JP5882187B2 - User attribute estimator construction device, method, user attribute estimation device, and program - Google Patents

Description

  The present invention relates to a user attribute estimator construction device, method, user attribute estimation device, and program, and in particular, a user attribute estimator construction device, method, user attribute estimation device, and program for estimating user attribute information of a user. About.

  A microblog (also called a miniblog) is a medium for posting the user's own status and communicating with others using short documents. There is no need to write a long document and it is possible to post immediately. In addition, as smartphones have become popular, they are posted anytime and anywhere, so it is easy to see the live information that is happening around you.

  Because live information can be obtained in real time, there is a need for companies to analyze their posts about their company and their products on microblogs and use them for marketing. In order to use it for marketing, it is necessary to correctly analyze what kind of person the microblogging user is, not to mention the contents of posting.

  In the conventional technique, as in Non-Patent Document 1, it is necessary to prepare learning data necessary for constructing an estimator by human observation. For example, if a user attribute estimator called gender is to be constructed, it is necessary to collect a certain number of microblog users who are considered to be male and microblog users who are considered to be female by human observation. This operation is called annotation. Annotation is a very time-consuming task, so automation is required.

Kazufumi Ikeda, Hajime Hattori, Kazunori Matsumoto, Tomohiro Ono, Teruo Higashino, “Twitter profile estimation method for market analysis”, Multimedia, Distributed, Collaboration and Mobile (DICOM2011) Symposium, July 2011, p .1308-1315

  However, if the profile is a free-form document, it is up to the microblog user whether or not specific user attributes are described. Furthermore, even if the user attribute is described, it is difficult to extract it correctly. For example, when searching for a profile document that includes “male” as a character string in order to find a male microblogging user, a microblogging user that includes “male voice actor” or the like in the profile document is also extracted. There are many cases that should be excluded, and it takes time to exclude them on a rule basis.

  Therefore, as described above, there is no effective method for automatically performing annotation regardless of human observation or complicated rule setting. In addition, since microblogs are posted with short documents, if the user attributes are estimated from only the posted content, there is a problem that the estimation accuracy is lowered due to insufficient information necessary for estimation depending on the user's posting situation.

The present invention has been made to solve the above-described problem, and a user attribute estimator construction device, method, and the like that can construct a user attribute estimator for accurately estimating user attribute information of a microblog user. And to provide a program.
Moreover, it aims at providing the user attribute estimation apparatus and program which can estimate the user attribute information of the user of a microblog accurately.

  In order to achieve the above object, a user attribute estimator construction device according to the present invention provides, for each of a plurality of users of a microblog site, a plurality of microblog documents created by the user obtained from the microblog site. An external site extraction unit that extracts a URL of an external site different from the microblog site used by the user from a set of microblog documents, and a user from which the URL is extracted by the external site extraction unit The profile collection means for collecting the user's profile information from the external site of the URL extracted for the user, and for each of the users whose profile information has been collected by the profile collection means, from the user's profile information , Said Yu User attribute extraction means for extracting user attribute information indicating attributes relating to the profile, and features extracted from the user's microblog document set for each of the users from which the user attribute information has been extracted by the user attribute extraction means User attributes for constructing a user attribute estimator for estimating the user attribute information of the user who created the microblog document set based on the amount and the user attribute information of the user extracted by the user attribute extraction means Estimator construction means.

  A user attribute estimator construction method according to the present invention is a user attribute estimator construction method in a user attribute estimator construction device including external site extraction means, profile collection means, user attribute extraction means, and user attribute estimator construction means. The external site extraction means uses the user from a microblog document set composed of a plurality of microblog documents created by the user obtained from the microblog site for each of a plurality of users of the microblog site. Extracting a URL of an external site different from the microblogging site, and for each of the users from which the URL has been extracted by the external site extraction means by the profile collection means, URL external site Collecting the profile information of the user, and, for each of the users whose profile information has been collected by the profile collection means by the user attribute extraction means, from the user profile information, attributes related to the user profile. Extracting the user attribute information shown, and the user attribute estimator construction unit extracting the user attribute information from the user's microblog document set for each of the users whose user attribute information has been extracted by the user attribute extraction unit Constructing a user attribute estimator for estimating the user attribute information of the user who created the microblog document set based on the amount and the user attribute information of the user extracted by the user attribute extraction means; including.

  According to the user attribute estimator construction device and the user attribute estimator construction method according to the present invention, a plurality of users created from the microblog site by the external site extraction unit for each of the plurality of users of the microblog site. The URL of the external site different from the microblog site used by the user is extracted from the microblog document set including the microblog documents.

  Then, for each of the users whose URLs are extracted by the external site extracting means, the profile information of the users is collected from the external site of the URL extracted for the user by the profile collecting means.

  For each user whose profile information has been collected by the profile collection means, user attribute information indicating attributes relating to the user's profile is extracted by the user attribute extraction means.

  Then, for each user whose user attribute information has been extracted by the user attribute extraction unit by the user attribute estimator construction unit, the feature amount extracted from the user's microblog document set, and the user extracted by the user attribute extraction unit The user attribute estimator for estimating the user attribute information of the user who created the microblog document set is constructed based on the user attribute information.

  In this way, the URL of the external site used by the user is extracted from the set of microblog documents created by the user, and the user attribute information is extracted by collecting the user's profile document from the external site of the URL. User attribute estimation for accurately estimating user attribute information of microblog users by constructing a user attribute estimator based on the feature amount extracted from the microblog document set and user attribute information A vessel can be constructed.

