Information processing device, information processing method, and information processing program

Abstract

An information processing device acquires log data that records a response operation by a second user with respect to presentation information presented by a first user, calculates, on the basis of the log data, an edge weight value in a graph that represents, using a plurality of nodes corresponding to a plurality of users including the first user and the second user and an edge connecting the individual nodes, a relationship between the users, and performs user profiling on the basis of the graph in which the weight value is calculated and, as the second user performs the response operation with a higher psychological barrier with respect to the presentation information of the first user, the calculation of the weight value includes calculating the larger edge weight value corresponding to the first user and the second user.

Description

Information processing device, information processing method, and information processing program

[0001] The present disclosure relates to user profiling techniques.

[0002] Patent Literature 1 discloses a social network graph composed of nodes and edges. In the social network graph shown in Patent Literature 1, the nodes include user nodes corresponding to each user, concept nodes related to specific concepts, and topic nodes related to specific topics, and edges define connections between pairs of nodes.

[0003] However, Patent Document 1 does not disclose how to calculate the edge weight value. Therefore, Patent Document 1 cannot calculate the weight value taking into account the fact that a user has made a response operation with a high psychological barrier to information presented by another user. This does not allow for user profiling that takes into account the level of psychological barrier in a user's response operation. In the present invention, the "level of psychological barrier in a user's response operation" can be rephrased as "the degree to which a user's response operation is positive or negative in relation to their own or others' behavior or behavioral changes."

[0004] Special Publication No. 2013-539568

[0005] The present disclosure is intended to solve such problems, and aims to provide a technology that can perform user profiling taking into account the level of psychological barriers in a user's response operation.

[0006] The information processing device of the present disclosure includes an acquisition unit that acquires log data that records a response operation of a second user to presented information presented by a first user; a weight calculation unit that calculates, based on the log data, a weight value of the edge in a graph that represents the relationship between users using a plurality of nodes corresponding to a plurality of users including the first user and the second user and edges connecting each node; and a profiling unit that performs user profiling based on the graph in which the weight value has been calculated, wherein the weight calculation unit calculates a larger weight value for the edge corresponding to the first user and the second user when the second user performs a response operation that poses a higher psychological barrier to the presented information of the first user.

[0007] According to the present disclosure, it is possible to perform user profiling that takes into consideration the level of psychological barriers in the user's response operation.

[0008] 1 is a block diagram showing an example of a configuration of an information processing system according to a first embodiment of the present disclosure; FIG. 2 is a histogram of the number of times a comment is input; FIG. 3 is a histogram of the number of times an agreement stamp is input; FIG. 4 is a flowchart showing an example of processing by the information processing system; FIG. 5 is a flowchart showing details of a weight value calculation process when a first method is adopted; FIG. 6 is a flowchart showing details of a weight value calculation process when a second method is adopted; FIG. 7 is a flowchart showing details of a first example of profiling processing; and FIG. 8 is a flowchart showing details of a second example of profiling processing.

[0009] (Foundation of the Present Disclosure) In recent years, the number of users who use social network service applications (hereinafter referred to as SNS applications) via the Internet using smartphones or personal computers has been increasing.

[0010] One of the main functions of SNS apps is the follow function. With the follow function, information such as posts from favorite users is preferentially displayed on the screen of the SNS app on a smartphone. The follow function also allows the followed user to communicate that the following user likes the user. Generally, SNS apps can display the follow status to both the follower and the followed user.

[0011] This follow relationship is information that represents the relationship between users. The SNS app uses this follow relationship to calculate weight values ​​for edges in a social graph. A social graph is composed of nodes corresponding to users and edges that represent the relationships between users. A social graph is generally composed of a directed graph in which edges are assigned directionality, or an undirected graph in which edges are not assigned directionality. In the present disclosure, a social graph refers to a graph (hereinafter referred to as a social graph) that represents the relationships between users using multiple nodes corresponding to multiple users and edges connecting each node.

[0012] However, there are also SNS apps that do not have a follow function. In such SNS apps, the lack of a follow function makes it impossible to calculate edge weights, making it difficult to create a social graph.

[0013] Therefore, the inventors focused on response operations performed by a user in response to information presented by another user (e.g., information introducing a visiting spot). Response operations include posting a comment, clicking a "Like" button, posting a "Like" stamp, etc. Response operations also include a browsing operation to browse the information introduced by a user.

[0014] Of these response operations, commenting is time-consuming because it requires text input. On the other hand, posting by clicking the "Like" button can be done with a single touch, so it is not as time-consuming as commenting. While response operations vary in the amount of effort required, users who perform time-consuming response operations are likely to have a higher level of interest in the user who provided the displayed information. In other words, users who perform time-consuming response operations with a high psychological barrier are likely to have a higher level of interest in the user who provided the displayed information. This level of interest can be an indicator of the relationship between users.

[0015] Therefore, the inventors discovered that if the weight value of an edge is calculated depending on whether or not another user has performed a response action that poses a high psychological barrier to information presented by a certain user, a social graph can be created without using the follow function, thereby enabling highly accurate user profiling, and this led to the conceiving of each aspect of the present disclosure.

[0016] Note that user profiling in the present disclosure refers to clarifying user characteristics. For example, user profiling refers to clarifying user characteristics such as the user's interests, personality (e.g., liking to actively communicate with others), relationships with others (being at the center of a community), and values ​​regarding one's own behavior (e.g., liking to walk), but the manner in which it is done is not particularly limited. For example, user profiling can suggest to a user who likes walking that they walk more for the sake of their health.

