Patent Application
20230098009
The present invention relates to an SNS analysis system, an SNS analysis device, an SNS analysis method, and a recording medium storing an SNS analysis program.
It is very important to predict in advance the occurrence of a crime such as a terror and prevent the occurrence in advance in order to construct a safe society. Therefore, a technology for predicting occurrence of a crime in advance is expected.
As a technology related to such a technology, PTL 1 discloses a system in which a crime prediction server that collects crime related information related to an incident is connected to a center device having a display unit that displays the crime related information. The crime prediction server in this system accesses a social networking service (SNS) server and collects pieces of posted information including a crime-related word from among pieces of posted information of ordinary people as crime related information. The crime prediction server calculates statistical data for each attribute including an occurrence point of a crime, an occurrence time, and a crime type regarding the crime related information to transmit the crime data and the map data extracted from the statistical data of the crime related information in response to a request from the center device. Then, the center device in this system superimposes the crime data for each attribute on the map data on the display unit, and plots and displays the crime data at a location relevant to the crime occurrence point on the map.
PTL 2 discloses a system that stores crime data and weather data, and determines a crime prediction by adjusting a past crime rate based on a correlation between predicted weather conditions and the crime data. The system further stores event data and determines a crime prediction by further adjusting a past crime rate based on a correlation between future events and crime data.
One of methods for predicting occurrence of a crime in advance, includes identifying a person requiring attention who is highly likely to commit a crime from communication contents related to activities on an SNS or an analysis result of an SNS account. Since a particularly highly dangerous crime is often performed systematically, it is important to identify a person requiring attention involved in an organized crime at an early stage by estimating an unknown relationship between the persons requiring attention from analysis results of activities and accounts on the SNS in order to prevent the crime in advance. The unknown relationship is, for example, a relationship in which a follow-follower relationship is not established on the SNS but an acquaintance relationship is established in the real world.
In order to estimate an unknown relationship between the persons (users) in an SNS with high accuracy, it is necessary to estimate the unknown relationship in consideration of various factors that complicatedly affect each other. Such factors include, for example, a feature of a time-series change (transition) in the content of communication performed by the person in the SNS, a feature of a time-series change in the attribute of the person, and the like. Therefore, in order to estimate the unknown relationship between the persons in the SNS with high accuracy, it is necessary to grasp and analyze the features of the time-series change regarding the activity on the SNS with high accuracy.
However, in a general system that analyzes communication performed in an SNS, a feature of a time-series change regarding contents of communication in such an SNS cannot be sufficiently grasped. Therefore, in a general system, in particular, in a case where the feature of the time-series change is an important factor in the estimation of the unknown relationship between the persons, the estimation accuracy is greatly reduced. It cannot be said that the techniques disclosed in PTLs 1 and 2 described above are sufficient to solve this problem.
A main object of the present invention is to provide an SNS analysis system and the like capable of improving the accuracy with which existence of an unknown relationship between a plurality of persons is estimated in an SNS.
An SNS analysis system according to an aspect of the present invention includes an estimation means configured to estimate, based on an estimation model representing a relationship between communication history information and attribute information related to a first plurality of persons and presence or absence of a relationship existing between the first plurality of persons and the communication history information and the attribute information related to a second plurality of persons, existence of an unknown relationship between the second plurality of persons, wherein the communication history information represents a time-series change in at least any of exchange of information between the first plurality of persons or the second plurality of persons via an SNS and transmission of information related to each other by the first plurality of persons or the second plurality of persons via the SNS, and the attribute information indicates a time-series change in attributes of the first plurality of persons or the second plurality of persons.
In another viewpoint of achieving the above object, an SNS analysis method according to an aspect of the present invention includes an information processing system estimating, based on an estimation model representing a relationship between communication history information and attribute information related to a first plurality of persons and presence or absence of a relationship existing between the first plurality of persons and the communication history information and the attribute information related to a second plurality of persons, existence of an unknown relationship between the second plurality of persons, wherein the communication history information represents a time-series change in at least any of exchange of information between the first plurality of persons or the second plurality of persons via an SNS and transmission of information related to each other by the first plurality of persons or the second plurality of persons via the SNS, and the attribute information indicates a time-series change in attributes of the first plurality of persons or the second plurality of persons.