  The external site may be at least one of a social networking service site, a blog site, a file sharing site, and an electronic bulletin board site.

  In addition, the external site extraction unit extracts, for each of the plurality of users, the URL of the external site used by the user from the microblog document set created by the user and extracts the URL. The user ID in the external site included in the URL is extracted, and the profile collection unit extracts the user for each of the users from which the URL and the ID are extracted by the external site extraction unit. The user's profile information may be collected from the external site of the URL using the extracted user ID.

Further, the user attribute estimator construction means, for each of the users for which the user attribute information has been extracted by the user attribute extraction means, a feature amount extracted from a profile document in the user's microblog document set, The user attribute estimator constructed based on the user attribute information of the user extracted by the user attribute extraction means, and the user for each of the users from which the user attribute information has been extracted by the user attribute extraction means At least one of the user attribute estimators constructed based on the feature amount extracted from the microblog post document in the set of microblog documents and the user attribute information of the user extracted by the user attribute extraction means Can be built.

  The user attribute estimation device according to the present invention includes a microblog document collection unit that collects the set of microblog documents created by the estimation target user obtained from the microblog site, and any one of claims 1 to 3. For each of a plurality of user attribute estimators constructed by the user attribute estimator construction device according to claim 1, a feature amount extracted from the set of microblog documents collected by the microblog document collection unit; Based on the user attribute estimator, the user attribute information of the estimation target user is estimated, and used to construct the user attribute estimator for each of the plurality of user attribute estimators. Based on the weights determined according to the number of types of feature quantities extracted from the microblog document set of feature quantities, Integrating the user attribute information estimated for each of the serial plurality of user attributes estimator is configured to include a estimating means for estimating the user attribute information of the estimation target user.

  Further, the user attribute estimation device according to the present invention includes a microblog document collection unit that collects the microblog document set created by the estimation target user obtained from the microblog site, and the microblog document collection unit. The feature amount extracted from the profile document of the collected microblog document set, and the user attribute estimator constructed by using the profile document by the user attribute estimator construction device according to claim 4. , Estimating the user attribute information of the estimation target user, the feature amount extracted from the microblog post document in the set of microblog documents collected by the microblog document collection means, and the user attribute estimator Using the microblog post document by the construction device The user attribute estimator is used to estimate the user attribute information of the estimation target user and to construct the user attribute estimator for each of the user attribute estimators. The user attribute information estimated for each of the user attribute estimators based on the weights determined according to the number of types of feature quantities extracted from the profile document or microblog post document of the quantity And estimation means for estimating the user attribute information of the estimation target user.

  According to the user attribute estimation apparatus according to the present invention, the microblog document collection unit collects the microblog document set created by the estimation target user and obtained from the microblog site.

  Then, for each of the plurality of user attribute estimators constructed by the above-described user attribute estimator construction device, the feature amount extracted from the microblog document collection collected by the microblog document collection means, and the user attribute estimation And the user attribute information of the estimation target user based on the device, and for each of the plurality of user attribute estimators, the microblog document among the feature values used to construct the user attribute estimator User attribute information estimated for each of a plurality of user attribute estimators is integrated based on weights determined according to the number of types of feature values extracted from the set, and the user attributes of the user to be estimated Estimate information.

  In this way, the user attribute is collected based on the feature amount extracted from the microblog document set collected by the user to be estimated and extracted from the microblog document set and a plurality of user attribute estimators built in advance. For each estimator, the user attribute information is estimated, and the estimated user attribute information is calculated based on weights obtained in advance according to the number of types of feature values used for constructing the user attribute estimator. By integrating and estimating the user attribute information of the estimation target user, it is possible to accurately estimate the user attribute of the user of the microblog.

  The program according to the present invention is a program for causing a computer to function as each unit of the user attribute estimator construction device according to the present invention or the user attribute estimation device according to the present invention.

  As described above, according to the user attribute estimator construction device, the user attribute estimator construction method, and the program of the present invention, from the set of microblog documents created by the user, the external site that the user uses Extract URL, collect user profile document from external site of the URL, extract user attribute information, and estimate user attribute based on feature quantity extracted from microblog document set and user attribute information By constructing a device, it is possible to construct a user attribute estimator for accurately estimating user attribute information of a microblog user.

  Further, according to the user attribute estimation device and the program of the present invention, a microblog document set created by a user to be estimated is collected, a feature amount extracted from the microblog document set, and a plurality of prebuilt Based on the user attribute estimator, the user attribute information is estimated for each user attribute estimator, and the weight determined in advance according to the number of types of feature values used to construct the user attribute estimator is used. Based on this, the estimated user attribute information is integrated, and the user attribute information of the estimation target user is estimated, whereby the user attribute of the microblog user can be accurately estimated. .

It is explanatory drawing which shows the example of a user account tied | linking at the time of using an external site. It is a block diagram which shows the example of 1 structure of the microblog user attribute estimator construction apparatus which concerns on embodiment of this invention. It is a block diagram which shows the example of 1 structure of the microblog user attribute estimation apparatus which concerns on embodiment of this invention. It is a flowchart which shows the content of the construction process routine in the microblog user attribute estimator construction apparatus which concerns on embodiment of this invention. It is a flowchart which shows the content of the microblog user attribute estimation process routine in the microblog user attribute estimation apparatus which concerns on embodiment of this invention. It is the figure which showed the user attribute estimation example which concerns on embodiment of this invention.

<Overview>
First, an outline of an embodiment of the present invention will be described.

