JP5883744B2 - User attribute estimation device, user attribute estimation method, and program - Google Patents

Description

  The present invention relates to a technique for estimating a user attribute, and more particularly to a technique for estimating an attribute of a user who does not disclose a user attribute in a social network.

  As a first conventional technique for estimating a user attribute, there is a user attribute estimation device that estimates an unknown user attribute from a given geographic attribute by learning and using a relationship between the geographic attribute and the user attribute (for example, , See Patent Document 1).

  In addition, as a second conventional technique, there is a user attribute estimation device that estimates an unknown user attribute by defining proximity on a social network and propagating an attribute from a close user (see, for example, Non-Patent Document 1). ).

JP2011-238169A

Alan Mislove, Bimal Viswanath, Krishna P. Gummadi, Peter Druschel, "You are who you know: inferring user profiles in online social networks". In Proceedings of the third ACM international conference on Web search and data mining, pages 251-260, 2010.

  By using a method similar to that of the first conventional technique, it is possible to estimate user attributes based on feature quantities such as words used by the user, geographic information, and time information. For example, a user who uses the word “final exam” has a high probability of being a student. However, user attributes cannot be estimated for inactive users who do not have such information.

  The second prior art responds to the problem by giving an attribute of a nearby user on a social network, or when using the first prior art, it can be estimated by using this to propagate an inactive user. It is possible to estimate the user attribute. However, since the attributes are propagated from users who have no direct relationship on the social network, for example, users such as friends of friends, it propagates to attributes that the user on the propagation side does not have, that is, noise There has been a problem of propagation.

  The present invention has been made in view of the above points, and by solving a user attribute by a propagation method more suitable for a social network, it solves the problem of propagation including noise and is unknown with high accuracy. An object of the present invention is to provide a technique that makes it possible to estimate user attributes.

In order to solve the above-described problem, the present invention provides a user based on social network information stored in a social network storage unit and user attribute sets included in the social network stored in a user attribute storage unit. A user attribute estimation device for estimating an attribute set of a user whose attributes are unknown,
User cost calculation means for calculating a user cost that is lower as the attribute is shared with other users who are related on the social network and higher as the user does not share the input user that is assumed to have a certain attribute. ,
User attribute estimation means for estimating a user attribute set that minimizes the total cost in the social network based on the user cost calculated by the user cost calculation means for each user included in the social network. It is comprised as a user attribute estimation apparatus characterized by the above.

The user cost calculation means is, for example,
Based on social network information stored in the social network storage unit, a related user set extracting means for extracting a related user set related to the input user;
An inter-user cost calculation unit that calculates an inter-user cost between the input user and the related user extracted by the related user set extraction unit for each related user of the related user set;
The user cost can be calculated by adding the inter-user costs calculated by the all related user cost calculation means to the related user set.

The all related user cost calculation means is, for example,
Based on the social network information stored in the social network storage unit, the degree of similarity between users between the input user and the related user extracted by the related user set extraction unit is determined for each related user set. A similarity calculation means for calculating between users,
Based on the user attribute stored in the user attribute storage unit, the inter-user attribute distance between the input user and the related user extracted by the related user set extraction means is set as each related user of the related user set. and inter-user attribute distance calculation means to be de San attached to,
An inter-user cost calculating means for calculating the inter-user cost based on the inter-user similarity calculated by the inter-user similarity calculating means and the inter-user attribute distance calculated by the inter-user attribute distance calculating means; It can comprise by providing.

  The user attribute estimation device may further include an output unit that outputs an attribute value based on the result calculated by the user attribute estimation unit.

  Moreover, this invention can also be comprised as a user attribute estimation method which the said user attribute estimation apparatus performs. The present invention can also be configured as a program for causing a computer to function as each means in the user attribute estimation apparatus.

  According to the present invention, an attribute is propagated only between related users on a social network, and it is recursively repeated, for example, for all users, resulting in noise of unrelated users on the social network. Since attributes can be propagated while reducing the influence of attributes, it is possible to solve the problem of propagating to attributes that the user on the propagation side does not have and to estimate unknown user attributes with high accuracy It becomes.