In a further viewpoint of achieving the above object, an SNS analysis program according to an aspect of the present invention causes a computer to execute an estimation process of estimating, based on an estimation model representing a relationship between communication history information and attribute information related to a first plurality of persons and presence or absence of a relationship existing between the first plurality of persons and the communication history information and the attribute information related to a second plurality of persons, existence of an unknown relationship between the second plurality of persons, wherein the communication history information represents a time-series change in at least any of exchange of information between the first plurality of persons or the second plurality of persons via an SNS and transmission of information related to each other by the first plurality of persons or the second plurality of persons via the SNS, and the attribute information indicates a time-series change in attributes of the first plurality of persons or the second plurality of persons.
Furthermore, the present invention can also be achieved by a computer-readable non-volatile recording medium storing the SNS analysis program (computer program).
According to the present invention, it is possible to obtain an SNS analysis system or the like capable of improving the accuracy with which existence of an unknown relationship between a plurality of persons is estimated in an SNS.
A system exemplifying the example embodiment to be described later uses a learned model (also referred to as an estimation model) generated by machine learning (for example, deep learning) when estimating a target event from certain input information. Then, the system uses, for example, a graph including a node and an edge (also referred to as an edge) representing the input information. The graph changes in structure over time. The idea of the system has come when applying an algorithm capable of analyzing features of such a graph. As this algorithm, for example, the following algorithm is known.
(1) TGFN (Temporal Graph Factorization Network)
It is an algorithm that extracts a static feature that is unchanged regardless of time and a dynamic feature unique to each time from a graph whose structure changes with the lapse of time, and analyzes the extracted feature. Since this algorithm is disclosed in NPL 1, the detailed description thereof will be omitted in the example embodiment described later.
(2) STAR (Spatio-Temporal Attentive RNN)
It is an algorithm for identifying and analyzing, from a graph whose structure changes with the lapse of time, a node that is important (that is, the degree of influence on estimation is high.) on estimation of a certain event, for example, on each of a time axis and a spatial axis among nodes constituting the graph. Since this algorithm is disclosed in NPL 2, the detailed description thereof will be omitted in the example embodiment described later.
(3) Netwalk
It is an algorithm for extracting a feature amount of a node constituting a graph from the graph whose structure changes with time. Since this algorithm is disclosed in NPL 3, the detailed description thereof will be omitted in the example embodiment described later.
The disclosure exemplifying the example embodiment to be described later achieves improvement in accuracy with which a target event is estimated by applying the above-described algorithm when generating a learned model and when estimating the target event from certain input information using the learned model.
Hereinafter, example embodiments of the present invention will be described in detail with reference to the drawings.
A management terminal device 20 (also referred to as a display device) is communicably connected to the SNS analysis system 10. The management terminal device 20 is, for example, a personal computer or another information processing apparatus used when a user (hereinafter, also referred to as an “administrator”) who uses the SNS analysis system 10 inputs information to the SNS analysis system 10 or confirms information output from the SNS analysis system 10. Management terminal device 20 includes a display screen 200 that displays the information output from the SNS analysis system 10.
The SNS analysis system 10 includes an acquisition unit 11, a graph generation unit 12, a model generation unit 13, an estimation unit 14, and a display control unit 15. The graph generation unit 12, the model generation unit 13, the estimation unit 14, and the display control unit 15 are examples of a graph generation means, a model generation means, an estimation means, and a display control means, respectively.
Next, an operation in which the SNS analysis system 10 according to the present example embodiment generates or updates an estimation model 130 for estimating existence of an unknown relationship between a plurality of persons and an operation in which the SNS analysis system estimates the unknown relationship using the estimation model 130 will be described.
<Operation of Generating (Updating) Estimation Model 130>
First, an operation in which the SNS analysis system 10 according to the present example embodiment generates or updates the estimation model 130 for estimating existence of an unknown relationship between a plurality of persons in the SNS will be described.