  FIG. 1 shows an example of user account linking using an external site. Microblog profiles do not have profile items such as age or gender, only a free-form profile document. When words that indicate age or gender are extracted from the profile document using regular expressions, words such as 3 years old and niece are extracted, but these indicate attributes of children and pets, and attributes of users It does not indicate. Conventionally, such confirmation is performed by human observation, and collecting learning data for constructing an estimator is a laborious operation.

  In the present embodiment, the profile of the external site can be acquired by extracting the user ID of the external site from the URL information included in the profile of the microblog. Thereby, even if there is no user attribute information in the profile of the microblog or it is difficult to extract, the user attribute can be automatically given by associating the user account with the external site. In the example of FIG. 1, the user attribute that the user is 25 years old and a woman can be associated.

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

<System configuration of microblog user attribute estimator construction device>
FIG. 2 is a block diagram showing the microblog user attribute estimator construction device 10 according to the embodiment of the present invention. The microblog user attribute estimator construction device 10 is composed of a computer including a CPU, a RAM, and a ROM storing a program for executing a construction processing routine described later. It is configured.

  As shown in FIG. 2, the microblog user attribute estimator construction device 10 according to the present embodiment includes a construction input unit 1, a construction calculation unit 2, and a storage unit 3.

  The construction input unit 1 accepts a plurality of sets of microblog documents composed of a plurality of input documents.

  A microblog document indicates a collection of microblog posting documents and profile information described in a microblog site created by a microblog user. The microblog document set indicates a set of microblog documents generated by the microblog user, and the microblog document set is input for each microblog user.

  Profile information indicates all items that a user can set to express himself / herself in a microblog or an external site, such as a profile document, URL, gender, age, icon image, and the like. The profile document refers to a user's self-introduction document expressed in a free description format written on the microblog site by the user. The microblog post document indicates a document posted to the microblog site by the user.

  An external site is anything other than a microblog site, and profile information can be acquired and any user attribute that is an estimation target is included in profile information. For example, the external site includes a site as social media, and includes at least one of a social networking service site, a blog site, a file sharing site, and an electronic bulletin board site.

  The construction calculation unit 2 includes a microblog document database 20, a construction profile document extraction unit 21, a construction posted document extraction unit 22, an external site user attribute collection unit 23, and a microblog user attribute estimator construction unit 24. It consists of and.

  The microblog document database 20 stores a plurality of microblog document sets input by the construction input unit 1.

  For each microblog document set stored in the microblog document database 20, the construction profile document extraction unit 21 extracts a profile document from the microblog document set.

  For each of the microblog document sets stored in the microblog document database 20, the construction posted document extraction unit 22 extracts a microblog post document from the microblog document set.

  The external site user attribute collection unit 23 includes an external site ID extraction unit 230, an external site user profile collection unit 232, and an external site user attribute extraction unit 234.

  The external site ID extraction unit 230 extracts, for each of a plurality of users, the URL of the external site used by the user from the profile information in the user's microblog document set, and the user included in the URL Extract the ID. The URL of the external site is extracted from the URL description field if it exists in the profile information, but if it does not exist, it may be extracted from the profile document in the free description format using a regular expression or the like. The user ID of the external site is extracted from the URL using a regular expression or the like. The external site ID extraction unit 230 is an example of an external site extraction unit.

  The external site user profile collection unit 232 accesses the external site 231 via the Internet 15 for each user whose URL and user ID are extracted by the external site ID extraction unit 230 based on the URL extracted for the user. To do. Then, the user's profile information is collected from the external site 231 using the extracted user ID. If an API (Application Programming Interface) exists in the external site 231, the profile information of the external site user may be acquired using the API. If the API does not exist, the profile information of the external site user is acquired from the HTML document by using a technique such as scraping that extracts only a necessary part of the data on the website. The external site user profile collection unit 232 is an example of profile collection means.

The external site user attribute extraction unit 234 determines, for each user whose URL is extracted by the external site ID extraction unit 230, an attribute related to the user profile from the profile information of the user collected by the external site user profile collection unit 232. The user attribute information shown is extracted.
For example, when the gender estimator is constructed in the microblog user attribute estimator construction unit 24 described later, the external site user attribute extraction unit 234 extracts the user attributes of male and female. It is determined in advance what kind of user attribute to construct the estimator, and the external site user attribute extraction unit 234 holds the extraction rule. The extraction rule is described by a set of user attributes, character strings, regular expressions, and the like. For example, a user attribute of male and a regular expression of (male [sex] | ♂) are used. For example, when an age estimator for estimating the age is constructed, user attributes may be extracted indirectly, such as extracting birthday information and calculating the age from the current time. The external site user attribute extraction unit 234 is an example of a user attribute extraction unit.

  The microblog user attribute estimator construction unit 24, for each user whose user attribute information is extracted by the external site user attribute extraction unit 234, features extracted from the user's microblog document set, and external site user attributes Based on the user attribute information of the user extracted by the extraction unit 234, a plurality of user attribute estimators for estimating the user attribute information are constructed.

  The microblog user attribute estimator construction unit 24 includes a profile document estimator construction unit 240 and a posted document estimator construction unit 241. Note that the microblog user attribute estimator construction unit 24 is an example of a user attribute estimator construction unit.

  The profile document estimator construction unit 240 extracts, for each user whose user attribute information has been extracted by the external site user attribute extraction unit 234, the profile document extracted from the user's microblog document set by the construction profile document extraction unit 21. A profile document estimator comprising estimators for each user attribute item to be estimated based on the feature amount extracted for the user and the user attribute information of the user extracted by the external site user attribute extraction unit 234 To do. Each estimator of the profile document estimator inputs a feature amount extracted from the profile document and outputs a probability value in which the user attribute item is each class.