It is a block diagram of the user attribute estimation apparatus 10 in embodiment of this invention. It is an example of the flowchart of operation | movement of the user attribute estimation part 12 in embodiment of this invention. It is a flowchart of operation | movement of the user cost calculation part 14 in embodiment of this invention. It is an example of the information stored in the social network memory | storage part 15 in embodiment of this invention. It is an example of the information stored in the user attribute memory | storage part 16 in embodiment of this invention.

  Embodiments of the present invention will be described below with reference to the drawings. The embodiment described below is only an example, and the embodiment to which the present invention is applied is not limited to the following embodiment.

(Outline of the embodiment)
First, an outline of the present embodiment will be described. In the present embodiment, modeling is performed such that each user can easily share attributes only with related users on the social network, and from the user to the related user so as to minimize the total cost of the entire given social network. By propagating the attribute, the attribute of the user whose user attribute is unknown is estimated. That is, intuitively, it is modeling in which recursive propagation is performed to a user B related to a certain user A and further to a user C related to the user B. However, in that case, instead of maximum likelihood estimation that propagates recursively and considers only the attribute with the highest probability, Bayesian estimation is performed to optimize the combination of all possible attributes. .

  Here, the cost of a certain user having a certain attribute is referred to as a user cost, and is assumed to be lower as the attribute is shared with a related user on the social network and higher as not shared. Moreover, the cost in the whole given social network is called total cost, and it is assumed that the user cost is totaled for all users.

(Configuration of User Attribute Estimation Device 10)
FIG. 1 shows a block diagram of a user attribute estimation device 10 for estimating a user attribute whose user attribute is unknown based on a given social network and a user attribute set included therein.

  As shown in the figure, the user attribute estimation device 10 includes an input unit 11, a user attribute estimation unit 12, a user cost calculation unit 14, a social network storage unit 15, a related user set extraction unit 17, an all related user cost calculation unit. 18, an inter-user similarity calculation unit 19, a user attribute storage unit 16, an inter-user attribute distance calculation unit 20, an inter-user cost calculation unit 21, and an output unit 13.

  The user attribute estimation device 10 according to the present embodiment includes a computer including a storage device (memory, hard disk, etc.) serving as each storage unit, an input unit 11, a user attribute estimation unit 12, a user cost calculation unit 14, and a social network storage. Unit 15, related user set extraction unit 17, all related user cost calculation unit 18, user similarity calculation unit 19, user attribute storage unit 16, user attribute distance calculation unit 20, user cost calculation unit 21, output unit This can be realized by executing a program corresponding to 13 processes. More specifically, in accordance with a program, data to be calculated is read from a memory or the like, an operation is performed by the CPU, and the process is executed while repeating the operation of storing in the memory or the like. The program may be stored in a storage medium such as a portable memory and distributed, installed on the computer, or downloaded from a server on a network and installed on the computer. Further, either or all of the social network storage unit 15 and the user attribute storage unit 16 may be provided in an external device that is accessible from the computer via a network without being provided in a computer that performs learning and inference processing.

  Below, each structure and operation | movement of the user attribute estimation apparatus 10 are demonstrated in detail.

<Input unit 11>
The input unit 11 receives information on a user set to be predicted as an input. Then, the information is converted into information U _in which becomes a user search key such as a user ID, and is output to the user attribute estimation unit 12.

The information to be accepted may be the social network S or the social network range S ′. In that case, the input user set U _in is obtained by extracting all user sets included in the social networks S and S ′ or user sets whose attributes are not disclosed.

<User attribute estimation unit 12>
The user attribute estimation unit 12 receives the input user set U _in from the input unit 11 as an input. The user set U _S attribute estimation result to the user the cost calculation unit 14 in the user u and attribute a and iteration t A _US, outputs _t, user cost C (u when the user u took the attribute a, a) receive. Here, the user set U _S is a set of all users included in a given social network S. Then, based on the received user cost C (u, a), the total cost in the modeling described in the outline of the embodiment is minimized, and the attribute A _Uin of the user set U _in is given to the output unit 13. Output. For minimizing the total cost, if the user's social network relationship is defined as directed, the variational Bayes method, sampling method, etc. are defined in addition to the above two methods. It is possible to use an effective programming method. Some methods do not perform the iteration.