The acquisition unit 11 acquires communication history information 100 and attribute information 103 regarding a plurality of persons (also referred to as first plurality of persons) to be learned in a predetermined period from a computer device (not illustrated) or a database via a network. Acquisition unit 11 may, for example, periodically acquire the communication history information 100 and the attribute information 103. Alternatively, for example, the acquisition unit 11 may acquire the communication history information 100 and the attribute information 103 according to an instruction input by the user via the management terminal device 20.
Acquisition unit 11 includes, for example, a communication circuit connected to one or a plurality of computer devices or databases that transmit the communication history information 100 and the attribute information 103, and a storage device that stores information acquired by the communication circuit. The storage device may be a hard disk 904 or a RAM 903 of an information processing system 900 illustrated in
The communication history information 100 is information indicating a time-series change (transition) in communication performed by a plurality of persons via the SNS. The communication history information 100 includes follow result information 101 and posted information 102.
Communication history information 100 includes SNS account information and SNS activity information of a plurality of SNS users.
The SNS account information is information related to an account of the SNS user. For example, the SNS account information includes identification information (name, nickname, ID, etc.), residential place information (address, etc.), work place information (company name, workplace address, etc.), a telephone number, an email address, and the like of the SNS user. The SNS account information is not limited thereto, and may include various pieces of information registered by the SNS user at the time of account creation.
The SNS activity information is information related to activity on the SNS performed by the SNS user via the SNS account. The SNS activity information includes, for example, the following information.
The SNS activity information is not limited thereto, and may include various pieces of information related to activity on the SNS or interaction with another user.
Note that the follow result information 101 and the posted information 102 may each include SNS account information and SNS activity information.
In the follow result information 101 illustrated in
The follow result information 101 is time-series changing information to which a follow result is added when the follow by a certain person is performed.
Although the posted information 102 illustrated in
The posted information 102 is information that changes in time series and to which a posted result is added when posting to the SNS by a certain person is performed.
The organization to which a person belongs and the status in the organization in the attribute information 103 are changed when the situation in which the person belongs to the organization changes, and the criminal record is added when the person newly commits a crime, so that the attribute information 103 is information that changes in time series.
The acquisition unit 11 stores the follow result information 101, the posted information 102, and the attribute information 103 acquired as described above in a storage device (not illustrated) (for example, a memory, a hard disk, or the like).
The graph generation unit 12 illustrated in
Each node in the graph 120 includes attribute information of a person. More specifically, the nodes in graph 120 include the attribute information 103. Therefore, each node is represented by a multi-dimensional function including the item (for example, an organization to which a person belongs, a status in the organization, a criminal record, and the like.) included in the attribute information 103 as an element with time t as a variable. A multi-dimensional function representing a node is stored in a storage device (not illustrated) (for example, the hard disk 904 or the RAM 903) in association with information indicated by the node.
More specifically, each edge in the graph 120 is associated with the follow result information 101 and the posted information 102. For example, an edge connecting a node indicating the person A and a node indicating the person B represents a result of following the person A by the person B indicated by the follow result information 101, and is represented by a function fAB (t) illustrated in
The relevance between the content posted by the person A and the content posted by the person B indicated by the posted information 102 is also represented by the function fAB (t) illustrated in
In a case where a posted content included in the posted information 102 is represented by a voice, the graph generation unit 12 may convert the posted content into a text using, for example, an existing voice recognition technique, and perform the above-described processing on the text for obtaining similarity. In a case where a posted content included in the posted information 102 is represented by an image, the graph generation unit 12 may convert the posted content into text using, for example, an existing image recognition technique, and perform the above-described processing on the text for obtaining similarity.
As described above, the function such as the function fAB (t) representing each edge is a multi-dimensional function including the item (for example, follow relationship) included in the follow result information 101 and the item (for example, the relevance of the posted content) included in the posted information 102 as elements with time t as a variable. A multi-dimensional function representing an edge is stored in a storage device (not illustrated) (for example, the hard disk 904 or the RAM 903) in association with the edge.
The graph generation unit 12 further assigns a label to the graph 120 for teacher data generated for a predetermined period and used when the model generation unit 13 described later performs machine learning. The graph generation unit 12 sets, as the label, the presence or absence of a relationship, existing between a plurality of persons, that is unknown in the predetermined period but is known after the predetermined period.
