the information content recommended by such comparison, there is a problem that there is no guarantee about the quality of the recommended information and the accuracy is low.

  In view of the above-described problems, an information recommendation device, an information recommendation method, and an information recommendation program suitable for recommending information in a field that is unexpected to the user and that can be easily accepted by the user are proposed.

  The above-mentioned problem was examined from a viewpoint different from the conventional proposal, and a solution was found by the following proposal. That is, an information recommendation device for recommending information possessed by another user to a predetermined user belonging to a community formed on the network, the community storage means for storing information on each community formed on the network And using the information about each community, the predetermined user and the other user in common and the characteristics of the community to which the predetermined user belongs and the characteristics of all the communities to which the predetermined user belongs A similarity calculating means for calculating a first index value representing a degree that the other user belonging to one or more communities in common with the user handles information having affinity for the predetermined user; Using the information about each community, the community to which the predetermined user and the other users belong commonly Among the users belonging to the user, the characteristics of the user who does not belong to the predetermined user and also belongs to the community to which the other user belongs, and the characteristics of the community to which the predetermined user and the other user belong in common Based on the unexpectedness calculation, a second index value representing a degree to which the other user who belongs to one or more communities in common with the predetermined user brings information in a field unexpected to the predetermined user is calculated. And recommending means for selecting the other user who recommends information to the predetermined user based on a result of calculating the first index value and the second index value for each user. To do.

  Here, the community formed on the network may be, for example, a community in which a theme provided in a social networking service (SNS) or the like is specified, or in a blog network without an explicit community, a community division method The community obtained by

  Examples of information related to the community include, but are not limited to, the number of users belonging to the community, the theme of the community, attributes such as sex and age of each user belonging to the community, and their distribution.

  Further, the user belongs to at least one community, but may belong to two or more communities.

  Further, the predetermined user means a user who receives information recommendation, and the other users mean users other than the predetermined user.

  ADVANTAGE OF THE INVENTION According to this invention, based on the characteristic of the community in a user's social network, the information of the field which is unexpected for a user and is easy for a user to accept can be recommended.

  Embodiments of the present invention will be described below with reference to the drawings.

[First Embodiment]
FIG. 1 is a diagram illustrating a configuration example of a first embodiment of the present invention. In FIG. 1, a server 3 is connected to a plurality of terminals 2 a to 2 n via a communication network 1 so as to be communicable. The server 3 has a function as an information recommendation device. As the communication network 1, for example, the Internet may be used. Further, it may be an intranet within a company. The terminals 2a to 2n may be personal computers or mobile phones. The server 3 includes community storage means 3a, similarity calculation means 3b, unexpected degree calculation means 3c, and recommendation means 3d. Such a server 3 can be realized by a CPU, a computer or the like having a program execution function. That is, in the case of a computer, the functions described below are realized by reading and executing an information recommendation program, which will be described later, stored in advance in an internal or external storage device (recording medium). The same applies to the second embodiment described below.

  In the server 3, the community storage unit 3 a stores information related to a community to which a user connected to the communication network 1 belongs using the terminals 2 a to 2 n. As information regarding the community, for example, there is a list of users belonging to each community, and FIG. 2 is an example thereof. FIG. 2 shows that users A, B, and C belong to the community c1, for example. Usually, users belong to multiple communities.

  The similarity calculation means 3b uses the information about each community stored in the community storage means 3a, and the characteristics of the community to which the predetermined user and other users belong in common and the characteristics of all the communities to which the predetermined user belongs Based on the above, a first index value representing the degree to which other users who commonly belong to the predetermined user and one or more communities handle information having affinity for the predetermined user is calculated.

  In the community described here, when there is an explicit theme such as SNS, information on the theme is mainly handled. In addition, there is no explicit theme in the community obtained by the community division method, but similar information is handled in the community. Therefore, the community can be seen as one field.

  The unexpectedness calculation means 3c uses the information about each community stored in the community storage means 3a, and the predetermined user does not belong to other users belonging to the community to which the predetermined user and other users belong. Based on the characteristics of the users who belong to the community to which the user belongs, and the characteristics of the community to which the predetermined user and other users belong in common, A second index value representing the degree to which the user brings information in a field unexpected for a given user is calculated.

  The recommendation unit 3d provides information to a predetermined user based on the results of calculating the first index value and the second index value calculated from the similarity calculation unit 3b and the unexpectedness calculation unit 3c, respectively. Select other users to recommend.