  Specifically, first, the profile document is divided into appropriate character string units by morphological analysis. From these character strings, characteristic character strings (hereinafter referred to as feature values) predetermined for each class of user attribute items to be estimated are extracted. For example, in the gender attribute, character strings that are characteristic for men (I, bride, etc.) and character strings that are characteristic for women (I, husband, etc.) are extracted as feature amounts. For each extracted feature quantity, the appearance frequency of the feature quantity is counted, and the values of TF-IDF (TF: Term Frequency, IDF: Inverse Document Frequency), Akaike information quantity standard, and chi-square value are the feature quantity. It is calculated as a value to have. Next, a certain number of feature values are extracted in order from the highest value of the feature value. The above feature amount extraction is performed for each user. Then, based on the feature quantity extracted for each user, the user attribute information of the item to be estimated, extracted for each user, as a teacher signal, a naive Bayes classifier (Reference (Andrew McCallum, Kamal Nigam, “A comparison of event models for Naive Bayes text classification ”, In proceedings of AAAI WORKSHOP ON LEARNING FOR TEXT CATEGORIZATION, 1998, p.41-48), or support vector machine (references (Corinna Cortes, Vladimir Vapnik,“ Support- (See Vector Networks, Machine Learning, 1995, p.237-297)). In the present embodiment, a naive Bayes classifier is constructed as an estimator. The estimator is constructed for each item of the estimated user attribute information.

  The posted document estimator construction unit 241 extracts, for each user whose user attribute information has been extracted by the external site user attribute extraction unit 234, a microblog extracted from the user's microblog document set by the construction posted document extraction unit 22. A posted document estimator comprising an estimator for each item of user attribute to be estimated based on the feature amount extracted for the posted document and the user attribute information of the user extracted by the external site user attribute extraction unit 234 Build up. Each estimator of the posted document estimator receives the feature amount extracted from the microblog posted document and outputs a probability value that the user attribute item is each class.

  Specifically, similar to the profile document estimator construction unit 240, after performing morphological analysis, the feature amount is extracted, the value of the feature amount is calculated for each feature amount, and a certain number of feature amounts are extracted. To do. Based on a certain number of feature quantities extracted for each user, an estimator is constructed using the user attribute information of the item to be estimated extracted for each user as a teacher signal. The estimator is constructed for each item of user attribute information to be estimated. In the following description, an example in which each of the profile document estimator and the posted document estimator is a gender estimator will be described.

  The storage unit 3 includes a profile document estimator storage unit 30 and a posted document estimator storage unit 31.

  The profile document estimator storage unit 30 stores the profile document estimator constructed by the profile document estimator construction unit 240.

  The posted document estimator storage unit 31 stores the posted document estimator constructed by the posted document estimator construction unit 241.

<System configuration of microblog user attribute estimation device>
FIG. 3 is a block diagram showing a microblog user attribute estimation apparatus 100 according to the embodiment of the present invention. This microblog user attribute estimation device 100 is composed of a computer including a CPU, a RAM, and a ROM storing a program for executing a microblog user attribute estimation processing routine described later. It is configured as shown.

  As shown in FIG. 3, the microblog user attribute estimation apparatus 100 according to the present embodiment includes an input unit 4, a storage unit 5, a calculation unit 6, and an output unit 7.

  The input unit 4 receives an input of a microblog user ID of a user of a microblog that is a user attribute estimation target.

  The storage unit 5 includes a profile document estimator storage unit 50 and a posted document estimator storage unit 51.

  The profile document estimator storage unit 50 stores the profile document estimator constructed by the microblog user attribute estimator construction apparatus 10.

  The posted document estimator storage unit 51 stores a posted document estimator constructed by the microblog user attribute estimator construction device 10.

  The calculation unit 6 includes a microblog document collection unit 60, a profile document extraction unit 62, a profile document estimation unit 63, a posted document extraction unit 64, a posted document estimation unit 65, and an estimation result integration unit 66. ing.

  Based on the microblog user ID input by the input unit 4, the microblog document collection unit 60 accesses the microblog site 61 corresponding to the microblog user ID via the Internet 15. Then, the microblog document included in the microblog site 61 is collected.

  The profile document extraction unit 62 extracts a profile document from the microblog document collected by the microblog document collection unit 60.

The profile document estimation unit 63 estimates user attribute information based on the profile document extracted by the profile document extraction unit 62 and the profile document estimator stored in the profile document estimator storage unit 50.
Specifically, it estimates the user attribute by profile document estimator relative profile document u _p in micro blog document. Estimation results of profile document estimator _p p _{(u p)} is outputted as a probability value for each class of items in the user attribute estimating. For example, when the user attribute item is gender, male 0.60 and female 0.40 are output for each class.

More specifically, profile document estimation unit 63, the profile document u _p extracted by the profile document extraction section 62 first performs morphological analysis. Then, based on the analysis result obtained by the morphological analysis, the profile document u _p, counting the number of occurrences of each feature quantity, and extracts the appearance feature amount. Here, the extracted feature quantities are all or a part of the feature quantities used when the profile document estimator is constructed by the microblog user attribute estimator construction apparatus 10.