  FIG. 2 is a flowchart showing the operation of the user attribute estimation unit 12 when Gibbs sampling is used when the attribute a included in the attribute set A is a discrete value as an example. 2 step numbers are indicated as appropriate.

First, for all user sets U _s in a given social network, user cost C (u, a) is calculated when attribute a included in all attribute sets A that each user can take (FIG. 2). Step 21). Specifically, the user cost C (u, a) is calculated by the user cost calculation unit 14, and the user attribute estimation unit 12 receives the user cost C (u, a) as the calculation result. Then, the user attribute estimation unit 12 calculates P (a | u) by normalizing C (u, a) as in the following equation (step 22 in FIG. 2).

このP(a|u)をもとにイテレーションtにおけるユーザuの属性推定結果a_u,tをサンプリングする（図２のステップ２３）。この際、最大のP(a|u)であるa_u,tを選ぶのではなく、P(a|u)に基づいてa_u,tを確率的に選ぶ点に注意されたい。これは、前述したように、とりうる全ての属性の可能性を考慮し、その組み合わせを最適化するようなベイズ推定を行うためである。P(a|u)はユーザコストを元に算出されているため、ユーザコストをより小さくするようなa_u,tが選ばれやすい。そのため、複数の可能性を考慮しながら、総コストを最適化することを可能としている。

Based on this P (a | u), the attribute estimation result a _{u, t} of the user u in the iteration _t is sampled (step 23 in FIG. 2). At this time, it should be noted that a _{u, t} is selected probabilistically based on P (a | u), rather than selecting a _{u, t} which is the maximum P (a | u). As described above, this is because Bayesian estimation is performed so as to optimize the combinations in consideration of all possible attributes. Since P (a | u) is calculated based on the user cost, a _{u, t} that makes the user cost smaller is likely to be selected. Therefore, it is possible to optimize the total cost while considering a plurality of possibilities.

Then, we sampled a _u, calculates the user cost _{C (u, a u, t} ) in _t, which is adding up for all user set U _s, the total cost _{C (U S, A U,} t) a Calculate (step 24 in FIG. 2). The process is repeated until the difference between the calculated total cost and the total cost of the iteration t−1 is equal to or less than the threshold value ε. If the difference is greater than or equal to ε, add 1 to the number of iterations.

If attribute a included in attribute set A is a continuous value, sampling is performed as follows. First, it is assumed that P (a | u) follows some distribution, and parameters of the distribution are estimated from an attribute a of the related user set U _rel described later and a user similarity sim (u, u ′) described later. For example, if it is assumed that the distribution is normal, the weighted average of the attribute a and the weighted variance with the user similarity sim (u, u ′) as weights are used as parameters. Sampling is performed from a normal distribution having an estimated parameter, and a box-Muller method or the like is used as a method for generating a random number from the normal distribution.

<User cost calculation unit 14>
The user cost calculation unit 14 receives the user u, the attribute a, and the user attribute estimation result A _{US, t} from the user attribute estimation unit 12 as inputs. Then, the inter-user cost C (u, u ′, a) with the user u ′ included in the related user set U _rel when the user u takes the attribute a, calculated by the all related user cost calculation unit 18. Based on this, the user cost C (u, a) when the user u takes the attribute a is calculated and output to the user attribute estimating unit 12 (steps 31 to 34 in FIG. 3).

C (u, a) is obtained by adding the inter-user cost C (u, u ′, a) with respect to the related user set U _rel as shown in the following equation.

図１に示すように、ユーザコスト算出部１４は、ソーシャルネットワーク記憶部１５、関連ユーザ集合抽出部１７、全関連ユーザ間コスト算出部１８、ユーザ属性記憶部１６を有する。

As shown in FIG. 1, the user cost calculation unit 14 includes a social network storage unit 15, a related user set extraction unit 17, a total related user cost calculation unit 18, and a user attribute storage unit 16.

  Below, each structure and operation | movement of the user cost calculation part 14 are demonstrated in detail. FIG. 3 is a flowchart showing the operation of the user cost calculation unit 14, and the corresponding step numbers in FIG.

<Social network storage unit 15>
An example of data stored in the social network storage unit 15 is shown in FIG.