(1) the person A (LEADER of a criminal organization) posted a statement urging execution of the terror.
(2) the person E followed the statement of execution of the terror by the person A.
(3) the person F posted a content related to execution of the terror.
(4) the person I has posted a content related to the content posted by the person F (However, there is no direct follow from the person I to the person F.).
Based on communication history information 100 indicating a time-series change in communication performed by the plurality of persons via the SNS, the graph generation unit 12 generates the graph 120 indicating a time-series change in the communication and used as teacher data. A graph 120-t1 and a graph 1204 illustrated in
Note that the graph generation unit 12 may generate (draw) a function graph instead of the graph structure data as described above. In this case, for example, the graph generation unit 12 may generate a graph (function) in which a horizontal axis represents time (date and time) and a vertical axis represents an index indicating an SNS activity.
In the example illustrated in
Such labeling may be performed, for example, by the user determining existence of an unknown relationship based on the content of the time-series change in communication performed via the SNS indicated by the communication history information 100 and the fact that the terror incident in which the person I participated occurred. Alternatively, the graph generation unit 12 may perform such labeling according to a predetermined labeling criterion based on the content of the time-series change in the communication performed via the SNS indicated by the communication history information 100 and the information indicating the fact that the terror incident in which the person I participated occurred. The graph generation unit 12 stores the configuration of the graph 120 to which a label is assigned as described above in the storage device. The graph generation unit 12 outputs the labeled graph 120, as teacher data, to the model generation unit 13.
Using the labeled graph 120, as teacher data, input from the graph generation unit 12, the model generation unit 13 generates the estimation model 130 (learned model) to be used when an estimation unit 14 described later estimates existence of an unknown relationship between the persons. The model generation unit 13 performs machine learning for generating the estimation model 130 (learned model) using the above-described teacher data by the processor.
Specifically, the model generation unit 13 extracts, from the input graph 120, features of time-series changes regarding communication between a plurality of persons via the SNS and attributes of the plurality of persons using a predetermined algorithm. The model generation unit 13 can use, for example, TGFN, STAR, Netwalk, or the like described above as the predetermined algorithm.
By using, for example, TGFN, the model generation unit 13 extracts, from the graph 120, static features and dynamic features that change with time regarding communication between a plurality of persons via the SNS and attributes of the plurality of persons. Alternatively, for example, by using STAR, the model generation unit 13 extracts a node that is important (that is, the degree of influence on estimation is high.) in estimating existence of an unknown relationship between the persons on each of a time axis (a viewpoint over a certain period of time) and a spatial axis (a viewpoint focusing on individual times). Alternatively, the model generation unit 13 extracts the feature amount of the node from the graph 120 by using, for example, Netwalk. When Netwalk is used, the model generation unit 13 may combine it with an existing prediction algorithm such as gradient boosting, for example.
Next, in the process of performing machine learning using the above-described teacher data, the model generation unit 13 determines an explanatory variable related to existence of the unknown relationship between the persons from the result of extracting the feature from the graph 120 as described above. A specific example of the explanatory variable will be described later. Specifically, the result of extracting the feature from the graph 120 is the static features and the dynamic features regarding communication between a plurality of persons via the SNS and attributes of the plurality of persons, or feature amounts of nodes. Then, the model generation unit 13 generates the estimation model 130 including a criterion for estimating existence of the unknown relationship between the persons based on the value of the explanatory variable. The model generation unit 13 determines the criterion by performing machine learning on the relationship between the value of the explanatory variable and the value of the label in the teacher data.
Model generation unit 13 determines an explanatory variable related to, for example, a time-series change in communication activity via the SNS, the explanatory variable being indicated by communication history information 100. The explanatory variable represents, for example, a relationship between a follower and a follow destination, a communication content, a place where communication is performed, and the like, but is not limited thereto. For example, the model generation unit 13 determines an explanatory variable related to, for example, the time-series change in the attribute of the person indicated by the attribute information 103. The explanatory variable represents, for example, an organization to which the person belongs, a status in the organization, and the like, but is not limited thereto.
