JP6113560B2 - Social network information processing apparatus, processing method, and processing program - Google Patents

Description

  The present invention relates to an information processing apparatus that uses information acquired in a social network and an information processing method thereof.

  In recent years, social networks (SNS) have been widely used, and individuals and companies send and receive a lot of information on the SNS. Attempts have been made to utilize the enormous amount of data in some form of business. Here, individuals and corporations are collectively referred to as persons.

  Information on the SNS (hereinafter also referred to as “SNS information”) such as a profile on a person's SNS, text such as a comment sent by the person on the SNS, and a link of friendship on the person's SNS The character of the person appears and can be used to estimate the character of the person.

  For example, Patent Document 1 describes a service that supports an individual who wants to select a career such as a school advancement destination, a job destination, or a job change destination by using SNS information. The result of natural language analysis of information on a predetermined item and information in an arbitrary format input by an individual on the SNS is matched with recruitment information subjected to similar processing, and the degree of match is determined as a priority.

  Patent Document 2 describes a system that enables a person search with a specified type by tagging a person with a type using SNS information. This makes it possible to search for a desired type of person by searching.

  Patent Document 3 describes a technique for determining whether or not an individual can be specified from information obtained from an SNS, and using the determination result for protecting personal anonymity. Individual information is recorded in a database, and anonymity of the individual is calculated based on the information and information input by the individual from the terminal.

JP 2009-9468 A JP 2010-79435 A JP 2010-97336 A

  In order to use the SNS information for a service to an actual person, it may be necessary to specify an account of a person on the same SNS as the actual person. If a target person and an account on the SNS of the person are given in advance, a person on the same SNS as the actual person can be easily specified.

  However, if it is not given, it is not easy to determine the identity between an actual person and a person on the SNS, or to search the SNS for the same person as the actual person. The reason that makes it difficult is that an accurate and complete profile is not disclosed on the SNS, and that information that is different from the actual information is disclosed or that some information is not disclosed. There is. For example, the address or name may have changed if time has passed since account registration. In some cases, information such as hometown or school is not disclosed. In addition, information may not be input in an item that is arbitrarily input. In addition, information that is different from actual information may be disclosed to the SNS consciously. In addition, there is a case where information of the same item as the information given for the actual person does not exist on the SNS or the item name is different.

  An object of the present invention is to provide a technique that enables the identity determination between an actual person and a person on the SNS.

  A social network information processing apparatus according to an aspect of the present invention provides an account on a social network based on at least a part of information of target person information that is a plurality of items of person information and each item is given a value of the target person. The candidate person social information that is information on the social network related to the account owner of the candidate account extracted by narrowing down is acquired, and the person information based on the acquired candidate person social information and information on the social network An estimation unit that generates candidate person-corrected social information including values of a plurality of items that can be compared with person information based on a person model that provides values that can occur for at least some of the items, and the target person information And the candidate person correction social information about each candidate account By comparison, has a, a determination unit determining the identity of said target person account holder for said each candidate account.

  According to the present invention, it is possible to determine the identity between an actual person and a person on the SNS.

It is a block diagram of the social network information utilization system by this embodiment. 2 is a block diagram of the identity determination device 11. FIG. It is a block diagram of the caching system by a present Example. 2 is a block diagram of an evaluation system 36. FIG. 4 is a block diagram of an identity determination device 41. FIG. 5 is a flowchart showing processing of a learning unit 51. It is a figure which shows an example of the format of person information. It is a figure which shows an example of evaluation object person information. It is a flowchart which shows the process of the estimation part 52. FIG. It is a figure which shows an example of candidate person social information. It is a figure which shows an example of the candidate person correction social information which complemented candidate person social information. 5 is a flowchart showing the operation of a determination unit 53. 3 is a block diagram of a credit determination device 42. FIG. 7 is a flowchart showing processing of a learning unit 81. 5 is a flowchart showing processing of an estimation unit 82. It is a figure which shows an example of the data set produced | generated by the estimation part. It is a figure which shows an example of the data set with which accident occurrence probability SS of the evaluation information 94 was complemented. It is a figure which shows an example of the candidate person correction | amendment social information which complemented the candidate person social information in Example 2. FIG. It is a block diagram of the identity determination apparatus by Example 3.

  A schematic embodiment of the present invention will be described with reference to the drawings.

  On social networks (SNS), individuals and companies with accounts are connected to each other and exchange opinions. For this reason, various information related to individuals and companies exists on the SNS, which is information that can be used effectively for information processing such as services for individuals and companies, or services related to individuals and companies.

  The system according to the present embodiment relates to an information use target, information related to an information use target obtained from a social network related to the account, using an individual or a company that can have an account on the SNS as an information use target. This system is used for information processing.

  Here, individuals and organizations that are information utilization targets are collectively referred to as persons.

  On the social network, there are information (SNS information) such as profiles, friendships, content of remarks, etc. of a specific information utilization target person (target person) and many other persons.

  Here, the target person is specified by filling each item of person information including a plurality of items. Further, it is assumed that information on the target person (target person information) is given, but the account of the target person on the SNS is not specified. Therefore, in order to acquire information on the SNS regarding the target person, it is first necessary to specify an account on the SNS of the target person.

  The system of the present embodiment identifies the target person's account by determining the identity between the designated target person and the account owner on the SNS, acquires information on the SNS related to the account, It is what you use.

  FIG. 1 is a block diagram of a social network information utilization system according to this embodiment. Referring to FIG. 1, the social network information utilization system 10 includes an identity determination device 11 and an information utilization device 12.

  The identity determination device 11 is a device that identifies the account owner who is the same person as the target person from the SNS by determining the identity between the designated target person and the account owner on the SNS.

  The information utilization device 12 is a device that executes information processing using information obtained from the SNS related to the account owner identified as the same person as the target person by the identity determination device 11.

  There are many fields where information obtained from SNS can be used.

  For example, when a financial institution makes a loan to an individual, the credit that gives credit to the individual, such as how much the individual has a repayment capability and the probability of occurrence of an accident such as repayment is delayed In addition, it is conceivable to use information related to the individual obtained from the SNS.