In social network storage unit 15 stores social network information of the user set U _S. The social network information is information indicating which user the user has an association with and information indicating the strength of the association. The relationship may be directed information such as Twitter (registered trademark) follow or undirected information such as Facebook (registered trademark) friend registration. In addition, whether a reply is a reply to a post between two users in a certain period, even if the user explicitly does mutual registration, such as whether friends are registered between the two users. It may be extracted from. The information indicating the strength of association is information such as the frequency of conversation, the number of characters, the number of times a reply continues, the time taken for replying, the number of common friends, and the like.

  The social network storage unit 15 is not limited to a specific one as long as the information is stored and can be restored. For example, a database, a database stored in a specific area of a general-purpose storage device (memory or hard disk device) provided in advance, a Web server holding a Web page, a database server including a database, or the like.

<User attribute storage unit 16>
An example of data stored in the user attribute storage unit 16 is shown in FIG.

In the user attribute storage unit 16 stores the attribute set A of user set U _S. The attribute a is a set of a plurality of attribute names a _i and attribute values a _ij . Attribute names include gender, age, place of residence, hometown, native language, available language, occupation, place of work, educational information (university or high school, etc.), group name, religion, instructions Political parties, experience sports, preferences (favorite food, favorite music, favorite books, etc., or the degree may be included), hobbies, married / unmarried, family composition, etc. For example, the example shown in FIG. 5 indicates that the attribute value of the attribute whose attribute name is “profession” is “student” for the user whose ID is ID _a .

  Further, the attribute set A stored in the user attribute storage unit 16 may be a user attribute calculated using the same method as the first conventional technique, or a probability value having each user attribute.

The user attribute storage unit 16 may be adapted to store the profile set P of a user set U _S. Profile P includes a self-introduction sentence and an introduction sentence from a friend or the like. In this case, as preprocessing, the profile is subjected to morphological analysis, and the attribute set A is extracted by matching whether the attribute name is included in the morphological analysis result.

  The user attribute storage unit 16 is not limited to a specific one as long as the information is stored and can be restored. For example, a database, a database stored in a specific area of a general-purpose storage device (memory or hard disk device) provided in advance, a Web server holding a Web page, a database server including a database, or the like.

<Related User Set Extraction Unit 17>
The related user set extraction unit 17 receives the user u from the user attribute estimation unit 12 as an input. Then, the related user set U _rel of the user u is extracted from the social network stored in the social network storage unit 15 and output to the all related user cost calculation unit 18 (step 31 in FIG. 3).

Since the related user set U _rel does not depend on the iteration t, it may be calculated and stored in advance when a social network is given.

<All related user cost calculator 18>
All related user cost calculation unit 18 receives user u, attribute a, and attribute estimation result A _{US, t} from user attribute estimation unit 12 and user u related user set U _rel from related user set extraction unit 17 as input. . Then, the inter-user cost C (u, u ′, a) calculated by the inter-user cost calculation unit 21 and the user u ′ when the user u has the attribute a is output to the user cost calculation unit 14 ( Steps 32-33 in FIG.

  As shown in FIG. 1, all related inter-user cost calculation unit 18 includes social network storage unit 15, inter-user similarity calculation unit 19, user attribute storage unit 16, inter-user attribute distance calculation unit 20, and inter-user cost calculation unit. 21.

  Hereinafter, each configuration and operation of the all related user cost calculation unit 18 will be described in more detail.

<Inter-user similarity calculation unit 19>
The inter-user similarity calculation unit 19 receives the user u from the user attribute estimation unit 12, the user u ′ from the related user set extraction unit 17, and information indicating the strength of association from the social network storage unit 15 as input. Then, the user similarity sim (u, u ′) between the user u and the user u ′ is calculated and output to the inter-user cost calculation unit 21 (step 32 in FIG. 3).

For example, in the second prior art, sim (u, u ′) is obtained by the following equation. In the following equation, all or a part of the feature amount included in the information indicating the strength of association is obtained. An added method may be used. Also, normalization is performed so that sim (u, u ′) is added to the related user set U _rel to be 1.

ここで、distは下記の関数である。

Here, dist is the following function.

ただし、Kは、ユーザuとユーザu'の間の、共通の近隣ユーザを経由した経路集合であるとする。また、strengthは下記の関数である。

However, K is assumed to be a set of routes between the user u and the user u ′ via a common neighboring user. Strength is the following function.