[0017] (1) An information processing device in the present disclosure includes an acquisition unit that acquires log data that records a response operation of a second user to presented information presented by a first user; a weight calculation unit that calculates, based on the log data, weight values ​​of edges in a graph that represents relationships between users using multiple nodes corresponding to multiple users including the first user and the second user and edges connecting each node; and a profiling unit that performs user profiling based on the graph in which the weight values ​​have been calculated, wherein the weight calculation unit calculates a larger weight value for the edge corresponding to the first user and the second user when the second user performs a response operation that poses a higher psychological barrier to the presented information of the first user.

[0018] According to this configuration, the higher the psychological barrier a second user has in responding to the information presented by the first user, the larger the weight value of the edge corresponding to the first user and the second user is calculated, thereby enabling user profiling that takes into account the level of the psychological barrier in the user's response.

[0019] (2) In the information processing device described in (1) above, the weight calculation unit may calculate the weight value using a value that is predetermined depending on the type of the response operation.

[0020] According to this configuration, the weight value is calculated using a predetermined value depending on the type of response operation, so that the number of processing steps and processing load required for calculating the weight value can be reduced.

[0021] (3) In the information processing device described in (1) or (2) above, the weight calculation unit may classify the log data according to the type of the response operation, calculate a histogram showing the relationship between the number of response operations and the frequency of the number of times for each type of response operation based on the classified log data, calculate an average value of the histogram for each type of response operation, calculate a weight value for each type of response operation based on the reciprocal of the average value calculated for each type of response operation, and calculate the weight value using the type-specific weight value.

[0022] Because response operations occur discretely and the higher the psychological barrier, the fewer users will perform that response operation, it can be assumed that the shape of the response operation histogram follows a negative binomial distribution. Furthermore, based on this assumption, the inventors have discovered that the type-specific weight values ​​for each type of response operation can be defined by using the reciprocal of the average value of the histogram for each type of response operation. With this configuration, a histogram is calculated for each type of response operation, the average value for each type of response operation is calculated from the calculated histogram, and the type-specific weight values ​​are calculated from the reciprocal of the average value. The type-specific weight values ​​are then used to calculate edge weight values. Therefore, statistically supported weight values ​​can be calculated.

[0023] (4) In the information processing device described in any of (1) to (3) above, the profiling unit may detect the shortest route between nodes for every combination of any two different nodes in the graph based on the weight value, detect the number of times a node corresponding to one user appears as an intermediate node in the shortest route, and calculate the influence of the one user on other users based on the number of times it appears.

[0024] According to this configuration, the influence of a single user is calculated based on the number of times that a node corresponding to the single user appears as an intermediate node on the shortest route between any two nodes. This allows users who can become communication hubs to be evaluated as highly influential users.

[0025] (5) In the information processing device described in any of (1) to (4) above, the profiling unit may calculate the degree of influence of a user corresponding to one node on other users based on the number of edges connected to one node and the weight value for the edges connected to the one node.

[0026] According to this configuration, the influence of a user corresponding to a node is calculated based on the number of edges connected to the node and the weight value for the edge, so that the stronger the relationship between a user and other users, the higher the calculated influence can be.

[0027] (6) The information processing device according to any one of (1) to (5) above may further include an output unit that outputs the results of the user profiling.

[0028] This configuration allows the results of profiling to be presented.

[0029] (7) The information processing device according to (4) or (5) above may further include an output unit that generates list information in which the plurality of users are arranged in descending order of the degree of influence, and outputs the list information.

[0030] With this configuration, each user can check the relative level of their own influence.

[0031] (8) In the information processing device described in any of (1) to (7) above, the response operation may include inputting a comment on the presented information, inputting a button or stamp indicating agreement with the presented information, and an operation of viewing the presented information.

[0032] According to this configuration, response operations include comments, input of buttons or stamps indicating agreement, and viewing of presented information, so that a weight value according to the level of the psychological barrier can be calculated with high accuracy.

[0033] (9) An information processing method in another aspect of the present disclosure is an information processing method in a computer, comprising: acquiring log data that records a response operation of a second user to presented information presented by a first user; calculating, based on the log data, weight values ​​of the edges in a graph that represents a relationship between users using a plurality of nodes corresponding to a plurality of users including the first user and the second user and edges connecting each node; and performing user profiling based on the graph in which the weight values ​​have been calculated, wherein calculating the weight values ​​includes calculating larger weight values ​​for the edges corresponding to the first user and the second user when the second user performs a response operation that poses a higher psychological barrier to the presented information of the first user.

[0034] According to this configuration, an information processing method is provided that can perform user profiling taking into consideration the level of psychological barriers in the user's response operation.

[0035] (10) In another aspect of the present disclosure, an information processing program causes a computer to acquire log data recording a response operation of a second user to presented information presented by a first user; calculate, based on the log data, weight values ​​of the edges in a graph representing relationships between users using multiple nodes corresponding to multiple users including the first user and the second user and edges connecting each node; and perform user profiling based on the graph in which the weight values ​​have been calculated. The calculating of the weight values ​​includes calculating a larger weight value for the edge corresponding to the first user and the second user when the second user performs a response operation that poses a higher psychological barrier to the presented information of the first user.

[0036] According to this configuration, an information processing program is provided that can perform user profiling taking into consideration the level of psychological barriers in the user's response operation.