  In addition, for example, it is conceivable to improve the marketing effect by using information related to an individual obtained from the SNS for marketing for an individual such as a product or a service.

  In addition, for example, in a job hunting or job change, it is possible to use information related to individuals obtained from the SNS to secure more suitable human resources or the desired type of human resources. .

  FIG. 2 is a block diagram of the identity determination device 11. Referring to FIG. 2, the identity determination device 11 includes a learning unit 21, an estimation unit 22, and a determination unit 23.

  The learning unit 21 holds in advance a person model that provides contents (or numerical values representing the contents (hereinafter also simply referred to as “values”) that may occur for at least some items of person information based on information on the SNS. Yes. The person model may be generated in advance by the learning unit 21 itself, or may be given in advance from the outside.

  First, the estimation unit 22 is a candidate person that is information on the SNS related to the account owner of the candidate account extracted by narrowing down the accounts on the SNS based on information of at least some items of the target person information. Get social information. It is preferable to narrow down the accounts so that the target person does not leak and the number of candidates is as small as possible. For example, if the name is included in the personal information, it may be possible to narrow down by using the name. This is because the name is information with little change over time and there is little duplication. Candidates can be narrowed down to only those with the same name as the target person. Other personal information items suitable for narrowing down accounts include email addresses, telephone numbers, and addresses.

  Subsequently, based on the acquired candidate person social information and the person model, the estimation unit 22 generates candidate person corrected social information including values of a plurality of items that can be compared with the person information. From the SNS, information that can be compared with the target person information is not necessarily acquired for all items of the person information. Therefore, the estimation unit 22 estimates information that cannot be acquired from the SNS by using a person model. As a result, for the target person, the given target person information can be compared with the information obtained from the SNS.

  The determination unit 23 determines the identity between the account owner and the target person for each candidate account by comparing the target person information with the candidate person correction social information for each candidate account. The identity determination result may be output, for example, in the form of an identity score indicating a high value when the possibility of being the same person is high.

  At this time, the determination unit 23 may determine the account having the highest identity with the target person among the candidate accounts by comparing the target person information with the candidate person correction social information. For example, the candidate account with the highest identity score may be estimated to be the same person as the target person.

  As described above, according to the present embodiment, it is possible to determine the identity between an actual target person to whom person information is given and a person on the SNS.

  In addition, when the learning unit 21 generates the person model in advance by itself, the learning unit 21 includes personal information regarding a plurality of learning target persons whose correspondence relationship with the account owner is known, and an account corresponding to the learning target person. The person model can be calculated by inputting information on the SNS related to the learning target account owner who is the owner as learning data and performing an operation on the learning data. For example, based on the learning data, performing an operation for formulating or quantifying the relationship between each item of the person information related to the learning target person and each item of information obtained from the SNS related to the learning target account owner Thus, a person model may be calculated.

  In addition, the person model provides information indicating a value that can occur for at least some items of person information and the occurrence probability of the value, and the candidate person correction social information includes at least some items, Information indicating the value given by the person model and the occurrence probability of the value may be included. In that case, the determination unit 23 calculates an identity score indicating the identity between the account owner and the target person for each candidate account based on the value indicated by the candidate person correction social information and the occurrence probability. Also good.

  Hereinafter, an example of a calculation using a person model will be described.

  Assume that no value is obtained from the SNS for one item of candidate person social information of a candidate person. In that case, by applying a person model to the candidate person, in relation to other information obtained from the SNS, a plurality of types of values and each value of the item of the candidate person social information whose value is not obtained The probability of occurrence is obtained. Candidate person-corrected social information can be obtained by applying information including a plurality of types of values obtained from the person model and the probability of occurrence of each value to an item whose value was not obtained from the SNS of the candidate person social information. . By considering this occurrence probability in calculating the identity score, a means for calculating a more appropriate identity score can be obtained.

  As an example, the candidate person social information acquired from the SNS by the estimation unit 22 includes the account owner's profile set in the candidate account, text input from the candidate account, and a link from the candidate account to a predetermined number of steps. A part or all of the related account connected to the candidate account and the text regarding the account owner of the candidate account input from the related account or another account may be included. Thereby, the various information which can be acquired from SNS can be utilized for determination of identity.

  In addition, if the information on the SNS related to the account owner of the candidate account exceeds a predetermined condition, the estimation unit 22 acquires a part of the information on the SNS related to the account owner by sampling, Candidate person social information may be used. Thereby, for example, when the information on the SNS regarding the account owner is enormous and the load is large, it is possible to reduce the amount of information while making it difficult for the information content to be biased by sampling. Here, sampling is to extract a part so that the characteristics of the population can be obtained. There are various methods for sampling, and a representative example is random sampling.

  In addition, the identity determination device 11 according to the present embodiment uses the target person information based on the candidate person social information or the candidate person corrected social information for the account determined by the determination unit 23 as having the highest identity with the target person. It is also possible to further include a person information correction unit (not shown) that generates correction target person information by correcting. According to this, since the person information (target person information) of the target person can be corrected to express the target person better, when the target person information is used in post-processing, the target person is more It can be reflected. For example, this is effective when the information using apparatus 12 performs information processing using not only the information obtained from the SNS related to the target person but also the personal information of the target person.

  Moreover, the determination part 23 will correct | amend by predetermined | prescribed correction | amendment processing, if the information obtained from SNS regarding the account owner of the account determined with the highest identity with an object person does not correspond with object person information. You may decide.

  As a specific example of the correction process, a process of matching the information obtained from the SNS regarding the account owner of the account determined to have the highest identity with the target person with the target person information can be considered. According to this, since the information obtained from the SNS related to the target person is corrected so as to coincide with the target person information, the information obtained from the SNS is more accurately obtained when the reliability of the target person information is high. Can be corrected to represent the target person. As a result, more accurate information can be used in post-processing using information obtained from the SNS.