Next, the profile document estimation unit 63 calculates a probability value for each class of user attribute items based on the extracted feature amount and the profile document estimator stored in the profile document estimator storage unit 50. and outputs as the estimated result _{p p (u p).}

  In addition, the profile document estimation unit 63 estimates user attributes and records information related to the feature amount when counting up the appearance of the feature amount. The information to be recorded conforms to the information used in the estimation result integration unit 66 described later. For example, when the number of feature quantities appears, the number of types of feature quantities that appear, the values of the feature quantities (TF-IDF, Akaike information criterion, chi-square value, etc.), or feature quantities are arranged in descending order Record the ranking. The profile document estimation unit 63 is an example of an estimation unit.

  The posted document extraction unit 64 extracts a set of microblog posted documents from the microblog document collected by the microblog document collection unit 60.

The posted document estimation unit 65 estimates user attribute information based on the microblog posted document set extracted by the posted document extraction unit 64 and the posted document estimator stored in the posted document estimator storage unit 51. . Specifically, the user attribute is estimated using a posted document estimator for the microblog posted document u _{t in the} microblog document. The estimation result p _t (u _t ) by the posted document estimator is output as a probability value for each class of user attribute items to be estimated.
More specifically, post document estimation unit 65, like the profile document estimation unit 63, after the morphological analysis, the micro blog post document u _t, counting the number of occurrences of each feature quantity, the appearance feature amount Extract. Here, the extracted feature quantities are all or a part of the feature quantities used when the posted document estimator is constructed by the microblog user attribute estimator construction apparatus 10.

Next, the posted document estimation unit 65 outputs a probability value as an estimation result p _t (u _t ) for each class of the user attribute item based on the extracted feature amount and the posted document estimator.

  Similarly to the profile document estimation unit 63, the posted document estimation unit 65 estimates user attributes and records information about the feature amount when counting up the appearance of the feature amount. The posted document estimation unit 65 is an example of an estimation unit.

The estimation result integration unit 66 out of the feature quantities used for constructing the profile document estimator, the number of types of feature quantities appearing in the profile document, and the features used for constructing the posted document estimator. The reliability as a weight for each of the profile document estimator and the posted document estimator is calculated according to the number of types of feature quantities that appear in the microblog posted document. Then, based on the calculated reliability, the user attribute information estimated by each of the profile document estimator and the posted document estimator is integrated to estimate the user attribute information of the estimation target user.
Specifically, the estimation result integration unit 66 calculates the reliability as a weight for each estimation result based on the information about the feature amount recorded together with the estimation by the profile document estimation unit 63 and the estimation by the posted document estimation unit 65. To do.
Then, based on the estimation result estimated by the profile document estimation unit 63, the estimation result estimated by the posted document estimation unit 65, and the calculated reliability, for each class of user attribute items, the final The estimation result p (u) is output. The calculation formula of the estimation result p (u) is shown in the following formula (1).

_{Here, r} p of (u) _{reliability, r} t estimation results _p p of the profile document estimator is output _{(u p)} (u) is estimated result of post document estimator outputs _{p t (u} t) It is reliability. r _p (u) and r _t (u) satisfy r _p (u) + r _t (u) = 1. In the case of using the number of types of used feature quantities as a reliability calculation method, r _p (u) and r _t (u) are calculated as in the following equations (2) to (5).

_Here, i p _{(u p)} is the total feature quantity genus has a profile document estimator | _{V p} | among the profile document _u number of types of feature quantities included in the _{p v p} (u _p) appearance Indicates the amount of information selected. i _t (u _t ) is the same for the microblog post document. α is a constant for adding 0 so as not to take a logarithm, and α = 1 or the like is used. As the number of types of feature quantities that appear increases, the number of clues necessary for estimation increases, and the estimation accuracy increases. The greater the number of feature types, the smaller the selected information amount. When the amount of selected information is small, the reliability becomes large because the denominator of reliability becomes small.
In the calculation of the selection information amount, if the ratio of the feature amount of the user to the entire feature amount can be calculated, the number of appearance of the feature amount, the value of the feature amount (TF-IDF, Akaike information amount standard and chi-square value), rankings when the feature amounts are arranged in descending order of the values, and the like may be used.

  The estimation result integration unit 66 outputs the class having the largest value as the estimation result of the user attribute based on the estimation result of each class calculated according to the above equations (1) to (5). The estimation result integration unit 66 is an example of an estimation unit.

  The output unit 7 outputs the estimation result calculated by the estimation result integration unit 66.

<Operation of Microblog User Attribute Estimator Construction Device>
Next, the operation of the microblog user attribute estimator construction device 10 according to the present embodiment will be described. First, when a microblog document set, which is a set of microblog documents for each of a plurality of microblog users obtained from the microblog site 61, is input to the microblog user attribute estimator construction device 10, a microblog user attribute estimator is constructed. The apparatus 10 stores a plurality of input microblog document sets in the microblog document database 20. Then, the construction processing routine shown in FIG. 4 is executed by the microblog user attribute estimator construction device 10.

  First, in step S100, a target user is set. Next, in step S102, a set of microblog documents created by the target user set in step S100 is extracted.

  In step S104, the external site ID extraction unit 230 extracts the URL of the external site used by the user from the profile information in the target user's microblog document set extracted in step S102. The user ID of the external site included in the URL is extracted.

  In step S106, it is determined whether or not the user ID has been extracted in step S104. If the user ID can be extracted, the process proceeds to step S108. If the user ID cannot be extracted, the process returns to step S100 to set the next target user.

  In step S108, the external site user profile collection unit 232 accesses the external site 231 via the Internet 15 based on the URL extracted in step S104. Then, the profile information of the target user is collected from the external site 231 using the user ID extracted in step S104.