ここで、X'_ijは下記のように求める。

Here, X ′ _ij is obtained as follows.

ただし、X_ijはユーザiとユーザjが過去に会話した回数であるとする。

Here, X _ij is the number of times user i and user j have talked in the past.

  Further, since the similarity sim (u, u ′) between users does not depend on the iteration t, it may be calculated and held first when a social network is given.

<User attribute distance calculation unit 20>
The inter-user attribute distance calculation unit 20 stores the attribute a and the attribute estimation result A _{US, t} from the user attribute estimation unit 12, the user u ′ from the related user set extraction unit 17, and the iteration stored in the user attribute storage unit 16. It accepts attribute a _{US, 0} of user set U _S at times 0 as input. Then, the attribute estimation result a _{u ', t} between the attribute a and the user u', the attribute distance diff (a, a _{u ', t} ) between the users _, and the attribute a and the user a's attribute a _{u, 0 in} the iteration count 0 The inter-attribute distance diff (a, a _{u, 0} ) is calculated and output to the inter-user cost calculation unit 21 (step 32 in FIG. 3). However, when the attribute estimation result a _{u ′, t} does not exist, the attribute a _{u ′, 0} is used. If the attribute a _u, 0 of the user u at the iteration count 0 does not exist, the inter-user attribute distance diff (a, a _{u, 0} ) is set to 0.

diff (a, a _′ ) is obtained by the following equation. That is, the number of shared attribute values. If the attribute value is a continuous value, it is determined by the distance such as the cosine similarity between the attributes and the reciprocal of the Pearson correlation. If the attribute value is a probability value, it is calculated by the sum of the probability values for the matched attributes. Desired.

＜ユーザ間コスト算出部２１＞
ユーザ間コスト算出部２１では、ユーザ間類似度算出部１９からsim(u,u')と、ユーザ間属性距離算出部２０からdiff(a,a_u',t)とdiff(a,a_u,0)を入力として受け取る。そして、ユーザ間コストC(u,u',a)を算出し、ユーザコスト算出部１４へと出力する（図３のステップ３３）。

<Inter-user cost calculation unit 21>
In the inter-user cost calculation unit 21, sim (u, u ′) from the inter-user similarity calculation unit 19 and diff (a, a _{u ′, t} ) and diff (a, a _u ) from the inter-user attribute distance calculation unit 20. _{, 0} ) as input. Then, the inter-user cost C (u, u ′, a) is calculated and output to the user cost calculation unit 14 (step 33 in FIG. 3).

  C (u, u ′, a) is obtained by the following equation.

ここで、αは重みパラメタであり、関連ユーザu'との属性距離とユーザuの初期値との属性距離のどちらをどの程度重視するかのパラメタである。

Here, α is a weight parameter, which is a parameter indicating how much importance is given to the attribute distance between the related user u ′ and the initial value of the user u.

<Output unit 13>
The output unit 13 _{receives the} attribute a _Uin of the user set U _in from the user attribute estimation unit 12 as an input. Then, the attribute value a _ij for each attribute name a _i is output.

  Here, output is a concept including display on a display, printing on a printer, sound output, transmission to an external device, and the like. The output unit may or may not include an output device such as a display or a speaker. The output unit can be realized by driver software of an output device, driver software of an output device, an output device, or the like.

(Concrete example)
Hereinafter, the processing of the present embodiment will be described using a specific example.

  Here, a specific example under the following conditions will be described.