  As another example of specific correction processing, items that do not match the target person information among the information obtained from the SNS regarding the account owner of the account determined to have the highest identity with the target person. A process of assuming that there is no value and replacing the value with only other information obtained from the SNS is also conceivable. According to this, among the information obtained from the SNS in relation to the target person, the item that does not match the target person information can be corrected to another value obtained from the SNS, so that the possibility of being inaccurate is higher than the others. For high items, the calculation can be performed again and corrected to other values. As a result, more accurate information can be used in post-processing using information obtained from the SNS.

  Hereinafter, more specific examples will be described.

  Example 1 illustrates a system that uses information about an individual obtained from an SNS for credit in caching.

  FIG. 3 is a block diagram of the caching system according to this embodiment. The cashing system has a credit system 31 and a lending system 32.

  The lending system 32 is a system that accepts a loan application from the user terminal 33 via the Internet 35 and provides a loan to the applicant if the applicant satisfies a predetermined condition. The loan system 32 uses the credit system 31 when determining whether the applicant satisfies a predetermined condition.

  Here, an example is shown in which an application for a loan is made via the Internet 35, but the embodiment is not limited to this, and personal information (evaluation target person information) of a predetermined item about the applicant is Any form can be used as long as it is obtained. Another example is application at a store.

  The credit system 31 receives the credit request by identifying the individual applicant from the lending system 32, performs credit processing on the applicant based on information (SNS information) obtained from the SNS, and determines the risk score or accident occurrence probability. It responds with the indicated credit information.

  The SNS information used by the credit system 31 may be information that can be acquired from the SNS 34 in real time, may be collected and accumulated information, or may be information of both. Here, both pieces of information are used.

  As shown in FIG. 3, the credit system 31 includes an evaluation system 36, a collection device 37, and an SNS information database (DB) 39.

  The collection device 37 collects SNS information from the SNS 34 and stores it in the SNS information DB 38. For example, the collection device 37 collects SNS information related to a predetermined individual and stores the SNS information in the SNS information DB 38 in association with the individual.

  FIG. 4 is a block diagram of the evaluation system 36. Referring to FIG. 4, the evaluation system 36 includes an identity determination device 41 and a credit determination device 42.

  The identity determination device 41 is a device corresponding to the identity determination device 11 in FIG. 1, and evaluates by determining the identity between the designated applicant (person to be evaluated) and the account owner on the SNS 34. The account owner having the highest identity with the target person is identified from the SNS 34.

  The credit determination device 42 is a device corresponding to the information use device 12 shown in FIG. 1 and is evaluated using information obtained from the SNS 34 related to the account owner specified by the identity determination device 41. Credit processing is executed for the target person.

  First, the configuration and operation of the identity determination device 41 will be described.

  FIG. 5 is a block diagram of the identity determination device 41. Referring to FIG. 5, the identity determination device 41 includes a learning unit 51, an estimation unit 52, and a determination unit 53.

  The learning unit 51 calculates a person model by performing learning processing using the given person model learning data. The person model learning data is data in which person information of the same person is associated with an account, and the person model receives information obtained from the SNS 34 in relation to an account of a person as an input, This is model data from which values (numerical values indicating contents) that can be compared with information are obtained. The learning process is a process of learning contents that can be included in each item and occurrence probability of each content based on information obtained from the SNS 34 related to the account. The content here includes not only the content or the value indicating the content itself but also information indicating the origin of the content on the SNS 34. For example, it is conceivable that the item “name” of the personal information is likely to have the same content as the “name” field included in the profile in the SNS 34. In addition, it is possible that the same information as the “Old School” column of the profile in the SNS 34 is likely to be included in the “Old School” item of the personal information. Although the possibility may be lower than that, it is considered that the same information as the “Old School” field of the friend's profile may be included in the “Occupational School” item of the personal information. Furthermore, if the number of friends from the same school is large, the possibility that the school will enter the “School” column will increase.

  FIG. 6 is a flowchart showing the processing of the learning unit 51. Referring to FIG. 6, the learning unit 51 first inputs human model learning data from the outside (step 101). As an example, the person model learning data is data that specifies person information about a person and an account on the SNS 34 of the same person.

  The learning unit 51 acquires information on the SNS 34 related to the account included in the person model learning data, and each piece of information (person social information) obtained by dividing information from the SNS 34 based on the acquired information. Fill in the items. At this time, for items corresponding to items included in the person information among items of the person social information, the person information and the item name are matched (step 102). For example, there is an item “name” in the personal information, and if there is information in the item “name” in the information from the SNS 34, the item name is changed to “item” in the personal social information.

  In order to make the item names match, for example, the learning unit 51 holds in advance a rule table storing rules that associate the corresponding item names with each other, and matches the item names by referring to the rule table. You can do it. The rule table may be obtained by calculation of information processing for analyzing information on the SNS 34, or may be set manually.

  Subsequently, the learning unit 51 calculates a possible value and an occurrence probability for each item (step 103). The value calculated here and the occurrence probability become the data of the person model.

  For example, a value that appears in the person social information as the content of each item may be a content that can be entered as each item. The content that can be included in one item can be plural. Since the contents of a certain item are related to the contents of another item, the occurrence probability of the value of a certain item can be calculated based on the value of the other item. For example, consider a case where there are two items that have the same value with a high probability. When the value of one item is obtained and the value of the other item is not obtained, the probability of occurrence of the same value as the one item in the other item is high.

  In this way, by calculating the possible values of each item and their occurrence probabilities, it is possible to estimate the contents of items that cannot be obtained from the SNS 34 based on information obtained from other SNSs 34. Become.

  When the applicant (evaluation target person) applies for a loan from the user terminal 33 in FIG. 3 to the loan system 32 in the state where learning of the person model is finished, the evaluation target person is transferred from the loan system 32 to the evaluation system 36. An evaluation request for the person to be evaluated is sent together with the information.

  FIG. 7 is a diagram showing an example of a format of person information. Here, the personal information 61 includes items such as name, address, work place, and job type. In addition, for example, when applying for a loan, the applicant inputs information of his / her own contents in each item of personal information.