  Next, the operation will be described.

  FIG. 3 is a flowchart showing an example of a process progress of the information recommendation program in the first embodiment of the present invention.

  First, the community storage means 3a memorize | stores the information regarding each community (step s11).

  The similarity calculation means 3b uses the information about each community stored in the community storage means 3a, and the characteristics of the community to which the predetermined user and other users belong in common and the characteristics of all the communities to which the predetermined user belongs And a first index value (hereinafter referred to as “a”) indicating the degree to which other users who commonly belong to the predetermined user and one or more communities handle information having affinity for the predetermined user. (Referred to as “similarity”) (step s12).

Below, the example at the time of making the feature of a community into the number of users who belong to the community is shown. In general, it is estimated that the smaller the number of users belonging to a community, the stronger the unity within that community, and the higher the number of communities to which a given user and other users belong in common, the higher the affinity between the two Is done. Therefore, the similarity s _i of another user i for a given user is determined by, for example, the number of users belonging to a community to which the given user and other users belong in common, and each of all communities to which the given user belongs. Using the relative ratio of the average number of users belonging to the following equation 1, the following equation 1 can be given. However, it is not limited to this.

In Equation 1, #c ¹ _m is the number of users belonging to the mth community to which the predetermined user belongs, M is the number of communities to which the predetermined user belongs, and #c ⁰ _k is common to the predetermined user and other users i The number of users belonging to the k-th community to which K belongs, and K represents the number of communities to which a predetermined user and other users belong in common. As another example, the logarithm of the above formula may be newly given as the similarity.

  The unexpectedness calculation means 3c uses the information about each community stored in the community storage means 3a, and the predetermined user does not belong to other users belonging to the community to which the predetermined user and other users belong. Based on the characteristics of the users who belong to the community to which the user belongs, and the characteristics of the community to which the predetermined user and other users belong in common, The user calculates a second index value (hereinafter referred to as “expected degree”) representing the degree of information that is unexpected for a given user (step s13).

  FIG. 4 is an example showing the state of the community and the users belonging to the community. In this example, the predetermined user U belongs to the community c1 but does not belong to the communities c2 and c3. The other users A to G belong to the community c1 in common with the predetermined user U. Other users C to F also belong to the community c2, and other users G also belong to the community c3. In this way, the other users C to G are unexpected because the community c1 that belongs to the predetermined user U exists in common and the community c2 or c3 to which the predetermined user U does not belong. This is a calculation target of the degree calculation means 3c. The other users A and B have a community c1 that belongs in common with the predetermined user U, but do not further belong to the community to which the predetermined user U does not belong. It is not necessary. Alternatively, the unexpectedness of other users A and B may be defined as 0. Since the other users X1 to X3 and Y1 to Y3 do not have a community that belongs to the predetermined user U, the unexpectedness calculating unit 3c does not calculate the unexpectedness for these other users.

Below, the example at the time of making the feature of a community into the number of users who belong to the community is shown. In FIG. 4, the number of users belonging to both the communities c1 and c3 is the other user G1, which is smaller than the total of the other users C to F belonging to both the communities c1 and c2. This is presumed to indicate that the information of the community c3 is more surprising information in the community c1 than the information of the community c2. That is, it is calculated that for the predetermined user U, the other user G has a higher degree of information that is unexpected to the predetermined user U than the other users C to F. The unexpectedness u _i of another user i for a given user U is determined by, for example, among users belonging to a community to which the given user U and other users belong in common, Using the relative ratio between the number of users belonging to the community to which the user belongs and the number of users belonging to the community to which the predetermined user and other users belong, it can be given as follows. However, it is not limited to this.

u _i = # (c1) / # (c1∩ck)
That is, # (c1) represents the number of users belonging to the community c1, and # (c1∩ck) represents the number of users belonging to both the communities c1 and ck (in FIG. 4, ck is c2 and c3). In this example, the unexpectedness of the other users C to F is calculated as 8/4 = 2, and the unexpectedness of the other user G is calculated as 8/1 = 8.