When determining the explanatory variable as described above, the model generation unit 13 also determines the degree of importance on estimation of existence of the unknown relationship between the persons (contribution to the estimation result) for each of the plurality of explanatory variables. The model generation unit 13 may weight the value of each explanatory variable by the degree of importance of the explanatory variable in the criterion for estimating existence of the unknown relationship between the persons described above. At this time, the model generation unit 13 may determine a different degree of importance for each target person from a difference in feature between the target persons related to the communication history information 100 and the attribute information 103 with respect to the same explanatory variable. That is, for example, with respect to a certain explanatory variable, the model generation unit 13 may set the importance on estimation of existence of the unknown relationship between the person A and the person B to be high, and may set the importance on estimation of existence of the unknown relationship between the person C and the person D to be low.
The model generation unit 13 stores the estimation model 130 generated or updated as described above in a non-volatile storage device (not illustrated). The model generation unit 13 can gradually improve the estimation accuracy by updating (also referred to as relearning) the estimation model 130, for example, every predetermined time.
Next, an operation (processing) of generating (performing machine learning) the estimation model 130 by the SNS analysis system 10 according to the present example embodiment will be described in detail with reference to a flowchart of
The acquisition unit 11 acquires, from the outside, the communication history information 100 and the attribute information 103 related to a plurality of persons used as teacher data (step S101). The graph generation unit 12 generates (updates) the graph 120 by using the communication history information 100 and the attribute information 103 acquired by the acquisition unit 11, and assigns, to the graph 120, the presence or absence of an unknown relationship between the persons as a label (step S102).
Using a predetermined algorithm, the model generation unit 13 extracts, from the graph 120 generated by the graph generation unit 12, a feature of a time-series change in the follow and transmission of related information on the SNS between the persons, and a feature of the attribute (step S103). The model generation unit 13 determines an explanatory variable related to existence of an unknown relationship between the persons based on the extraction result (step S104).
The model generation unit 13 determines the degree of importance on estimation of existence of the unknown relationship between the persons for each explanatory variable using a predetermined algorithm, generates (updates) the estimation model 130 including the explanatory variable (step S105), and ends the entire processing.
<Operation of Estimating Existence of Unknown Relationship Between a Plurality of Persons>
Next, an operation in which the SNS analysis system 10 according to the present example embodiment estimates existence of an unknown relationship between a plurality of persons using the generated or updated estimation model 130 will be described.
The acquisition unit 11 acquires the communication history information 100 and the attribute information 103 from an external device (not illustrated) as in the case where the SNS analysis system 10 generates the estimation model 130. The acquisition unit 11 does not acquire these pieces of information as the above-described teacher data, but acquires these pieces of information as data for estimating existence of an unknown relationship between the persons.
For example, as described above, it is assumed that the estimation model 130 is generated based on the communication history information 100 and the attribute information 103 regarding a plurality of persons (also referred to as a first plurality of persons) involved in a certain crime. In this case, the acquisition unit 11 acquires the communication history information 100 and the attribute information 103 regarding another plurality of persons (also referred to as a second plurality of persons) who is dangerous to commit a crime according to an instruction input by the user via the management terminal device 20, for example. The form of the communication history information 100 and the attribute information 103 related to a plurality of persons to be estimated are similar to that of the communication history information 100 and the attribute information 103 used for generating the estimation model 130 illustrated in
The graph generation unit 12 generates the graph 120 representing the communication history information 100 and the attribute information 103 about a plurality of persons to be estimated. Note that the configuration of the graph 120 is as described above with reference to
The estimation unit 14 illustrated in
As in the case where the model generation unit 13 generates or updates the estimation model 130, the estimation unit 14 extracts, from the graph 120 input from the graph generation unit 12, the feature of the time-series change regarding the communication between the plurality of persons via the SNS and the attributes of the plurality of persons. At this time, the estimation unit 14 may use a predetermined algorithm such as TGFN, STAR, or Netwalk described above, for example.