  FIG. 8 is a diagram illustrating an example of evaluation target person information. The evaluation target person information 62 is information in which the contents relating to the evaluation target person (applicant) or a numerical value indicating the contents is set in each item of the person information in FIG. In the example of FIG. 8, the value SN is set in the name field of the evaluation target person information 62, the value SA is set in the address field, the value SC is set in the work field, and the job type field is set. A value of SJ is set.

  FIG. 9 is a flowchart showing the processing of the estimation unit 52.

  In the evaluation system 36 that has received the evaluation request together with the evaluation target person information, the estimation unit 52 of the identity determination device 41 narrows down candidate accounts using some items of the evaluation target person information (step 201). According to the evaluation target person information 62 shown in FIG. 8, since the name of the evaluation target person is SN, the estimation unit 52 extracts an account whose owner name is SN from among the accounts on the SNS 34. A candidate account.

  Next, the estimation unit 52 matches the information acquired from the SNS 34 regarding the candidate account with the item name for the item having the item corresponding to the person information (step 203), and the content of the item from the SNS 34. Candidate person social information in which information to be acquired is set is generated.

  The range which the estimation part 52 acquires as candidate person social information is not specifically limited. For example, the candidate person social information includes the account owner's profile set in the candidate account, the text entered from the candidate account, and the account of a friend connected to the candidate account with a link from the candidate account to a predetermined number of steps. Contains some or all of the related account and text about the account owner of the candidate account entered from that related account or other accounts.

  In addition, there may be a large amount of candidate person social information related to the candidate account on the SNS 34. For example, the candidate account may have many friends on the SNS 34. It is also conceivable that the candidate account frequently speaks on the SNS 34.

  If the information on the SNS 34 related to the account owner of the candidate account exceeds a predetermined condition, the estimation unit 52 acquires a part of the information on the SNS 34 related to the account owner by sampling, and the candidate person Social information may be used.

  FIG. 10 is a diagram illustrating an example of the candidate person social information. FIG. 10 shows candidate person social information 71, 72, and 73 for three accounts A, B, and C. For each item of name, address, place of work, and job type, the personal information and the item name match.

  In each item, a value indicating the content acquired from the SNS 34 is set. For example, looking at the candidate person social information 71 of the candidate account A, a value of AN is set in the name item. The AN is equal to the SN of the name of the person to be evaluated. The item of the address of the candidate account is blank because the information entering there is not obtained from the SNS 34. The value of AC is set in the work field. A value of AJ is set in the item “job type”.

  In addition, as an item having no item corresponding to the person information, a value of AS is set for the item “school”. However, the item “major” is blank because the information contained therein was not obtained from the SNS 34.

  Further, as shown in FIG. 10, the candidate person social information may include information on the person's remark, friend, and remark of the friend.

  In addition, a value CN is set in the name field of the candidate person correction social information 76 of the candidate account C. This CN is similar to the name SN of the person to be evaluated. For example, it can be assumed that the kanji notation is similar to the hiragana notation.

  At the stage where step 203 is completed, candidate person social information as shown in FIG. 10 is generated.

  Next, the estimation part 52 estimates the value which enters the blank item of candidate person social information using a person model, and complements a value to the item (step 204). In that case, the estimation part 52 calculates the value put into a blank item using a person model based on the information obtained from SNS34. Specifically, the estimation unit 52 applies information obtained from the SNS 34 to the person model, and sets a value having the highest occurrence probability as a blank item value as a blank item.

  FIG. 11 is a diagram illustrating an example of candidate person correction social information complemented with candidate person social information. FIG. 11 shows candidate person-corrected social information 74, 75, and 76 that complement the candidate person social information 71, 72, and 73 of FIG. For example, in the candidate person correction social information 74 of the candidate account A, a value of estimated AA is set in the address field that was blank. The blank estimation may be performed for the entire candidate person social information, or may be performed only for an item having an item corresponding to the person information. In the example of FIG. 11, blank estimation is performed only for items with items corresponding to person information.

  At the stage where step 204 is completed, the candidate person correction social information can be compared with all items of the evaluation target person information.

  At that stage, the determination unit 53 compares the evaluation target person information with the candidate person correction social information of each candidate account, and calculates the identity score of each candidate account.

An example of the condition for calculating the identity score is shown below.
<Name> Match: 1.0, similarity: 0.5, mismatch: 0.0 / weight: 2.0
<Address> Match: 1.0, number difference: 0.8, mismatch: 0.0 / weight: 1.0
<Workplace> Match: 1.0, affiliated company: 0.6, mismatch: 0.0 / weight: 0.7
<Occupation> Match: 1.0, Similar: 0.5, Mismatch: 0.0 / Weight: 1.0
A calculation example based on the above calculation conditions is shown below.

  It is assumed that the candidate account A has the same name, the address does not match, the work place is an affiliated company, and the job type does not match. Then, the identity score Score_A of the candidate account A is 2.0 × 1.0 + 1.0 × 0.0 + 0.7 × 0.6 + 1.0 × 0.0 = 2.42.

  It is assumed that the candidate account B has the same name, does not match the address, does not match the work place, and matches the job type. Then, the identity score Score_B of the candidate account B is 2.0 × 1.0 + 1.0 × 0.0 + 0.7 × 0.0 + 1.0 × 1.0 = 3.00.

  It is assumed that the candidate account C has a similar name, a different address, a mismatched work place, and a matching job type. Then, the identity score Score_C of the candidate account C is 2.0 × 0.5 + 1.0 × 0.8 + 0.7 × 0.0 + 1.0 × 1.0 = 2.8.

  When the identity score is calculated for each candidate account in this way, next, the determination unit 53 determines a candidate account that is highly likely to be the same person as the evaluation target person, and determines the evaluation target person and the determined candidate account. The credit association device 42 is notified of the association information to be associated.

  FIG. 12 is a flowchart showing the operation of the determination unit 53.