  Next, in step S110, the external site user attribute extraction unit 234 extracts user attribute information indicating attributes related to the profile of the target user from the profile information of the target user collected in step S108.

  In step S112, it is determined whether user attribute information has been extracted in step S110. If the user attribute information can be extracted, the process proceeds to step S114. If the user attribute information cannot be extracted, the process returns to step S100 to set the next target user.

  In step S114, the construction profile document extraction unit 21 extracts a profile document from the set of microblog documents of the target user extracted in step S102.

  Next, in step S116, feature amounts are extracted from the profile document extracted in step S114, and a set of the extracted feature amounts and user attribute information extracted in step S110 is stored in a memory (not shown). Record.

  Next, in step S118, the construction posted document extraction unit 22 extracts a microblog posted document set from the target user's microblog document set.

  Next, in step S120, feature amounts are extracted from the microblog post document set extracted in step S118, and a set of the user attribute information extracted in step S114 and the extracted feature amounts is stored in a memory (illustrated). (Omitted).

  Next, in step S122, it is determined whether or not the processing in steps S100 to S120 has been executed for all user microblog document sets stored in the microblog document database 20. If there is a set of microblog documents of users who have not executed the processes in steps S100 to S120, the process returns to step S100, and the user is set as a target user. On the other hand, when the above steps S100 to S120 are executed for the microblog document set of all users, the process proceeds to step S124.

  In step S124, the profile document estimator construction unit 240 creates a profile based on the combination of the feature amount extracted from the profile document and the user attribute information for each of all users recorded in step S116. Build a document estimator.

  In step S126, the profile document estimator constructed in step S124 is stored in the profile document estimator storage unit 30.

  In step S128, the posted document estimator construction unit 241 uses the feature amount extracted from the microblog posted document and the user attribute information for each of all users recorded in step S120. Build a posted document estimator.

  Next, in step S130, the posted document estimator constructed in step S128 is stored in the posted document estimator storage unit 31, and the construction processing routine is terminated.

<Operation of Microblog User Attribute Estimation Device>
Next, the operation of the microblog user attribute estimation apparatus 100 according to the present embodiment will be described. First, the profile document estimator stored in the profile document estimator storage unit 30 of the microblog user attribute estimator construction device 10 and the posted document estimator stored in the posted document estimator storage unit 31 are micro-blogs. When input to the blog user attribute estimation device 100, the profile document estimator is stored in the profile document estimator storage unit 50, and the posted document estimator is stored in the posted document estimator storage unit 51. When the microblog user ID of the user to be estimated is input to the microblog user attribute estimation device 100, the microblog user attribute estimation device 100 executes the microblog user attribute estimation processing routine shown in FIG.

  First, in step S200, the microblog user ID of the user who is the user attribute estimation target input by the input unit 4 is received.

  Next, in step S202, the microblog document collection unit 60 accesses the microblog site 61 via the Internet 15, and from the microblog site 61 using the microblog user ID input in step S200. A set of microblog documents corresponding to the microblog user ID is collected.

  Next, in step S204, the profile document extraction unit 62 extracts profile documents from the microblog document set collected in step S202.

  In step S206, the profile document estimation unit 63 extracts the feature amount from the profile document extracted in step S204, and the extracted feature amount and the profile document stored in the profile document estimator storage unit 50. A user attribute is estimated based on the estimator.

  In step S208, the posted document extracting unit 64 extracts a microblog posted document set from the microblog document set collected in step S202.

  In step S210, the posted document estimation unit 65 extracts feature amounts from the set of microblog posted documents extracted in step S208, and the extracted feature amounts are stored in the posted document estimator storage unit 51. The user attribute is estimated based on the posted document estimator.

  In step S212, the profile document estimation unit 63 and the posted document estimation unit 65 record information on the feature values used for the estimations in step S206 and step S210 in a memory (not shown).

  In step S214, the estimation result integration unit 66 calculates the reliability as the weight for each estimator according to the information (number of types of feature amounts) related to the feature amount recorded in step S212. Based on the calculated reliability, the user attributes estimated by each of the profile document estimator and the posted document estimator are integrated to estimate the user attribute of the estimation target user.

  In step S216, the output unit 7 outputs the estimation result estimated in step S214, and the microblog user attribute estimation processing routine ends.

<Example of attribute estimation for microblogging users>
Next, an example of user attribute estimation by the microblog user attribute estimation apparatus according to the present embodiment is shown.

  FIG. 6 shows an example of user attribute estimation by the microblog user attribute estimation apparatus. Profile documents contain words that are specific to a certain age, such as “children”, and words used by young people to some extent, such as emoticons, so 0.25 for teens, 0.45 for 20s, and 30s The profile document estimator estimates that the probability is 0.20 and the probability of ages 40 and above is 0.10. On the other hand, since the microblog post document does not contain emoticons, it is estimated that the age is a little older than the profile document, 0.05 for teens, 0.40 for 20s, 0.45, 40s and above is a probability of 0.10. Only with the estimation result of the microblog post document, the user is estimated to be in their thirties, and an erroneous estimation result different from that in the twenties that is the actual age is presented. Further, the estimation results of the profile document estimator and the posted document estimator are different, and both of the estimation results have a probability of 0.45, and it is not possible to adopt the age with the maximum probability.