・ Input is input user u ・ Attribute name a _i includes occupation and residence ・ Attribute value a _{occupation j} includes students and office workers ・ Attribute value a _{residence j} includes Tokyo and Kanagawa Included-The user attribute storage unit 16 stores user attributes calculated using the same method as the first prior art-In the user attribute estimation unit 12 that uses Gibbs sampling for the total cost minimization method, First, the attributes a _{u, t} possessed by the user u are sampled. For this purpose, the user cost C (u, a) is obtained through the user cost calculation unit 14 for all the attributes a that the user u can take. For example, C (u, a _{occupation: student, residence: Tokyo} ) is 2, C (u, a _{occupation: student, residence: Kanagawa} ) is 3, C (u, a _{occupation: office worker, residence: Tokyo)} ) Is 6, and C (u, a _{occupation: office worker, residence: Kanagawa} ) is 1, P (u, a _{occupation: student, residence: Tokyo} ) is (1/2) / (1/2 + 1 /3+1/6+1/1)=0.25, similarly, P (u, a _{occupation: student, residence: Kanagawa} ) is 0.167, P (u, a _{occupation: office worker, residence: Tokyo} ) is 0.083, P (u, a _{occupation: company employee, residence: Kanagawa} ) can be calculated as 0.5. Based on this probability, for example, a _profession as an attribute with the user _{u: company employee, residence:} When _Kanagawa has been sampling, attribute estimation result of the user u in iteration _t a _u, t to a _{Occupation: Employee ， Resident: Kanagawa} is substituted. This sampling is repeated for all user sets U _S , and the total cost C (U _S , A _{US is} obtained by adding the cost C (u, a _{occupation: company employee, residence: Kanagawa} ) in the sampled attributes for all users. _{, t} ). The above steps are iterated and repeated until the difference from the total cost C (U _S , A _{US, t-1} ) in the previous iteration is equal to or less than ε.

When the attribute value is a continuous value, for example, the degree of preference for IT, for example, the related user is u ₁ , u ₂ , u ₃ , sim (u, u ₁ ) = 0.6, a _u1 = 0.8 , Sim (u, u ₂ ) = 0.3, a _u1 = 0.6, sim (u, u ₃ ) = 0.1, a _u1 = 0.4, this weighted average is (0.6 × 0.8 + 0.3 × 0.6 + 0.1 × 0.4) /(0.6+0.3+0.1)=0.6, square of weighted variance is (0.6 × 0.8 ^ 2 + 0.3 × 0.6 ^ 2 + 0.1 × 0.4 ^ 2) / (0.6 + 0.3 + 0.1) −0.6 ^ 2 = 0.148 And can be calculated. Then, sampling is performed by generating random numbers using the Box-Muller method from the normal distribution with these two parameters.

An example in which user u, attribute a _{occupation: student, and residence: Tokyo} are input in the user cost calculation unit 14 will be described. First, the related user set extraction unit 17 obtains a related user set U _rel from the social network storage unit 15. Then, the inter-user cost C (u, u ′, a _{occupation: student, residence: Tokyo} ) is obtained for all users u ′ included in the related user set U _rel through the all related user cost calculation unit 18. By adding this to the related user set U _rel , the user cost C (u, a _{occupation: student, residence: Tokyo} ) is calculated and output to the user attribute estimation unit 12.

An example in which user u, user u ′, attribute a _{occupation: student, and residence: Tokyo} are input in all related user cost calculation unit 18 will be described. First, the inter-user similarity calculation unit 19 calculates the inter-user similarity between the user u and the user u ′ by adding all the conversations between common friends divided by the number of hops. For example, the common friends of the user u and the user u ′ are u _a and u _b , the user u and the user u ′ are 5 times, the user u and the user u _a are 3 times, and the user u ′ and the user u _a are 1 If the user u and the user u _b have conversations 4 times and the user u ′ and the user u _b have conversations 6 times, 5+ (3 + 1) / 2 + (4 + 6) / 2 = 12. By normalizing this, for example, it is assumed that it becomes 0.6. At the same time, in the attribute distance calculation unit 20 between users, attribute attribute diff (a, a _{u ', t} ) between attributes a _{occupation: student, residence: Tokyo} and user u' attribute estimation results a _{u ', t} The inter-user attribute distance diff (a, a _{u, 0} ) of the attribute a _u, 0 of the user u at the attribute a and the iteration count 0 is obtained. For example, if the attribute estimation result a _{u ', t of} the user u' is a _{occupation: company employee and residence: Kanagawa} , the attribute a and the attribute estimation result a _{u ', t} are attribute values of two attribute names. Diff (a, a _{u ', t} ) is 2 because Similarly, it is assumed that diff (a, a _{u, 0} ) is obtained as 1. Next, the inter-user cost calculation unit 21 calculates C (u, u ′, a _{occupation: student, residence: Tokyo} ). If α is 0.8, it can be calculated as 0.6 × 2 + 0.8 × 1 = 2.