[0037] The present disclosure can also be realized as an information processing system operated by such an information processing program. Needless to say, such a computer program can be distributed on a non-transitory computer-readable recording medium such as a CD-ROM or via a communication network such as the Internet.

[0038] Note that each of the embodiments described below represents a specific example of the present disclosure. The numerical values, shapes, components, steps, and step orders shown in the following embodiments are merely examples and are not intended to limit the present disclosure. Furthermore, among the components in the following embodiments, components that are not described in the independent claims that represent the highest concept are described as optional components. Furthermore, in all of the embodiments, the respective contents can be combined.

[0039] 1 is a block diagram showing an example of a configuration of an information processing system 100 according to a first embodiment of the present disclosure. The information processing system 100 includes an information processing device 1 and a terminal 4.

[0040] The terminal 4 is a device that acquires information about a user who is the subject of information processing. Examples of the terminal 4 include a smartphone, a personal computer, a wearable terminal, and a physical measurement device. Examples of the wearable terminal include smart glasses and a smart watch. Examples of the physical measurement device include a pedometer. Note that the form of the terminal 4 is not limited to these examples. There may be one or more terminals 4. The terminal 4 inputs the acquired log data to the information processing device 1 at a predetermined timing.

[0041] The log data is data that records a response operation by a second user in response to presentation information presented by a first user. The log data includes a timestamp indicating the date and time of the response operation, type information indicating the type of the response operation, the user ID of the user who performed the response operation, the user ID of the user who performed the response operation, and the content of the response operation. The content of the response operation includes text data indicating the content of a comment on the presentation information.

[0042] The response operation includes, for example, inputting a comment on the presented information, a button indicating agreement with the presented information, and inputting a stamp indicating agreement with the presented information. The button indicating agreement is, for example, a "Like" button. The stamp is, for example, an image that symbolically conveys agreement or favorability. The presented information is, for example, posted information posted on an SNS app. For example, posted information corresponds to information introducing recommended places to visit. The presented information also includes comments posted by users on the posted information.

[0043] In this way, a response operation is an operation that can be observed by other users. For example, if a user writes a comment on information presented by another user, the user can recognize that the comment was made by the other user.

[0044] The presented information may be self-introduction information that presents a self-introduction of a certain user. In this case, the response operation is a viewing operation in which another user views the self-introduction information of a certain user. Although the viewing operation is not a viewable operation, at least the user who performed the viewing operation is interested in the user who posted the self-introduction information, and therefore, in the present disclosure, the viewing operation is treated as a response operation.

[0045] An SNS app is, for example, an app for revitalizing a local community. This SNS app may have a function for inviting users to go for a walk to improve their health. The SNS app accepts posts from users about recommended places to visit and posts the places to visit on a bulletin board that can be viewed by all users. The SNS app accepts comments from other users about the places to visit and posts the comments on the bulletin board in association with the places to visit. In this way, the SNS app can motivate users to visit the places to visit and take a walk. Furthermore, users can get to know other users through the comments and build a personal network.

[0046] The information processing device 1 is a device for performing user profiling using log data, and is configured with a computer including a processor 2, a memory 3, and an interface circuit (not shown).

[0047] The processor 2 is hardware for performing user profiling using log data, and is, for example, a central processing unit. The processor 2 includes an acquisition unit 21, a weight calculation unit 22, a profiling unit 23, and an output unit 24. The acquisition unit 21 to the output unit 24 may be realized by the central processing unit executing an information processing program stored in the memory 3, or may be configured by a dedicated hardware circuit such as an ASIC. The acquisition unit 21 to the output unit 24 may be distributed among multiple computers. Some of the acquisition unit 21 to the output unit 24 may be included in the terminal 4.

[0048] The memory 3 is configured by a rewritable nonvolatile storage device such as a solid state drive (SSD) or a hard disk drive (HDD). The memory 3 includes a log storage unit 31. The log storage unit 31 stores the log data acquired by the acquisition unit 21 from the terminal 4.

[0049] The acquisition unit 21 acquires the log data stored by the terminal 4 and stores the acquired log data in the log storage unit 31. The log storage unit 31 stores the log data input from the acquisition unit 21.

[0050] The weight calculation unit 22 performs preprocessing on the log data stored in the log storage unit 31. The weight calculation unit 22 calculates the weight of the user U ｋ Day d ｊ The weight calculation unit 22 calculates the total number of inputs for each type of response operation. k is an index for identifying the user. j is an index for identifying the day. ｋ The statistical value can be an average value, a median value, or a total value. The weight calculation unit 22 performs this process for all users U ｋ This allows each user U to ｋ Here, the statistics of the number of inputs per day is obtained. ｊ However, the total number of inputs over a predetermined period (for example, two days, one week, one month, etc.) may be calculated. ｋ The statistical value calculated for each response is treated as the number of times the response operation is input.

[0051] The weight calculation unit 22 uses the log data stored in the log storage unit 31 to create a graph (social graph) representing the relationships between users using multiple nodes corresponding to multiple users including a first user and a second user and edges connecting each node. Each of the multiple users is a user registered in the above-mentioned SNS app. Each of the multiple users is a user belonging to a community in a certain area. The weight calculation unit 22 uses the log data stored in the log storage unit 31 to calculate weight values ​​of the edges.

[0052] The weight value is a non-negative value, and the larger the value, the stronger the connection between the nodes.