  In the present embodiment, the profile document estimator and the posting are calculated by considering how many feature quantities are used to output the estimation result with respect to the number of prepared feature quantities, and calculating the reliability as a result. The estimation results of the document estimator are weighted and integrated. As a result, it is estimated that the probability is 0.1000 for the teens, 0.4125 for the 20s, 0.3875 for the 30s, 0.1000 for the 40s and above, and the estimated 20s is the actual age. Is consistent with Estimating more reliable by emphasizing the results of the estimator estimated with confidence by integrating the estimated results of the profile document estimator and the estimated results of the posted document according to the feature quantities used. The result has been obtained.

  As described above, according to the microblog user attribute estimator construction device according to the present embodiment, from the set of microblog documents created by each user of the microblog site, the external site that the user uses Extracting URL and user ID, collecting user profile information from an external site of the URL, extracting user attribute information, and extracting feature value from profile document or microblog post document in microblog document set And a user attribute estimator for estimating the user attribute information based on the extracted user attribute information, so that the profile field of the microblog is a document in a free description format. User attributes for obtaining user attribute information of users even if they are unclear It is possible to build a Joki.

  The profile document estimator learns “what user” has “what profile document”, and the posted document estimator has “what user” has “what posted document”. As a result of learning, “what kind of user” is learned well by using an external site, the user attributes can be estimated with high accuracy.

  Further, according to the microblog user attribute estimation device according to the present embodiment, a microblog document set created by the estimation target user is collected and extracted from the profile document and the microblog post document of the microblog document set. And user attribute information for each user attribute estimator based on a plurality of user attribute estimators constructed in advance and types of feature values extracted from profile documents and microblog posting documents Based on the weights determined according to the number, the estimated user attribute information is integrated to estimate the user attribute information of the user to be estimated, so that the microblog profile field is a free-form document. Even user attributes can be estimated.

  Also, by discovering users who use both microblogging and external sites, linking user accounts, and diverting profile information from external sites, automatic annotation that does not require human observation is possible. Become.

  In addition, since automatic annotation that does not require human observation is possible, a user attribute estimator can be automatically constructed.

  In addition, it is possible to output a highly accurate estimation result by setting both the profile document and the microblog posting document as estimation targets and integrating both estimation results by weighting according to the reliability.

  Note that the present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the gist of the present invention.

  For example, the profile document estimator construction unit 240 and the posted document estimator construction unit 241 in the microblog user attribute estimator construction device, and the profile document estimation unit 63 and the posted document estimation unit 65 in the microblog user attribute estimation device perform morphological analysis. However, the present invention is not limited to this. For example, the document may be divided into character strings by other methods such as n-gram division. .

  Profile documents in the profile document estimator construction unit 240 and the posted document estimator construction unit 241 in the microblog user attribute estimator construction device, and the profile document estimation unit 63 and the posted document estimation unit 65 in the microblog user attribute estimation device. Each of the estimator and the posted document estimator has been described by way of example of a gender estimator for gender estimation, but is not limited thereto. Each of the profile document estimator and the posted document estimator may be composed of a plurality of estimators for a plurality of items of user attributes.

  Further, the microblog user attribute estimator construction unit 24 in the microblog user attribute estimator construction device has been described as an example of constructing the profile document estimator and the posted document estimator, but is not limited thereto. Only one of the profile document estimator and the posted document estimator may be constructed. In that case, in the microblog user attribute estimation device, the user attribute is estimated using either the profile document estimator or the posted document estimator.

  Further, the profile document estimator construction unit 240 and the posted document estimator construction unit 241 in the microblog user attribute estimator construction device, and the profile document estimation unit 63 and the posted document estimation unit 65 in the microblog user attribute estimation device include TF−. The case where the IDF, the Akaike information criterion, and the value of the chi-square value are calculated as values possessed by the feature amount has been described as an example. However, the present invention is not limited to this, and the value possessed by the feature amount is calculated by another method. You may do it.

  Further, the estimation result integration unit 66 in the microblog user attribute estimation apparatus has been described as an example in which the estimation result of the class having the largest value is output based on the estimation result calculated by each estimator. The class is not limited, and all classes whose estimation results have a value equal to or greater than a certain threshold may be output as estimation results. Alternatively, probability values may be output as estimation results instead of classes.

  Moreover, although the above-mentioned microblog user attribute estimator construction apparatus demonstrated the case where the profile document estimator memory | storage part 30 and the contribution document estimator memory | storage part 31 were provided, for example, the profile document estimator memory | storage part 30 and a contribution document An estimator storage unit 31 is provided in an external device of the microblog user attribute estimator construction device, and the microblog user attribute estimator construction device communicates with the external device using a communication means, thereby obtaining a profile document estimator storage unit. 30 and the posted document estimator storage unit 31 may be referred to.

  In addition, the above-described microblog user attribute estimation device also communicates with the profile document estimator storage unit 50 and the posted document estimator storage unit 51 provided in an external device by using a communication unit, thereby providing a profile document estimator. The storage unit 50 and the posted document estimator storage unit 51 may be referred to.

  In the above embodiment, the microblog user attribute estimator construction device and the microblog user attribute estimation device are described as separate devices. However, the microblog user attribute estimator construction device and the microblog are described. The user attribute estimation device may be configured as one device.

  Although this embodiment is particularly effective for microblogging, it can be applied to any site where a URL can be described in a profile, a profile document is in a free description format, and a document can be posted. It is.

In addition, as described in the above embodiment, the marketing tool that estimates the microblog user attribute and analyzes the posting of the microblog user, the product, service, information recommendation service according to the microblog user attribute, etc. It is also possible to use it.

  The microblog user attribute estimator construction device and the microblog user attribute estimation device described above have a computer system inside, but if the “computer system” uses a WWW system, a homepage is provided. It also includes the environment (or display environment).