  First, the determination unit 53 selects one candidate account (step 301). Next, the determination unit 53 compares the identity score of the candidate account with a predetermined threshold (step 302). If the identity score is greater than the threshold (Yes in Step 303), the determination unit 53 determines that the owner of the candidate account is highly identical to the evaluation target person (Step 304). On the other hand, if the identity score is not greater than the threshold (No in Step 303), the determination unit 53 determines that the owner of the candidate account is low in identity with the person to be evaluated (Step 305).

  Subsequently, the determination unit 53 determines whether or not the identity score has been compared with the threshold for all candidate accounts (step 306). If all are not completed, the determination unit 53 returns to Step 301. If all are completed, the determination unit 53 notifies the credit determination device 42 of association information that associates the evaluation target person with the candidate account determined to be highly identical (step 307).

  Here, the determination unit 53 determines that all the candidate accounts having the identity score exceeding the threshold value have high identity with the person to be evaluated, but the present invention is not limited to this. . As another example, a candidate account having the highest identity score may be estimated to be the same person as the person to be evaluated.

  Furthermore, the determination unit 53 performs a predetermined correction process if the information obtained from the SNS 34 related to the account owner of the account determined to have the highest identity with the evaluation target person does not match the evaluation target person information. You may decide to correct by. For example, the address of the account B shown in FIG. 10 is different from the address of the person to be evaluated. The determination unit 53 may correct this address.

  At that time, as a correction process, the determination unit 53 matches information obtained from the SNS 34 regarding the account owner of the account determined to have the highest identity with the evaluation target person with the evaluation target person information. May be.

  Alternatively, as the correction process, the determination unit 53 selects items that do not match the evaluation target person information from among the information obtained from the SNS 34 regarding the account owner of the account determined to have the highest identity with the evaluation target person. It is assumed that there is no value, and the value may be replaced with a value estimated only from other information obtained from the SNS 34. At that time, the person model described above can be used for estimation.

  Next, the configuration and operation of the credit determination device 42 will be described.

  FIG. 13 is a block diagram of the credit determination device 42. The credit determination device 42 includes a learning unit 81 and an estimation unit 82.

  The learning unit 81 receives the credit model learning data given from the outside, calculates the credit model learning data, and calculates a credit model that provides the accident risk score of the person to be evaluated.

  The credit model learning data includes learning target person information, learning target person social information, and whether or not the learning target person has an accident. The learning target person information is personal information related to a plurality of learning target persons whose presence or absence of an accident in a past transaction is known. The learning target person social information is information related to an account connected to the account on the SNS 34 of the learning target person up to a predetermined number of links. For example, if the learning target person social information includes up to a direct friend of the learning target person, the account connected to the learning target person through a link up to one level is the target. In addition, if the learning target person social information includes even the friend of the learning target person's friend, the account connected to the learning target person through links up to two levels is the target.

  The credit model is based on personal information of an arbitrary person and personal social information that is information on the SNS 34 within an account connected to the account on the social network of the person by a predetermined number of links. Data providing an outbreak risk score.

  The estimation unit 82 calculates an accident risk score based on the evaluation target person information, the evaluation target person social information, and the credit model. The evaluation target person information is the person information of the evaluation target person. The evaluation target person social information is information on the SNS 34 related to an account connected to the account on the SNS 34 of the evaluation target person up to a predetermined number of steps. The accident occurrence risk score is an index representing the accident occurrence probability in the person to be evaluated.

  The estimation unit 82 is given an association between the person to be evaluated and the account on the SNS 34 of the person who is estimated to be the same person from the identity determination device 41. Based on the information, Person social information can be acquired from SNS34.

  The evaluation target person social information includes, for example, a profile set in the account of the evaluation target person, a text input from the account, a related account connected to the account by a link from the account to a predetermined number of stages in SNS 34, It includes all the text entered from the related account and the text about the person to be evaluated entered from the related account or another account.

  In addition, when the related account of the account of the evaluation target person is greater than or equal to a predetermined threshold, the estimation unit 82 uses the evaluation target person social information based on the account sampled from the related account below the threshold.

  FIG. 14 is a flowchart showing the processing of the learning unit 81. Referring to FIG. 14, first, external credit model learning data is input (step 401). Next, the learning unit 81 performs an operation on the credit model learning data, and calculates a credit model that provides an accident occurrence risk score indicating the accident occurrence probability of the person to be evaluated (step 402).

  FIG. 15 is a flowchart showing the processing of the estimation unit 82.

  First, the estimation unit 82 acquires evaluation target person social information, which is information related to an account estimated to be the same person as the evaluation target person, from the SNS 34 based on the association information acquired from the identity determination device 41. The evaluation target person information and the evaluation target person social information are combined (step 501).

  FIG. 16 is a diagram illustrating an example of a data set generated by the estimation unit 82. Referring to FIG. 16, evaluation target person information 91 is combined with evaluation target person social information 92 that is information acquired from the SNS 34 regarding the account C estimated to be the same person as the evaluation target person.

  Furthermore, the estimation unit 82 prepares evaluation information 93 including an accident occurrence probability item for which no value is set, and adds the information to the evaluation target person information 91 and the evaluation target person social information 92 to further combine them (step 502). . Referring to the data set of FIG. 16, evaluation information 93 indicated by a broken line is added.

  Subsequently, the estimation unit 82 performs a process of complementing an item (blank item) in which the value of the data set in FIG. 16 is not set by estimation using the credit model generated by the learning unit 81 (step) 503). In that case, the estimation part 82 estimates the item of the blank of a data set using a credit model based on the evaluation object person social information 92. FIG. However, since the minimum requirement is the probability of an accident occurring, even if there are other blanks, it is not always necessary to supplement to that extent. Here, the estimation unit 82 ends the complementing process when only the accident occurrence probability SS is calculated.

  FIG. 17 is a diagram illustrating an example of a data set in which the accident occurrence probability SS of the evaluation information 94 is complemented. The data set in FIG. 16 is corrected as shown in FIG. 17 at the stage where the processing in step 503 is completed.