[0053] The weight calculation unit 22 calculates a weight value for the edge corresponding to the first user and the second user to be larger as the second user performs a response operation that poses a higher psychological barrier to the information presented by the first user.

[0054] The weight calculation unit 22 calculates the weight values ​​using the following first or second method.

[0055] (First Method) The weight calculation unit 22 calculates a weight value using a predetermined value according to the type of response operation. Hereinafter, the predetermined value according to the type of response operation will be referred to as a "type-specific weight value." Response operations include inputting a comment, inputting an agree button or an agree stamp, and an operation to view self-introduction information. These response operations are ranked in order of increasing psychological barrier, as follows: inputting a comment, inputting an agree button or an agree stamp, and an operation to view. Here, it is assumed that a type-specific weight value of "3" is preset for inputting a comment, a type-specific weight value of "2" is preset for inputting an agree button and an agree stamp, and a type-specific weight value of "1" is preset for the operation to view. However, these values ​​are merely examples, and other values ​​may be used as appropriate.

[0056] For example, user U １ User U (first user) ２ When there is data in which a comment is input once, the agree button is input twice, and a browsing operation is performed three times from the (second user), the weight calculation unit 22 calculates the weight of the user U １ Node N １ and user U ２ Node N ２ Edge ED that connects １ The weight value W １ W １ = 3 × 1 + 2 × 2 + 1 × 3 = 10. The weight calculation unit 22 calculates weight values ​​for the other edges that make up the social graph by performing similar calculations.

[0057] The type-specific weight values ​​are determined in advance by the service provider. In this way, higher type-specific weight values ​​are preset for response operations with higher psychological barriers, so the weight calculation unit 22 can calculate larger weight values ​​for edges that represent stronger relationships between users. This allows a social graph to be constructed that takes into account the level of psychological barriers.

[0058] (Second Method) The weight calculation unit 22 classifies the log data stored in the log storage unit 31 according to the type of response operation. Based on the classified log data, the weight calculation unit 22 calculates a histogram showing the relationship between the number of response operations and the frequency of the number of times for each type of response operation. The weight calculation unit 22 calculates the average value of the histogram for each type of response operation. The weight calculation unit 22 calculates a type-specific weight value for the response operation based on the reciprocal of the average value calculated for each type of response operation. The weight calculation unit 22 calculates a weight value using the type-specific weight value. For example, if the average values ​​of the histograms for comments, agree buttons, agree stamps, and browsing operations are A1, A2, A3, and A4, respectively, the weight calculation unit 22 calculates the type-specific weight value w １ , w ２ , w ３ , w ４ For example, if there is data in which a comment is input from a second user to a first user once, an agree button is input twice, an agree stamp is input twice, and a browsing operation is performed three times, the weight calculation unit 22 calculates the weight of the node N １ and node N ２ Edge ED that connects １ The weight value W １ W １ =w １ ×1+w ２ ×2 + w3 ×2 + w4 ×3. The weight calculation unit 22 calculates weight values ​​for the other edges that make up the social graph by performing similar calculations. The calculation principle of the second method will be described later.

[0059] The profiling unit 23 performs user profiling based on the social graph for which the weight values ​​have been calculated. The profiling unit 23 detects the shortest route between nodes for every combination of any two different nodes in the social graph based on the weight values. For example, if there are five users in total, there are 10 possible combinations of any two users. In the social graph, the i-th edge ED ｉ The weight value of W ｉ In this case, the profiling unit 23 calculates the weight value W ｉ The inverse of the edge ED ｉ Distance d ｉ As a result, the weight value W ｉ The larger the distance d ｉ is calculated briefly.

[0060] For example, node N １ and node N ２ The shortest route R １ Then, the profiling unit 23 selects a node N １ and node N ２ By searching for the shortest route connecting １ Detect.

[0061] The profiling unit 23 is ｋ The profiling unit 23 detects the number of times that a node corresponding to the user U appears as an intermediate node on the shortest route. ｋ Calculate the impact of

[0062] For example, a shortest route R １ Node N is an intermediate node. ３ If the node N ３ User U corresponding to ３ The number of occurrences of is counted up by one. １ Node N is an intermediate node. ４ If the node N ４ User U corresponding to ４The number of occurrences of is counted up by one. １ is a node located between one tip node and the other tip node in the

[0063] The profiling unit 23 determines the other shortest route R ｉ Also in node N ３ , N ４ appears, the number of times that it appears, ３ , N ４ The profiling unit 23 performs this process on all the shortest routes R ｉ Execute about.

[0064] The profiling unit 23 calculates the shortest route R ｉ The ratio of the number of occurrences to the total number of occurrences is calculated for each user, and the obtained ratio is calculated as the influence of each user.

[0065] This allows the profiling unit 23 to calculate the betweenness centrality of each user in the social graph. A user with high betweenness centrality is a user who serves as a hub for other users among the users who make up the social graph, and represents a central person in communication.

[0066] The profiling unit 23 may calculate the degree of influence by the following method. The profiling unit 23 calculates the degree of influence of a user corresponding to one node on other users based on the number of edges connected to one node and the weight value of the edges connected to one node. For example, for node N ３ Three edges are connected to the １ , W ２ , W ３ In this case, the profiling unit 23 ３ The influence of W １ +W ２ +W ３ It is calculated as follows.

[0067] The output unit 24 outputs profiling data indicating the results of user profiling calculated by the profiling unit 23. The profiling data is output to, for example, a user terminal used by the user or a management terminal used by an administrator who provides the service. The user terminal may be the terminal 4.