The estimation unit 14 obtains a value of the explanatory variable identified by the estimation model 130 in the graph 120 based on the feature extracted from the graph 120. The estimation unit 14 collates the obtained values of the explanatory variables with a criterion for estimating existence of an unknown relationship between a plurality of persons included in the estimation model 130, thereby estimating existence of the unknown relationship. The features extracted from the graph 120 include, for example, a degree of similarity of persons in the attribute information 103, a degree of similarity of each other's follow results in the follow result information 101, and a time-series feature regarding a time-series change in the SNS activity. The time-series features include, for example, posting timings of posted contents with the same content being similar, following a certain SNS user at the same time, defollowing a certain SNS user at the same time, or the like. Note that the feature extracted from the graph 120 is not limited thereto.
The estimation unit 14 outputs, to the display control unit 15, the result of estimating existence of the unknown relationship between the plurality of persons and information indicating the reason for estimating the existence. The information indicating the reason for estimating the existence is, for example, the value of the explanatory variable in the graph 120 for estimating existence of the unknown relationship, the degree of importance of the explanatory variable, and the like.
The display control unit 15 displays, on the display screen 200 of the management terminal device 20, the result of estimating existence of the unknown relationship between the plurality of persons and the information indicating the reason for estimating the existence, which are input from the estimation unit 14. That is, the display control unit 15 causes the management terminal device 20 to display the estimation result and the estimation reason by the estimation unit 14 on the display screen 200 of the management terminal device 20.
The display screen 200 illustrated in
1. A content highly related to the content posted by the person K who follows the post suggesting the terror by the person A (leader of the organization P) is posted by the person L.
(The estimation reason in this case is that “the posted content is similar to the post following the person requiring attention”. That is, in this case, the estimation reason is the relationship between the similarity of the posted content with the post following the person requiring attention and existence of an unknown relationship.)
2. In the above 1, the posting by the person K and the posting by the person L are performed at substantially the same time.
(The estimation reason in this case is that “the posting times are similar”. That is, in this case, the relationship between the similarity of the posting times and the existence of an unknown relationship is the estimation reason.)
3. In the above 1, both the posting by the person K and the posting by the person L are performed from the region Z.
(The estimation reason in this case is that “the posting places are similar”. That is, in this case, the relationship between the similarity of the posting places and the existence of an unknown relationship is the estimation reason.)
The SNS analysis system 10 visibly presents the explanatory variable as the estimation reason to the administrator, thereby achieving an effect of improving the explanatory property. The SNS analysis system 10 can also visibly present the relationship between the explanatory variables contributing to the estimation as the reason for estimating existence of the unknown relationship. At this time, the SNS analysis system 10 may visibly present the estimation reason by a mode that is not a natural language sentence as long as the estimation reason can be visually recognized.
Although not illustrated in
The display screen 200 illustrated in
In the case of the example illustrated in
Next, an operation (processing) of estimating existence of an unknown relationship between a plurality of persons by the SNS analysis system 10 according to the present example embodiment will be described in detail with reference to a flowchart of
The acquisition unit 11 acquires the communication history information 100 and the attribute information 103 to be estimated from the outside (step S201). The graph generation unit 12 generates (updates) the graph 120 using the communication history information 100 and the attribute information 103 acquired (step S202).
The estimation unit 14 extracts, from the graph 120 generated by the graph generation unit 12, a feature of a time-series change in the follow and transmission of related information on the SNS between the persons and a feature of the attribute by using a predetermined algorithm (step S203).
The estimation unit 14 estimates existence of the unknown relationship between the persons based on the feature extraction result from the graph 120 and the estimation model 130, and identifies the reason for estimating the existence (step S204). The display control unit 15 displays the estimation result of the existence of the unknown relationship between the plurality of persons and the reason for estimating the existence by the estimation unit 14 on the display screen 200 of the management terminal device 20 (step S205), and the entire process ends.
The SNS analysis system 10 according to the present example embodiment can improve accuracy with which existence of an unknown relationship between a plurality of persons is estimated in the SNS. This is because the SNS analysis system 10 estimates existence of the unknown relationship between the plurality of persons based on the estimation model 130 generated by using the result of extracting the feature of the time-series change from the information related to the communication between the plurality of persons via the SNS.
Hereinafter, effects achieved by the SNS analysis system 10 according to the present example embodiment will be described in detail.