  In the present specification, the embodiment has been described in which the program is installed in advance. However, the program can be provided by being stored in a computer-readable recording medium.

2 construction computing unit 6 computing unit 10 microblog user attribute estimator construction device 20 microblog document database 21 construction profile document extraction unit 22 construction posted document extraction unit 23 external site user attribute collection unit 24 microblog user attribute estimator Construction unit 30 Profile document estimator storage unit 31 Posted document estimator storage unit 50 Profile document estimator storage unit 51 Posted document estimator storage unit 60 Microblog document collection unit 62 Profile document extraction unit 63 Profile document estimation unit 64 Post document extraction Unit 65 posted document estimation unit 66 estimation result integration unit 100 microblog user attribute estimation device 230 external site ID extraction unit 232 external site user profile collection unit 234 external site user attribute extraction unit 240 profile document estimator construction unit 241 Written estimator constructing unit

Claims (8)

For each of a plurality of users of a microblog site, the microblog site used by the user from a microblog document set made up of a plurality of microblog documents created by the user obtained from the microblog site An external site extraction means for extracting URLs of different external sites;
Profile collection means for collecting profile information of the user from each external site of the URL extracted for the user for each of the users whose URLs have been extracted by the external site extraction means;
User attribute extraction means for extracting user attribute information indicating attributes relating to the user profile from the user profile information for each of the users whose profile information has been collected by the profile collection means;
For each user from which the user attribute information has been extracted by the user attribute extraction unit, a feature amount extracted from the user's microblog document set, and the user attribute information of the user extracted by the user attribute extraction unit, A user attribute estimator construction means for constructing a user attribute estimator for estimating the user attribute information of the user who created the microblog document set,
A user attribute estimator construction device including:   2. The user attribute estimator construction device according to claim 1, wherein the external site is at least one of a social networking service site, a blog site, a file sharing site, and an electronic bulletin board site. The external site extracting means extracts, for each of the plurality of users, the URL of the external site used by the user from the set of microblog documents created by the user, and the extracted URL The ID of the user at the external site included in
The profile collection unit uses the extracted user ID from the external site of the URL extracted for the user for each of the users from which the URL and the ID are extracted by the external site extraction unit. The user attribute estimator construction device according to claim 1, wherein the user profile information is collected. The user attribute estimator construction means includes a feature amount extracted from a profile document in the user's microblog document set for each of the users whose user attribute information has been extracted by the user attribute extraction means, and the user The user attribute estimator constructed based on the user attribute information of the user extracted by the attribute extraction means, and the user's micro for each of the users from which the user attribute information has been extracted by the user attribute extraction means Construct at least one of the user attribute estimators constructed based on the feature amount extracted from the microblog post document in the blog document set and the user attribute information of the user extracted by the user attribute extraction means The user attribute estimator according to any one of claims 1 to 3. Construction device. Microblog document collection means for collecting the microblog document set created by the estimation target user obtained from the microblog site;
The microblog document collected by the microblog document collection means for each of a plurality of user attribute estimators constructed by the user attribute estimator construction device according to any one of claims 1 to 3. The user attribute information of the estimation target user is estimated based on the feature amount extracted from the set and the user attribute estimator, and the user attribute estimation is performed for each of the plurality of user attribute estimators. For each of the plurality of user attribute estimators, based on the weights determined according to the number of types of feature values extracted from the microblog document set among the feature values used to construct a device. And estimating means for estimating the user attribute information of the estimation target user by integrating the user attribute information estimated in
A user attribute estimation device including: Microblog document collection means for collecting the microblog document set created by the estimation target user obtained from the microblog site;
The feature amount extracted from the profile document in the microblog document collection collected by the microblog document collection unit, and the user constructed using the profile document by the user attribute estimator construction device according to claim 4. Based on the attribute estimator, the user attribute information of the estimation target user is estimated, and the feature amount extracted from the microblog post document in the microblog document collection collected by the microblog document collection unit And based on the user attribute estimator constructed using the microblog post document by the user attribute estimator construction device, the user attribute information of the estimation target user is estimated, and each of the user attribute estimators For the feature used to construct the user attribute estimator The user attribute information estimated for each of the user attribute estimators is integrated based on weights determined according to the number of types of feature values extracted from the profile document or microblog post document. Estimating means for estimating the user attribute information of the estimation target user;
A user attribute estimation device including: A user attribute estimator construction method in a user attribute estimator construction device including external site extraction means, profile collection means, user attribute extraction means, and user attribute estimator construction means,
For each of a plurality of users of the microblog site by the external site extraction means, the user uses from a microblog document set consisting of a plurality of microblog documents created by the user obtained from the microblog site. Extracting a URL of an external site different from the microblogging site,
Collecting the user's profile information from the external site of the URL extracted for the user for each of the users whose URL has been extracted by the external site extraction means by the profile collection means;
Extracting user attribute information indicating an attribute related to the user's profile from the user's profile information for each of the users whose profile information has been collected by the profile collecting means by the user attribute extracting means;
Features extracted from the user's microblog document set for each user whose user attribute information has been extracted by the user attribute extraction unit by the user attribute estimator construction unit, and extracted by the user attribute extraction unit Building a user attribute estimator for estimating the user attribute information of the user who created the microblog document set based on the user attribute information of the user,
A user attribute estimator construction method including:   The program for functioning a computer as each means of the user attribute estimation device construction apparatus of any one of Claims 1-4, or the user attribute estimation apparatus of Claim 5 or 6.