[0068] The output unit 24 may generate list information in which multiple users are arranged in descending order of influence, and include the generated list information in the profiling data. The profiling results are not limited to list information. For example, the profiling results may include a social graph indicating the strength of the weight value W. Alternatively, the profiling results may include the influence of each individual user. In this case, the output unit 24 may transmit the influence of each individual user only to the user terminal corresponding to the relevant user. The profiling results may include the value of the influence of each user or each individual user.

[0069] (Calculation Principle of Second Method) Next, the calculation principle of the second method will be described in detail. User response operations occur discretely. Poisson distribution is known as a probability distribution of such discretely occurring events. The shape of the Poisson distribution is expressed by a parameter λ. Furthermore, the higher the psychological barrier, the gradually decreasing frequency of occurrence of response operations. Gamma distribution is known as a probability distribution that represents such continuous events. When the parameter λ of the Poisson distribution follows the gamma distribution, these two distributions can be expressed using a negative binomial distribution.

[0070] FIG. 2 is a histogram of the number of times a comment is entered. In FIG. 2, the horizontal axis represents the number of times a comment is entered, and the vertical axis represents the relative frequency of the number of times. When a histogram is created for the number of times a comment is entered, there are fewer users who post many comments, as shown in FIG. 2, and more users who post few comments, resulting in a distribution that can be considered a negative binomial distribution. The relative frequency is the total number of users using the app divided by the number of users that fall into each bin of the histogram. The bin width of the histogram is predetermined, and is, for example, 5. However, this is just an example, and the bin width can be any appropriate value other than 5, such as 1, 3, or 10.

[0071] The probability mass function of the negative binomial distribution is expressed by equation (1) for the random variable X (k=0, 1, . . . ).

[0072]

[0073] Here, p is the success probability, which represents the probability of k failures before r (r>0) successes. The number of times a comment is entered on the horizontal axis in Figure 2 corresponds to k. Note that while it may seem unnatural at first glance to "fail" after entering a comment k times, this is not a problem because if the relationship between success and failure is reversed in the negative binomial distribution, the success probability p can also be interpreted as the probability of k successes before r failures.

[0074] The expected value E of the negative binomial distribution is expressed by equation (2).

[0075]

[0076] The variance V of the negative binomial distribution is expressed by equation (3).

[0077] If the number of comment inputs follows a negative binomial distribution, the expected value E is the mean value of the histogram, and the variance V is the variance of the histogram. From equations (2) and (3), since r>0, the success probability p and failure probability r are each expressed by equation (4).

[0078] The negative binomial distribution is a mixed distribution of the Poisson distribution and the gamma distribution. Therefore, the parameter λ of the Poisson distribution can be calculated from the expected value E and variance V calculated from the histogram. First, the parameter λ of the gamma distribution is calculated from the expected value E and variance V.

[0079] The probability density function of the gamma distribution is expressed by equations (5) and (6). Here, α is a shape parameter, β is a scale parameter, and Γ represents the gamma function. The probability mass function of the Poisson distribution is expressed by equation (6).

[0080]

[0081] Since the mixed distribution of the gamma distribution and the Poisson distribution is a negative binomial distribution, if the discrete variable k of the Poisson distribution in equation (6) is regarded as a continuous variable x, the integral over the entire interval of the parameter λ is expressed by equation (7).

[0082]

[0083] Transforming the right side of equation (7) yields equation (8).

[0084]

[0085] Here, if the part after the integral sign in equation (8) is replaced by A and q=x+α is replaced by s=1 / β+1, equation (9) is obtained.

[0086]

[0087] This is because the interior of the integral of A is a gamma distribution with shape q and scale s, and therefore the integral from 0 to positive infinity is 1 according to the definition of the gamma function. Furthermore, for the gamma function, equation (10) holds for a positive integer n.

[0088]

[0089] Therefore, equation (7) can be expressed as equation (11) using equations (8), (9), and (10).

[0090]

[0091] When the formula (11) is expressed in combination symbols, the formula (12) is obtained.

[0092]

[0093] By replacing r with α and k with x in equation (1), we can convert the negative binomial distribution into a continuous distribution P ｎｂ Considering (x)', we obtain equation (13).

[0094]

[0095] Since both equations (12) and (13) represent a negative binomial distribution, the relationship in equation (14) is obtained.

[0096]

[0097] Furthermore, the shape parameter α of the gamma distribution is equal to r by the definition of the negative binomial distribution.

[0098] When we focus on the relationship between users and responses with high psychological barriers, we believe that responses with high psychological barriers strongly reflect the relationship between users. This is because, for example, if a user posts a comment to another user that is a response with high psychological barriers, it can be considered that the comment was made despite the high psychological barrier.

[0099] By using the reciprocal of the expected value E of the parameter λ of the Poisson distribution generated from the gamma distribution as the type-specific weight w, the magnitude of the psychological barrier can be reflected in the weight W. Here, the expected value E of the parameter λ of the Poisson distribution is the expected value E of the gamma distribution, and is therefore expressed as αβ. The larger the parameter λ, the higher the occurrence probability. Therefore, to express the height of the psychological barrier, it is necessary to take the reciprocal of the expected value E. Note that the height of the psychological barrier is not limited to a simple reciprocal, and the square of the reciprocal may also be used. In this case, the influence of the psychological barrier can be more strongly reflected in the type-specific weight w.

[0100] The type-specific weight value w is expressed by equation (15) using equations (4) and (14).