  FIG. 5 shows a situation where a predetermined user U and other users P and Q commonly belong to the community c4, the other user P belongs to the community c5, and the other user Q also belongs to the communities c5 and c6. It is an example which shows. In this example, since the other user Q has more communities than the other user P, the unexpectedness for the predetermined user U is calculated that the other user Q is higher than the other user P. . The unexpectedness of the user Q for the predetermined user U in this example can be given as follows, for example, but is not limited thereto.

u _Q = # (c4) / # (c4∩c5) + # (c4) / # (c4∩c6)
When the unexpectedness is given in this way, it is obvious that the unexpectedness u _P = # (c4) / # (c4∩c5) of the other user P becomes higher.

FIG. 6 is an example showing a situation where a predetermined user U and another user R commonly belong to the communities c7 and c8, and the other user R also belongs to the community c9. In this example, the unexpectedness of the other user R is u _R = # (c7) / # (c7∩c9), when only the communities c7 and c9 are considered, and the community c8 and Applying the above calculation formula taking only c9 into consideration, u _R = # (c8) / # (c8∩c9). If # (c7) / # (c7∩c9) <# (c8) / # (c8∩c9), the information of the community c9 for the community c7 is more surprising than the information of the community c9 for the community c8. No, ie easy to spread. Therefore, in this example, the unexpectedness of the other user R is preferably u _R = # (c7) / # (c7 （c9).

In view of the above, the unexpectedness u _i of another user i for a given user can be given, for example, as in the following Equation 2, but is not limited to this.

In Equation 2, #c ⁰ _k is the number of users belonging to the kth community to which the predetermined user and other user i belong in common, and #c ⁱ _j does not belong to the predetermined user but other user i belongs to This represents the number of users belonging to the jth community. As another example, the logarithm of the above-described formula may be newly given as an unexpected degree.

  The recommendation unit 3d selects other users who recommend information to a predetermined user based on the results of calculating the similarity and the unexpectedness calculated from the similarity calculation unit 3b and the unexpectedness calculation unit 3c, respectively. A selection is made (step s14).

  First, another user with a high degree of surprise is selected first. Here, a plurality of other users whose unexpectedness is higher than a predetermined value may be selected, or a predetermined number of other users may be selected in descending order of unexpectedness.

  Here, it is estimated that other users having a high degree of similarity are similar to the predetermined user, and information on such other users is easily accepted by the predetermined user. Therefore, it is desirable to preferentially select other users with high similarity. If the condition is not satisfied, other users do not need to be selected.

  Next, the effect of the first embodiment will be described.

  In the first embodiment, a community to which the predetermined user does not belong, that is, information on a field unexpected for the predetermined user, characteristics of one or more communities to which the predetermined user and other users belong in common, It is possible to make a recommendation based on the characteristics of users who belong to a community to which a predetermined user and other users belong and belong to a community to which the predetermined user does not belong and other users.

  By recommending information of other users who have a high degree of surprise, it is possible to recommend information in a field unexpected for the user. Furthermore, when the degree of similarity is high among other users having a high degree of unexpectedness, the predetermined user can easily accept the recommendation.

  In the prior art, the user was not able to appropriately recommend information in an unknown field, but in this embodiment, as described above, the community recommends appropriate information even in an unknown field. It is secured by the unexpectedness and similarity calculated from the characteristics of.

[Second Embodiment]
FIG. 7 is a diagram showing a configuration example of the second embodiment of the present invention. Differences from the first embodiment will be described. The server 3 ′ includes a user desire storage unit 3e in addition to the community storage unit 3a, the similarity calculation unit 3b, the unexpected degree calculation unit 3c, and the recommendation unit 3d.

  The user desire storage means 3e uses the input device such as a keyboard to make a request about a predetermined user connected to the communication network 1 using the terminals 2a to 2n with respect to the first index value and the second index value. Accept and store input.

  Here, it is not always necessary to input the user's wish, and a predetermined index value prepared in advance may be used as a default.

  The recommendation unit 3d is obtained by the user's desired storage unit 3e and the result of calculating the first index value and the second index value calculated from the similarity calculation unit 3b and the unexpectedness calculation unit 3c, respectively, for each user. Based on the index value desired by the predetermined user, another user who recommends information to the predetermined user is selected.

  Next, the operation will be described.

  FIG. 8 is a flowchart illustrating an example of processing progress of the second embodiment. The same steps as those in the first embodiment described with reference to FIG. 3 are denoted by the same step numbers, and description thereof is omitted.

  The user desire storage means 3e accepts and stores the input of the predetermined index of the first index value and the second index value via an input device such as a keyboard (step s21).