  Furthermore, the estimation unit 82 converts the calculated accident occurrence probability into a risk score that is an index indicating the risk of accident occurrence, and notifies the loan system 32 (step 504). The conversion of the accident occurrence probability risk score can be realized, for example, by preparing a correspondence table in advance and referring to it.

  Here, as an example, the risk score obtained by converting the accident occurrence probability is notified to the lending system 32. However, the calculated accident occurrence probability may be notified as the risk score as it is.

  As described above, the identity determination device 41 may determine that there are a plurality of accounts of the person to be evaluated.

  In this case, as an example in this embodiment, the estimation unit 82 calculates the accident occurrence probability for each account, and generates the accident occurrence probability of the person to be evaluated by statistical processing from the plurality of accident occurrence probabilities. For example, a weighted average with the identity score of the accident occurrence probability of each account as a weight is calculated, and the result is set as the accident occurrence probability of the person to be evaluated.

  However, this is an example, and the embodiment is not limited to this. As another example, the estimation unit 82 may acquire information from the SNS 34 for a plurality of accounts and statistically process the information to generate one evaluation target person social information. In this case, for example, a weighted average with the identity score as a weight may be used as statistical processing.

  In the present embodiment, as an example, the collection device 37 may continuously collect SNS information for the person (transaction execution person) for whom the loan has been executed, and store the SNS information in the SNS information DB 38. At this time, the collection device 37 may accumulate time information indicating the time when the collected information is registered, input, or collected. As a result, time-series information can be used as information from the SNS 34 regarding the transaction executing person. Also, time-series information can be used as credit model learning data.

  Therefore, the learning unit 81 may use a change of a predetermined item of the learning target person social information with time as another item of the learning target person social information and use it as factor information in the credit model. For example, the learning unit 81 may add the person who caused the accident among the transaction executing persons to the person to be learned, and use the information collected by the collection device 37 for the calculation of the credit model.

  The estimation unit 82 calculates the accident risk score of the evaluation target person based on the evaluation target person information, the evaluation target person social information accumulated in time series, and the credit model in which the time change is included in the factor information. You may decide to do it. As a result, a time change of a predetermined matter becomes factor information in the credit model, which affects an accident occurrence probability calculated by the credit model.

  For example, the estimation unit 82 may calculate an accident occurrence probability using information collected by the collection device 37 as a credit on the transaction executing person. Since the information on the SNS 34 that changes with the passage of time is used for the credit on the way, the change in the creditworthiness of the person executing the transaction can be favorably reflected in the accident occurrence probability. For the person who executed the loan, the time change of predetermined information on the SNS 34 such as the content of the statement and the frequency of the statement can be reflected in the accident occurrence probability. Specifically, the probability of occurrence of an accident can be increased if a person's utterance to the SNS 34 decreases rapidly. In addition, if the information on the SNS 34 of all persons is accumulated in time series, the amount of data becomes enormous. However, since the information is accumulated in time series only for the person who executed the transaction, the amount of data to be accumulated Can be reduced.

  In this embodiment, the learning target person social information and the evaluation target person social information may include, as an item, a network structure formed by a link connecting the target person and a friend. In general, social networks have a function to connect and associate friends with links. Therefore, the network structure of the link centering on the target person can be acquired from the SNS 34.

  For example, there are few friends, the network is small, conversely there are many friends, the network is large, friends are linked to each other in a mesh, or a star topology with a certain person at the center. The network configuration will affect the calculation of the accident probability.

  The learning unit 81 calculates a credit model using the learning target person social information, and the estimation unit 82 calculates an accident occurrence probability of the evaluation target person using the credit model. The processes of the learning unit 81 and the estimation unit 82 at that time are not different from the processes described above. According to this example, by including the network structure as factor information in the credit model, information specific to social networks is reflected in the calculation of the accident occurrence probability, and risk calculation considering factors that could not be obtained by conventional methods Is possible.

  Further, when the accident occurrence probability or the accident occurrence risk score is lower than a certain reference value, the black list may be generated by including the person to be evaluated in the black list. By using the accident occurrence probability that cannot be calculated without using the information of the SNS 34, it is possible to register a person with a high accident occurrence probability in the black list and call attention before the accident actually occurs.

  Moreover, when the accident occurrence probability or the accident occurrence risk score is lower than a certain reference value, information on the SNS 34 related to the person to be evaluated may be continuously collected and accumulated in time series. . In that case, it is preferable to calculate the accident occurrence probability continuously as the credit on the way using the time-series information. By using the accident occurrence probability that cannot be calculated without using the information of the SNS 34, it is possible to register a person with a high accident occurrence probability in the black list and call attention before the accident actually occurs.

  In Example 1, as shown in FIG. 11, the example which complements one blank of candidate person social information with one value was shown. This is because the blank is complemented with a value having the highest probability of occurrence as a value entering the blank using the person model. However, the embodiment of the present invention is not limited thereto. Here, as Example 2, an example in which one blank of candidate person social information is complemented with a plurality of types of values and the occurrence probability of each value is shown.

  In this embodiment, when the estimation unit 52 supplements the blank of the candidate person social information, each value that can be generated as the value of the blank item by applying the information obtained from the SNS 34 to the person model, and each value Each occurrence probability is obtained, and a combination of a plurality of values and occurrence probabilities is set as the content of the blank item.

  FIG. 18 is a diagram illustrating an example of candidate person correction social information obtained by complementing candidate person social information in the second embodiment. Referring to FIG. 18, in the candidate person correction social information A1 of the candidate account A, a plurality of values AA1, AA2, and AA3 and occurrence probabilities 305, 15%, and 13% are set in the address item. Yes.

  Similarly, in the candidate person social information A2 of the candidate account B, a plurality of values BC1 and BC2 and their occurrence probabilities 50% and 20% are set in the work field.

  Further, in the candidate person social information A2 of the candidate account B, a plurality of values BJ1 and BJ2 and their respective occurrence probabilities of 70% and 10% are set in the item of occupation.

  In addition, in the candidate person social information A3 of the candidate account C, a plurality of values CA1, CA2, and CA3 and respective occurrence probabilities of 38%, 10%, and 8% are set in the address item.