[0101]

[0102] Therefore, the type-specific weighting value w based on the magnitude of the psychological barrier is determined from the histogram's expected value E (average value) and variance V. This allows highly accurate user profiling to be achieved using the type-specific weighting value w that is statistically supported.

[0103] Generally, when estimating the distribution of a population using a small amount of data, the expected value E has higher estimation accuracy than the variance V. This is because the variance V includes a square calculation and is therefore more affected by a small number of outliers than the average value. Because the type-specific weights w are composed only of the expected values ​​E estimated from the histogram, the type-specific weights w can be determined with high accuracy even with a small amount of data.

[0104] The weight calculation unit 22 calculates the expected value E of the histogram shown in FIG. 2, and substitutes the expected value E into equation (15) to calculate the type-specific weight value w for the comment.

[0105] Figure 3 is a histogram of the number of times consent stamps were entered. The vertical and horizontal axes are the same as those in Figure 2. In the histogram in Figure 3, the peak of relative frequency is shifted to the right compared to Figure 2. This is because there is a lower psychological barrier to entering consent stamps compared to entering comments.

[0106] The weight calculation unit 22 also calculates the expected value E for the histogram of FIG. 3, and substitutes the calculated expected value E into equation (15) to calculate the weight value w for each type of consent stamp.

[0107] Furthermore, the weight calculation unit 22 calculates the expected value E from the histogram for each of the number of times the agree button is entered and the number of times the operation to view the introduction information is entered, and by substituting the obtained expected value E into equation (15), calculates the type-specific weight value w for each of the number of times the agree button is entered and the operation to view the introduction information is entered.

[0108] Then, the weight calculation unit 22 applies the type-specific weight value w to each edge ED of the social graph to calculate the weight value W of the edge ED.

[0109] Hereinafter, User U １ User U ２ The node will be explained using an example where user U makes a comment. ２ User U １ When the user U makes a comment on the presented information, the weight calculation unit 22 １ Node N corresponding to １ and user U ２ Node N corresponding to ２Edge ED between １ Here, the weight calculation unit 22 calculates the weight of the edge ED １ A directed graph or an undirected graph may be used to build the graph. A directed graph can distinguish between the commenter and the commentee, so a directed graph can perform more accurate user profiling than an undirected graph.

[0110] Also, Edge ED １ When a directed graph is used as the weight calculation unit 22, the weight calculation unit 22 distinguishes between comments and agreement stamps, for example, and calculates the edge ED １ In this case, the weight calculation unit 22 may calculate the edge ED between the comment and the agreement stamp based on the direction of recognition. １ In this way, by determining the direction of the edge for each type of response operation, accurate user profiling is possible. ｋ The edge ED is such that the node of the user who has commented or stamped the agreement is the start point and the node of the user who has commented or stamped the agreement is the end point. １ Just put it up.

[0111] 4 is a flowchart showing an example of processing performed by the information processing system 100. Apart from this flowchart, the acquisition unit 21 constantly acquires log data from the terminal 4 and stores the log data in the log storage unit 31.

[0112] In step S1, the weight calculation unit 22 acquires log data to be processed from the log storage unit 31. For example, if the flowchart of Fig. 4 is performed at regular intervals (for example, once a day), the weight calculation unit 22 may acquire unprocessed log data from the log storage unit 31. Note that the regular interval is not limited to one day, and any suitable value such as two days, one week, one month, or one year may be used.

[0113] Next, in step S2, the weight calculation unit 22 performs the above-described preprocessing on the log data acquired in step S1, and calculates the weights of each user U ｋ Hereinafter, the weight calculation unit 22 treats this statistical value as the number of times a response operation is input.

[0114] Next, in step S3, the weight calculation unit 22 creates a social graph from the log data acquired in step S1. For example, the weight calculation unit 22 calculates the weight of a user U in the current social graph. ｋ When adding a new user U ｋ The weight calculation unit 22 then adds an edge ED to the added node. ｉ In this case, the weight calculation unit 22 identifies, from the log data, to which user the user of the added node has performed a response operation or from which user a response operation has been performed. The weight calculation unit 22 establishes an edge ED between the added node and the identified user. ｉ Just put it up.

[0115] Next, in step S4, the weight calculation unit 22 uses the first or second method described above to calculate the weight value W. A detailed flowchart of this process will be described later.

[0116] Next, in step S5, the profiling unit 23 executes the above-mentioned user profiling process, the detailed flowchart of which will be described later.

[0117] Next, in step S6, the output unit 24 transmits the profiling results to the user terminal or the management terminal.

[0118] 5 is a flowchart showing the details of the calculation process of the weight value W when the first method is adopted. First, in step S11, the weight calculation unit 22 calculates the weight value W of the target edge ED ｘ Here, x=1, 2, ..., i, ..., N. The target edge ED ｘ is any one edge ED that constitutes the social graph. ｉ Refers to...

[0119] Next, in step S12, the weight calculation unit 22 calculates the target edge ED ｘ For example, the weight value w for the edge of interest ED is specified. ｘIf the edge ED is established based on the number of times comments and agreement stamps are input, the weight calculation unit 22 identifies the type-specific weight value w of the comment as "3" and the type-specific weight value w of the agreement stamp as "2".