  FIG. 9 is an example of a screen for inputting an index value desired by the user. On the input screen 4, a scale 41a relating to the degree of similarity, a slider 42a that can be continuously moved on the scale 41a, a scale 41b relating to the unexpectedness, and a slider 42b that can be continuously moved on the scale 41b are displayed. . The user moves the sliders 42a and 42b using an input means such as a mouse and inputs a desired index value. The user can appropriately change the input in accordance with his / her wish at that time, and the user desire storage means 3e stores the changed desired index value as needed. Further, the input method is not limited to this example, and may be replaced by other modes. For example, a desired index value may be input numerically using an input means such as a keyboard.

  The recommendation means 3d calculates the result of calculating the similarity and the unexpectedness calculated from the similarity calculation means 3b and the unexpectedness calculation means 3c for each user, and the predetermined user obtained by the user desire storage means 3e In view of the index value to be determined, it is determined that the information of other users close to the hope is recommended to the predetermined user (step s22).

  Here, “close to hope” may be calculated by calculating the Euclidean distance between the desired unexpectedness / similarity and the unexpectedness / similarity of other users, and may be set to be a small value, or simply Alternatively, a predetermined upper limit and lower limit may be provided for the desired unexpectedness and similarity, respectively, and it may be set that the unexpectedness / similarity of other users falls within the range. In addition, only the information of other users who are closest to the hope may be recommended, or information of a plurality of users that are close to the hope may be recommended by weighting them according to the closeness of the hope.

  Next, effects of the second embodiment will be described.

  In the second embodiment, by accepting a request from a predetermined user, when a predetermined user wants to spread insights about a field that he / she has not noticed before, input a higher degree of surprise This makes it possible, and when a predetermined user wants to obtain information on a field in which he / she is familiar, he / she can input it with a lower degree of surprise. As described above, the quality of information recommendation can be adjusted to a predetermined user's desire.

  For example, by incorporating an e-commerce system, the present invention can provide product information in a field unexpected for the user, and can promote further purchase by the user.

  In addition, it is possible to contribute to improvement of user activity by providing an unexpected field for the user in an information providing site such as news.

It is a block diagram which shows schematic structure of the information recommendation apparatus which concerns on the 1st Embodiment of this invention. It is an example of a list of users belonging to each community. It is a flowchart which shows the example of process progress which concerns on the 1st Embodiment of this invention. It is the figure which showed the example showing the mode of the community and the user who belongs to it. It is the figure which showed the example showing the mode of the community and the user who belongs to it. It is the figure which showed the example showing the mode of the community and the user who belongs to it. It is a block diagram which shows schematic structure of the information recommendation apparatus which concerns on the 2nd Embodiment of this invention. It is a flowchart which shows the example of the process progress which concerns on the 2nd Embodiment of this invention. It is the figure which showed the example of the screen which inputs the index value which a user desires.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Communication network 2a-2n Terminal 3, 3 'Server 3a Community storage means 3b Similarity calculation means 3c Unexpectedness calculation means 3d Recommendation means 3e User wish storage means 4 Input screen 41a, 41b Scale 42a, 42b Slider

Claims (15)