  In this case, when calculating the identity score, the determination unit 53 weights and accumulates the score of each complemented value by the occurrence probability of that value. For example, it is assumed that the value AA1 of the address item of the candidate account A matches the address of the person to be evaluated, the value AA2 is different in number, and the value AA3 does not match. Then, in the address item part, 1.0 (address weight) × (1.0 (match) × 0.3 (30%) + 0.8 (differing in number) × 0.15 (15%) + 0. 0 (mismatch) × 0.13 (13%)) = 0.42.

  Thus, according to the sail embodiment, a more stable identity score can be calculated.

  In the first embodiment, the credit determination device 42 uses the evaluation target person information given from the lending system 32 without being processed. However, the embodiment of the present invention is not limited thereto, You may decide to use the evaluation object person information which processed.

  In the third embodiment, when the identity determination device 41 specifies an account of an evaluation target person, the evaluation target person information is corrected based on information obtained from the SNS 34 in relation to the account. The credit determination device 42 performs credit processing using the corrected evaluation target person information.

  FIG. 19 is a block diagram of the identity determination apparatus according to the third embodiment. Referring to FIG. 19, the identity determination device B <b> 0 according to the third embodiment includes a learning unit 51, an estimation unit 52, a determination unit 53, and a person information correction unit B <b> 1. The learning unit 51, the estimation unit 52, and the determination unit 53 are the same as those in the first embodiment shown in FIG.

  Association information that is a determination result of the determination unit 53 is input to the person information correction unit B1. The association information indicates a correspondence relationship between the evaluation target person information and the account of the person having the highest identity with the evaluation target person indicated by the evaluation target person information. The person information correction unit B1 acquires information on the SNS 34 relating to the account indicated in the input association information. Subsequently, the person information correction unit B1 corrects the evaluation target person information based on the information acquired from the SNS 34.

  For example, since it is not an essential item, it is conceivable that information on items that the applicant did not fill out at the time of application is supplemented based on information from the SNS 34. In addition, there is a possibility that incorrect information is set because the applicant made a mistake during application. Based on the information from the SNS 34, it is conceivable to correct the erroneous information to correct information.

  The person information correction unit B1 notifies the credit determination device 42 of the corrected evaluation target person information (correction evaluation target person information). In this case, the credit determination device 42 executes a process of performing credit determination of the evaluation target person using the corrected evaluation target person information instead of the evaluation target person information.

  In the present embodiment, the person information correction unit B1 acquires information from the SNS 34 and corrects the evaluation target person information based on the information. However, the embodiment of the present invention is limited to this. There is nothing. As another example, the person information correction unit B1 may acquire the candidate person correction social information output from the estimation unit 52, and correct the evaluation target person information based on the candidate person correction social information.

  Further, the above-described embodiments and examples of the present invention are exemplifications for explaining the present invention, and the scope of the present invention is not intended to be limited only to those embodiments or examples. Those skilled in the art can implement the present invention in various other modes without departing from the gist of the present invention.

  In the first to third embodiments, the estimation unit 82 uses the credit model to estimate the value of the accident occurrence probability item, which is a blank in the data set, based on the evaluation target person social information 92. However, the embodiment of the present invention is not limited thereto. As another method, the estimation unit 82 holds a risk score table indicating a score for an accident occurrence risk for each content included in items other than the accident occurrence probability as a credit model, and each item of the evaluation target person social information 92 The accident occurrence risk score of the person to be evaluated may be calculated by accumulating scores corresponding to the contents of.

  Moreover, although the learning part 21 showed the example which calculates a person model previously in Embodiment 1-3 mentioned above, the Example of this invention is not limited to this. As another example, the person model may be calculated when a request for determining the identity of the person to be evaluated and the account is generated. As yet another example, each time new data is added as person model learning data, the person model may be calculated and updated.

  Similarly, in the first to third embodiments, the learning unit 81 calculates the credit model in advance. However, the embodiment of the present invention is not limited to this. As another example, a credit model may be calculated when a request for credit determination occurs. As another example, each time new data is added as credit model learning data, the credit model may be calculated and updated.

  Moreover, in Examples 1-3, the example which calculates the accident occurrence risk score of an evaluation object person only by the information obtained from SNS34 was shown, without using the result of the credit obtained by the existing evaluation method. However, the embodiment of the present invention is not limited to this. As another example, using both the credit result (social credit result) obtained from information on the SNS 34 and the credit result (non-social credit result) obtained by other evaluation methods, the credit result is comprehensively obtained. You may decide to calculate. For example, a weighted average value of a score that is a social credit result and a score that is a non-social credit result may be used. At that time, weighting may be performed depending on how much importance is given to the social credit result and the non-social credit result. Or you may decide to change the weighting ratio of a social credit result and a non-social credit result according to the reliability of social credit results, such as an identity score.

  In the first to third embodiments, a caching system that uses a combination of an identity determination device 41 that extracts an account of a person having high identity with the target person from the SNS 34 and a credit determination device 42 that performs credit processing on the target person. However, the embodiments of the present invention are not limited thereto.

  The identity determination device 41 can be combined with other devices and used in systems for other purposes. For example, a marketing system may be constructed by combining an advertisement distribution device (not shown) that distributes advertisement information of products and services suitable for individuals with an identity determination device 41. It is conceivable to improve the marketing effect by using information related to individuals obtained from the SNS 34 for individual marketing such as products and services. In that case, the identity determination device 41 can extract an account having high identity with the target person, and the advertisement distribution device can select an advertisement to be sent to each individual using information obtained from the SNS 34 related to the account. .

  Further, for example, a recruitment system is constructed by combining the identity determination device 41 with a job information providing device (not shown) that provides job information and job search information that suits each person's suitability and desire, and the identity determination device 41. Also good. Employment and job changes such as personal information obtained from the SNS 34 can be used to secure human resources that suit the content of the job or the type of job that the company wants, and individuals can get a job that suits them. It is possible to make it possible. In that case, the identity determination device 41 extracts an account having high identity with the target person, and the job information providing device selects job information to be sent to each individual using information obtained from the SNS 34 related to the account. be able to.