[0120] Next, in step S13, the weight calculation unit 22 calculates the weight value w by type determined in step S12 and the target edge ED ｘ and log data related to the target edge ED ｘ For example, the weight value W of the target edge ED ｘ If the log data relating to the target edge ED indicates that the number of times a comment is input is three and the number of times an agreement stamp is input is two, the weight calculation unit 22 calculates the weight of the target edge ED ｘ The weight value W is calculated as 3×3+2×2=13.

[0121] Next, in step S14, the weight calculation unit 22 calculates the weight of all edges ED ｉ It is determined whether the calculation of the weight value W of all edges ED has been completed. ｉ If the calculation of the weight values ​​W for all edges ED has not been completed (NO in step S14), the process returns to step S11. On the other hand, if the calculation of the weight values ​​W for all edges ED has been completed (YES in step S14), the process proceeds to step S5 in FIG.

[0122] 6 is a flowchart showing details of the calculation process of the weight value W when the second method is adopted. In step S21, the weight calculation unit 22 classifies the log data by type of response operation.

[0123] Next, in step S22, the weight calculation unit 22 calculates a histogram for each type of response operation using the classified log data. Note that, when calculating the histogram, the weight calculation unit 22 may use the log data added this time, may use all the log data stored in the log storage unit 31, or may use log data from a certain period of time from the present to the past.

[0124] Next, in step S23, the weight calculation unit 22 calculates the average value (expected value E) of the histogram for each type of response operation.

[0125] Next, in step S24, the weight calculation unit 22 substitutes the expected values ​​E of the histograms calculated for each type in step S23 into equation (15) to calculate type-specific weight values ​​w. This calculates type-specific weight values ​​w for each of the viewing operations of the comment, the agree button, and the agree stamp.

[0126] The process of step S25 is the same as step S11 in Fig. 5. Next, in step S26, the weight calculation unit 22 calculates a weight value W of the target edge EDx using the type-specific weight value w calculated in step S24 and the comments, the agree button, the agree stamp, and the number of inputs of the browsing operation related to the target edge EDx.

[0127] The process of step S27 is the same as step S14 in Fig. 5. As a result, the weight values ​​W of all edges ED that make up the social graph are calculated.

[0128] 7 is a flowchart showing the details of a first example of the profiling process. The profiling process of the first example calculates the betweenness centrality of a user as the degree of influence.

[0129] In step S31, the profiling unit 23 detects the shortest route for every combination of any two nodes from the social graph.

[0130] Next, in step S32, the profiling unit 23 identifies users located at intermediate nodes for each of all the shortest routes.

[0131] Next, in step S33, the profiling unit 23 calculates the number of times each user identified in step S32 appears as an intermediate node in all the shortest routes.

[0132] Next, in step S34, the profiling unit 23 calculates the number of appearances calculated for each user in step S33 as the influence of each user, thereby calculating the betweenness centrality of each user.

[0133] 8 is a flowchart showing the details of a second example of the profiling process. The profiling process of the second example calculates the influence of each user using the number of edges ED connected to each node and the weight value W of the edges ED.

[0134] In step S41, the profiling unit 23 identifies the number of edges ED connected to each node in the social graph.

[0135] Next, in step S42, the profiling unit 23 calculates the influence of each user by multiplying the number of edges ED connected to each node by the weight value W of the edges ED.

[0136] It should be noted that both the profiling process shown in FIG. 7 and the profiling process shown in FIG. 8 may be executed.

[0137] (First Example of Application Scenarios) A first example of application scenario of user profiling in the present disclosure will be described below. In the first example, the processing of the present disclosure is applied to a social networking app that revitalizes local communities. Users in this social networking app are people who belong to a specific community (for example, residents of a town). Therefore, it is considered that there are no influencers such as celebrities. Therefore, it is difficult to build a social graph based on the follow function from this social networking app.

[0138] On the other hand, the present disclosure makes it possible to build a social graph even without a follow function. Identifying central figures within a community is important for revitalizing the community. The present disclosure can identify users with high betweenness centrality as central figures in a community through user profiling. Detailed information, such as information about local events, is then provided to the central figures. This allows local event information to be effectively disseminated to various users within the community through the central figures' daily lives. In this way, the present disclosure can perform effective user profiling and contribute to regional revitalization, even for SNS apps that do not have a follow function, because a social graph is created based on response operations.

[0139] A specific process of the first example will be described below. A resident of a local community uses a smartphone as the terminal 4. A social networking app designed to revitalize the local community is installed on the terminal 4. The acquisition unit 21 acquires log data of the user's SNS app. The log storage unit 31 is, for example, a database built on the cloud. The log storage unit 31 stores the log data acquired by the acquisition unit 21 from the terminal 4. User profiling is performed in batch processing, for example, at 11:00 PM every day. The weight calculation unit 22 calculates the total number of inputs per day for each type of response operation for the log data acquired from the log storage unit 31. Furthermore, the weight calculation unit 22 calculates a type-specific weight value w for each type of response operation, and, as described above, calculates the weight value W of each edge ED using the type-specific weight value w. This allows for a social graph that reflects the relationships between the residents in the community to be obtained.

[0140] In this social graph, users with high betweenness centrality often appear as intermediate nodes on the shortest route between any two users. The profiling unit 23 determines whether each user is a central figure in the local community based on the number of times each user appears at intermediate nodes. The output unit 24 notifies the terminal of the organization manager of the results of user profiling. From the results of user profiling, the manager can estimate which users in the local community have high betweenness centrality. Therefore, the manager notifies event information for the local community starting from users with high betweenness centrality. This allows event information to be spread effectively.