An information recommendation device that recommends information held by other users to a predetermined user belonging to a community formed on a network,
Community storage means for storing information about each community formed on the network;
Using the information on each community, based on the characteristics of the community to which the predetermined user and the other users belong in common and the characteristics of all the communities to which the predetermined user belongs, Similarity calculation means for calculating a first index value representing the degree to which the other users belonging in common in one or more communities handle information having affinity for the predetermined user;
Of the users belonging to the community to which the predetermined user and the other user belong in common using the information on each community, the user belonging to the community to which the predetermined user does not belong and the other user belongs Based on the characteristics and characteristics of the community to which the predetermined user and the other user belong in common, the other user who belongs to the predetermined user and one or more communities in common An unexpectedness calculating means for calculating a second index value representing a degree of bringing information in a field unexpected for the user;
Recommending means for selecting the other user who recommends information to the predetermined user based on a result of calculating the first index value and the second index value for each user;
An information recommendation device comprising: The information recommendation device according to claim 1,
The information recommendation apparatus, wherein the information on each community is the number of users belonging to each community. The information recommendation device according to claim 2,
The similarity calculation means includes: a number of users belonging to a community to which the predetermined user and the other users commonly belong; and an average of the number of users belonging to each of all communities to which the predetermined user belongs. An information recommendation device that calculates the first index value using a relative ratio. The information recommendation device according to claim 2 or 3,
The unexpectedness calculating means includes the number of users belonging to a community to which the predetermined user does not belong and a user to which the other user belongs. The information recommendation apparatus calculates the second index value by using a relative ratio between the predetermined user and the number of users belonging to a community to which the other user belongs in common. The information recommendation device according to any one of claims 1 to 4,
Further, for the first index value and the second index value, comprising a user desire storage means for receiving and storing an input about the predetermined user's desire,
The information recommendation device, wherein the recommendation means selects another user using an index value desired by the predetermined user obtained by the user desire storage means. An information recommendation method for recommending information held by other users to a predetermined user belonging to a community formed on a network,
Storing information about each community formed on the network;
Using the information on each community, based on the characteristics of the community to which the predetermined user and the other users belong in common and the characteristics of all the communities to which the predetermined user belongs, Calculating the first index value indicating the degree to which the other users belonging in common in one or more communities handle information having affinity for the predetermined user;
Of the users belonging to the community to which the predetermined user and the other user belong in common using the information on each community, the user belonging to the community to which the predetermined user does not belong and the other user belongs Based on the characteristics and characteristics of the community to which the predetermined user and the other user belong in common, the other user who belongs to the predetermined user and one or more communities in common Calculate a second index value that represents the degree of information that is unexpected for the user,
Based on the result of calculating the first index value and the second index value for each user, the other user who recommends information to the predetermined user is selected.
An information recommendation method characterized by that. The information recommendation method according to claim 6,
The information recommendation method, wherein the information on each community is the number of users belonging to each community. The information recommendation method according to claim 7,
In the calculation of the first index value, the number of users belonging to the community to which the predetermined user and the other users belong in common, and the number of users belonging to each of all the communities to which the predetermined user belongs are calculated. An information recommendation method, wherein the first index value is calculated using a relative ratio with an average. The information recommendation method according to claim 7 or 8,
In the calculation of the second index value, among the users belonging to the community to which the predetermined user and the other user belong, the predetermined user does not belong and also belongs to the community to which the other user belongs. The information recommendation is characterized in that the second index value is calculated using a relative ratio between the number of users and the number of users belonging to a community to which the predetermined user and the other users belong in common. Method. An information recommendation method according to any one of claims 6 to 9,
Further, the first index value and the second index value are received and stored with respect to the predetermined user's wish,
An information recommendation method, wherein, when selecting the other user, the other user is selected using the stored index value desired by the predetermined user. An information recommendation program for causing a computer to execute an information recommendation process for recommending information held by another user to a predetermined user belonging to a community formed on a network,
Storing information about each community formed on the network;
Using the information on each community, based on the characteristics of the community to which the predetermined user and the other users belong in common and the characteristics of all the communities to which the predetermined user belongs, A process of calculating a first index value representing a degree to which the other users belonging in common in one or more communities handle information having affinity for the predetermined user;
Of the users belonging to the community to which the predetermined user and the other user belong in common using the information on each community, the user belonging to the community to which the predetermined user does not belong and the other user belongs Based on the characteristics and characteristics of the community to which the predetermined user and the other user belong in common, the other user who belongs to the predetermined user and one or more communities in common A process of calculating a second index value representing a degree of bringing in information in a field unexpected for the user;
A process of selecting the other user who recommends information to the predetermined user based on a result of calculating the first index value and the second index value for each user;
Information recommendation program to execute An information recommendation program according to claim 11,
The information recommendation program characterized in that the information on each community is the number of users belonging to each community. An information recommendation program according to claim 12,
In the calculation process of the first index value, the number of users belonging to the community to which the predetermined user and the other user belong in common, and the number of users belonging to each of all the communities to which the predetermined user belongs An information recommendation program for calculating the first index value by using a relative ratio between the average and the average. The information recommendation program according to claim 12, wherein
In the calculation process of the second index value, among the users belonging to the community to which the predetermined user and the other user belong, the community to which the predetermined user does not belong and the other user belongs The second index value is calculated using a relative ratio between the number of users belonging and the number of users belonging to a community to which the predetermined user and the other users belong in common. Recommendation program. The information recommendation program according to any one of claims 11 to 14,
Further, for the first index value and the second index value, a process of receiving and storing an input about the predetermined user's wish is executed,
An information recommendation program for selecting another user using the stored index value desired by the predetermined user in the process of selecting the other user.

Images