  Also, the credit determination device 42 can be used in combination with other devices or as a single unit in a credit evaluation system. For example, when the loan system 32 accepts the application, it may be requested to log in (social login) using the account of the SNS 34. By doing so, the loan system 32 can determine the association between the target person and the account on the SNS 34 at the time of application. If the lending system 32 notifies the credit determination device 42 of information associated with the credit determination device 42, the credit determination device 42 knows the correspondence between the target person and the account without the identity determination device 41. be able to.

  In addition, each part of the identity determination apparatuses 11 and 41, the information utilization apparatus 12, and the credit determination apparatus 42 of each embodiment and Example mentioned above is implement | achieved by making a computer run the software program which prescribed | regulated the process procedure of each part. You can also Further, each part (including the case of using software) constituting the identity determination devices 11 and 41, the information use device 12, and the credit determination device 42 does not necessarily exist in the same computer, and is distributed to a plurality of computers. May be.

DESCRIPTION OF SYMBOLS 10 ... Social network information utilization system, 11 ... Identity determination apparatus, 12 ... Information utilization apparatus, 21 ... Learning part, 22 ... Estimation part, 23 ... Determination part, 31 ... Credit system, 32 ... Loan system, 33 ... User terminal 34 ... SNS, 35 ... Internet, 36 ... evaluation system, 37 ... collection device, 38 ... SNS information DB, 41 ... identity judgment device, 42 ... credit judgment device, 51 ... learning unit, 52 ... estimation unit, 53 ... Judgment part 61 ... Person information 62 ... Evaluation target person information 71-73 ... Candidate person social information 74-76 ... Candidate person correction social information 76 ... Candidate person correction social information 81 ... Learning part 82 ... Estimation 91: Evaluation object person information, 92 ... Evaluation object person social information, 93 ... Evaluation information, 94 ... Evaluation information, A1 ... Candidate person correction social information A2 ... candidate person social information, A3 ... candidate person social information, B0 ... identity determination device, B1 ... person information correcting unit

Claims (12)

Account owners of candidate accounts extracted by narrowing down accounts on social networks based on at least part of the target person information with multiple items of person information, each of which is given a target person value Obtaining candidate person social information, which is related information on the social network, and providing values that may occur for at least some items of person information based on the acquired candidate person social information and information on the social network An estimation unit that generates candidate person-corrected social information including values of a plurality of items that can be compared with person information based on the person model
A determination unit that determines the identity of the account owner and the target person for each candidate account by comparing the target person information and the candidate person-corrected social information for each candidate account. Network information processing device.   Calculation is performed using, as learning data, personal information related to a plurality of learning target persons whose correspondence with the account owner is known, and social information related to the learning target account owner who is the account owner corresponding to the learning target person. The social network information processing apparatus according to claim 1, further comprising: a learning unit that calculates the person model. The person model provides information indicating a value that can occur for at least some items of person information and an occurrence probability of the value,
The candidate person correction social information includes information indicating a value given by the person model and an occurrence probability of the value for at least some items;
The determination unit calculates an identity score indicating the identity between the account owner and the target person for each candidate account based on the value indicated by the candidate person correction social information and the occurrence probability.
The social network information processing apparatus according to claim 1.   The said determination part determines the account with the highest identity with the said target person in the said candidate accounts by comparing the said target person information and the said candidate person correction social information. Social network information processing device.   The candidate person social information acquired by the estimation unit includes the profile of the account owner set in the candidate account, text input from the candidate account, and a link from the candidate account to a predetermined number of steps. The social network of claim 1, comprising some or all of an associated account connected with a candidate account and text about the account owner of the candidate account entered from the associated account or other account. Information processing device.   If the information on the social network related to the account owner of the candidate account exceeds a predetermined condition, the estimating unit samples a part of the information on the social network related to the account owner by sampling. The social network information processing apparatus according to claim 1, wherein the social network information processing apparatus acquires the candidate person social information. Based on the candidate person social information or the candidate person correction social information for the account determined to have the highest identity with the target person by the determination unit, the target person information is corrected to generate correction target person information. A personal information correction unit;
The social network information processing apparatus according to claim 4. The determination unit corrects by a predetermined correction process if information obtained from a social network related to the account owner of an account determined to have the highest identity with the target person does not match the target person information. To
The social network information processing apparatus according to claim 4. The correction process is a process of matching information obtained from a social network regarding the account owner of an account determined to have the highest identity with the target person with the target person information.
The social network information processing apparatus according to claim 8. In the correction process, among items obtained from the social network regarding the account owner of the account determined to have the highest identity with the target person, items that do not match the target person information are assumed to have no value. , Replace with values estimated solely from other information from social networks,
The social network information processing apparatus according to claim 8. The estimation means provided in the computer is extracted by narrowing down the accounts on the social network based on at least a part of the target person information, which is a plurality of items of person information and each item is given the value of the target person The candidate person social information that is information on the social network related to the account owner of the candidate account is acquired, and at least a part of the person information based on the acquired candidate person social information and the information on the social network Generating candidate person-corrected social information including values of multiple items that can be compared with person information based on a person model that provides possible values for
The determination means provided in the computer determines the identity between the account owner and the target person for each candidate account by comparing the target person information and the candidate person correction social information for each candidate account. Social network information processing method. The estimation means is a candidate account extracted by narrowing down the accounts on the social network based on at least a part of the target person information in which multiple items of person information and the value of the target person are given to each item The candidate person social information, which is information on the social network related to the account owner, is obtained, and the candidate person social information obtained and the information on the social network are generated for at least some items of the person information. Generating a candidate person-corrected social information including values of a plurality of items that can be compared with person information based on a person model that provides a possible value;
The determination means compares the target person information with the candidate person correction social information for each candidate account, thereby determining the identity between the account owner and the target person for each candidate account. , A social network information processing program for causing a computer to execute.

Images