[0141] (Second Example of Application Scene) The present disclosure can also be applied to an office, since an office can be considered as a community made up of multiple users.

[0142] In the second example, the present disclosure is applied to a service that activates office employees. The users of this service are employees who work in a specific office. Therefore, as in the first example, it is thought that influencers such as celebrities do not exist, and it is difficult to build a social graph based on the follow function.

[0143] To revitalize employees in an office, it is important to identify key employees. This disclosure makes it possible to identify individuals with high betweenness centrality. Then, analyzing the behavior of key employees and encouraging similar behavior to other employees will lead to revitalization of the entire office. Examples of the behavior of key employees include working hours, frequency of conversations with other users in real space, and time spent using chat apps.

[0144] The user profiling of the present disclosure is useful in that it can extract not only people with high betweenness centrality, but also people who are not conspicuous but are central in the office.

[0145] A specific process of the second example will be described below. An employee working in an office uses a smartphone as terminal 4. A social networking app designed to stimulate interpersonal relationships in the office is installed on terminal 4. The acquisition unit 21 acquires the user's log data for this SNS app. As in the first example, batch processing is then executed every day at 11:00 PM. The weight calculation unit 22 calculates the total number of inputs per day for each type of response operation for the log data acquired from the log storage unit 31. Furthermore, the weight calculation unit 22 calculates a type-specific weight value w for each type of response operation, and, as described above, calculates the weight value W of each edge ED using the type-specific weight value w. This results in a social graph that reflects the relationships between employees in the office.

[0146] In this social graph, users with high betweenness centrality often appear as intermediate nodes on the shortest route between any two employees. The profiling unit 23 determines whether each employee is a central figure in the office based on the number of times each employee appears in intermediate nodes. The output unit 24 notifies the terminal of the manager of the organization, who aims to revitalize human relationships in the office, of the results of user profiling. Therefore, the manager can estimate which employees are users with high betweenness centrality from the results of employee profiling. Therefore, the manager analyzes the behavior of employees with high betweenness centrality and recommends similar behavior to other employees. This helps revitalize the entire office.

[0147] Thus, according to this embodiment, the higher the psychological barrier a second user has in responding to the information presented by the first user, the larger the weight value of the edge corresponding to the first user and the second user is calculated, which makes it possible to perform user profiling that takes into account the level of the psychological barrier in the user's response.

[0148] The information device of the present disclosure is useful for user profiling.

Claims

1. An information processing device comprising: an acquisition unit that acquires log data that records a second user's response operations to presented information presented by a first user; a weight calculation unit that calculates, based on the log data, weight values ​​of the edges in a graph that represents the relationships between users using multiple nodes corresponding to multiple users including the first user and the second user and edges connecting each node; and a profiling unit that performs user profiling based on the graph in which the weight values ​​have been calculated, wherein the weight calculation unit calculates a larger weight value for the edge corresponding to the first user and the second user the higher the psychological barrier the second user's response operation to the presented information of the first user.

2. The information processing device according to claim 1, wherein the weight calculation unit calculates the weight value using a predetermined value depending on the type of the response operation.

3. The information processing device according to claim 1 or 2, wherein the weight calculation unit classifies the log data according to the type of response operation, calculates a histogram showing the relationship between the number of response operations and the frequency of the number of times for each type of response operation based on the classified log data, calculates an average value of the histogram for each type of response operation, calculates a weight value for each type of response operation based on the reciprocal of the average value calculated for each type of response operation, and calculates the weight value using the type-specific weight value.

4. The information processing device described in claim 1 or 2, wherein the profiling unit detects the shortest route between nodes for every combination of any two different nodes in the graph based on the weight value, detects the number of times a node corresponding to one user appears as an intermediate node in the shortest route, and calculates the influence of the one user on other users based on the number of times it appears.

5. An information processing device as described in claim 1 or 2, wherein the profiling unit calculates the degree of influence of a user corresponding to a node on other users based on the number of edges connected to the node and the weight value for the edges connected to the node.

6. The information processing device according to claim 1 or 2, further comprising an output unit that outputs the results of the user profiling.

7. The information processing device according to claim 4, further comprising an output unit that generates list information in which the plurality of users are arranged in descending order of the degree of influence, and outputs the list information.

8. The information processing device according to claim 1 or 2, wherein the response operation includes inputting a comment on the presented information, inputting a button or stamp indicating agreement with the presented information, and viewing the presented information.

9. An information processing method in a computer, comprising: acquiring log data that records a second user's response operation to presented information presented by a first user; calculating, based on the log data, weight values ​​of the edges in a graph that represents the relationships between users using a plurality of nodes corresponding to a plurality of users including the first user and the second user and edges connecting each node; and performing user profiling based on the graph in which the weight values ​​have been calculated, wherein calculating the weight values ​​includes calculating larger weight values ​​for the edges corresponding to the first user and the second user as the second user's response operation to the presented information of the first user poses a higher psychological barrier.

10. An information processing program that causes a computer to perform the following steps: acquire log data that records a second user's response operations to presented information presented by a first user; calculate, based on the log data, weight values ​​of the edges in a graph that represents the relationships between users using multiple nodes corresponding to multiple users including the first user and the second user and edges connecting each node; and perform user profiling based on the graph in which the weight values ​​have been calculated; wherein calculating the weight values ​​includes calculating larger weight values ​​for the edges corresponding to the first user and the second user as the second user's response operation to the presented information of the first user poses a higher psychological barrier.

Images