  Note that the attribute value is a probability value, for example, user u's occupation: student probability is 0.4, occupation: office worker probability is 0.6, user u 'occupation: student probability is 0.2, occupation: office worker probability. In the case of 0.8, the attribute distance between users is obtained by taking the reciprocal of these two cosine similarity and Pearson correlation.

  The present invention is not limited to the above-described embodiments, and various modifications and applications are possible within the scope of the claims.

DESCRIPTION OF SYMBOLS 10 User attribute estimation apparatus 11 Input part 12 User attribute estimation part 13 Output part 14 User cost calculation part 15 Social network memory | storage part 16 User attribute memory | storage part 17 Related user set extraction part 18 All related user cost calculation part 19 Inter-user similarity Calculation unit 20 User-to-user attribute distance calculation unit 21 User-to-user cost calculation unit

Claims (7)

In order to estimate the attribute set of a user whose user attribute is unknown based on the social network information stored in the social network storage unit and the user attribute set included in the social network stored in the user attribute storage unit The user attribute estimation device of
User cost calculation means for calculating a user cost that is lower as the attribute is shared with other users who are related on the social network and higher as the user does not share the input user that is assumed to have a certain attribute. ,
User attribute estimation means for estimating a user attribute set that minimizes the total cost in the social network based on the user cost calculated by the user cost calculation means for each user included in the social network;
A user attribute estimation device comprising: The user cost calculation means is
Based on social network information stored in the social network storage unit, a related user set extracting means for extracting a related user set related to the input user;
An inter-user cost calculation unit that calculates an inter-user cost between the input user and the related user extracted by the related user set extraction unit for each related user of the related user set;
The user attribute estimation device according to claim 1, wherein the user cost is calculated by adding the inter-user cost calculated by the all-related user cost calculation unit to the related user set. The cost calculation means for all related users is
Based on the social network information stored in the social network storage unit, the degree of similarity between users between the input user and the related user extracted by the related user set extraction unit is determined for each related user set. A similarity calculation means for calculating between users,
Based on the user attribute stored in the user attribute storage unit, the inter-user attribute distance between the input user and the related user extracted by the related user set extraction means is set as each related user of the related user set. and inter-user attribute distance calculation means to be de San attached to,
An inter-user cost calculating unit that calculates the inter-user cost based on the inter-user similarity calculated by the inter-user similarity calculating unit and the inter-user attribute distance calculated by the inter-user attribute distance calculating unit;
The user attribute estimation apparatus according to claim 2, comprising: In order to estimate the attribute set of a user whose user attribute is unknown based on the social network information stored in the social network storage unit and the user attribute set included in the social network stored in the user attribute storage unit A user attribute estimation method executed by the user attribute estimation device of
A user cost calculation step for calculating a user cost that is lower as the attribute is shared with other users who are related on the social network and higher as the user does not share the input user assuming that the attribute has an attribute; ,
A user attribute estimation step for estimating a user attribute set that minimizes the total cost in the social network based on the user cost calculated by the user cost calculation step for each user included in the social network;
A user attribute estimation method comprising: The user cost calculating step includes:
A related user set extraction step for extracting a related user set related to the input user based on social network information stored in the social network storage unit;
A cost calculation step between all related users for calculating, for each related user in the related user set, an inter-user cost between the input user and the related user extracted by the related user set extraction step;
The user attribute estimation method according to claim 4, further comprising: calculating the user cost by adding the inter-user cost calculated by the all related user cost calculation step to the related user set. . The all related user cost calculation step includes:
Based on the social network information stored in the social network storage unit, the degree of similarity between users between the input user and the related user extracted by the related user set extraction step is determined for each association of the related user set. A user-to-user similarity calculation step for calculating a user;
Based on the user attribute stored in the user attribute storage unit, the inter-user attribute distance between the input user and the related user extracted by the related user set extraction step is set as each related user of the related user set. and inter-user attribute distance calculation step to be de San attached to,
An inter-user cost calculation step for calculating the inter-user cost based on the inter-user similarity calculated by the inter-user similarity calculation step and the inter-user attribute distance calculated by the inter-user attribute distance calculation step;
The user attribute estimation method according to claim 5, further comprising:   The program for functioning a computer as each means in the user attribute estimation apparatus of any one of Claims 1 thru | or 3.

Images