In order to predict the occurrence of a crime in advance, it is necessary to estimate the relationship in consideration of various factors that complicatedly affect each other in order to estimate an unknown relationship between the persons in the SNS with high accuracy. Such factors include, for example, a feature of a time-series change in the content of communication performed by the person in the SNS, a feature of a time-series change in the attribute of the person, and the like. Therefore, in order to estimate the unknown relationship between the persons in the SNS with high accuracy, it is necessary to analyze the feature of the time-series change related to communication in the SNS with high accuracy. However, in a general system that analyzes communication performed in an SNS, there is a problem that high estimation accuracy cannot be obtained because a feature of a time-series change related to communication in such an SNS cannot be sufficiently grasped.
For such a problem, the SNS analysis system 10 according to the present example embodiment includes the estimation model 130 and the estimation unit 14, and operates as described above with reference to
The SNS analysis system 10 according to the present example embodiment generates the graph 120 that represents the communication history information 100 and the attribute information 103, includes nodes and edges, and has a structure changing in time series. Then, the SNS analysis system 10 uses the above-described algorithm such as TGFN, STAR, or Netwalk capable of extracting and analyzing the feature of the generated graph 120, thereby achieving grasping the feature of the time-series change regarding the communication in the SNS with high accuracy. Thus, the SNS analysis system 10 can improve the accuracy with which the unknown relationship between the persons is estimated in the SNS.
In the process of generating the estimation model 130, the SNS analysis system 10 according to the present example embodiment determines explanatory variables regarding the estimation of the unknown relationship between the persons, and further determines the degree of importance (contribution) on estimation of the unknown relationship between the persons for each explanatory variable. Then, the SNS analysis system 10 weights the explanatory variable by its degree of importance to estimate the unknown relationship between the persons. As a result, since the SNS analysis system 10 performs estimation in which the feature of communication in the SNS are captured accurately as compared with, for example, a case where estimation is performed without calculating the degree of importance, accuracy with which an unknown relationship between the persons in the SNS is estimated can be enhanced.
When generating the estimation model 130, the model generation unit 13 may exclude a node (person) having an influence on the relationship between the communication history information 100 and the attribute information 103 related to a plurality of persons and the presence or absence of the relationship existing between the plurality of persons smaller than the reference. That is, when estimating a relationship existing between a plurality of persons, the model generation unit 13 may ignore a person who does not affect the estimation and is obviously unrelated to the plurality of persons as a node that is noise. The model generation unit 13 can use, for example, a Graph Denoising Policy Network (GDPNet) as an existing algorithm for excluding a node that is such noise. Then, the SNS analysis system 10 can reduce the processing load by excluding a node that is noise.
In a general system that estimates an event using a learned model, an estimation process is a black box, and only an estimation result is presented without presenting an estimation reason. Therefore, it is difficult for a user to grasp the basis of the estimation result output by the system. On the other hand, the SNS analysis system 10 according to the present example embodiment displays the reason for estimating the unknown relationship between the persons in the SNS based on the value of the explanatory variable on the display screen 200 of the management terminal device 20, for example, as illustrated in
Communication, via the SNS, to be analyzed by the SNS analysis system 10 is not limited to communication between the persons requiring attention who may commit a crime. For example, in a criminal investigation, the SNS analysis system 10 may estimate an unknown relationship existing between a crime victim and a certain person.
The estimation model 31 represents a relationship between communication history information 310 and attribute information 313 regarding the first plurality of persons (persons to be targets of machine learning) and the presence or absence 314 of a relationship existing between the first plurality of persons. As in the estimation model 130 according to the first example embodiment, for example, the estimation model 31 is a learned model representing a result of performing machine learning on a relationship between the communication history information 310, the attribute information 313, and the presence or absence 314 of a relationship existing between the first plurality of persons.
Communication history information 310 represents a time-series change in at least any of exchange of information between the first plurality of persons via the SNS and transmission of information related to each other by the first plurality of persons via the SNS. The communication history information 310 may be, for example, information similar to the communication history information 100 described with reference to
The attribute information 313 represents a time-series change in the attributes of the first plurality of persons, and may be, for example, information similar to the attribute information 103 described with reference to
The estimation unit 32 estimates existence of the unknown relationship between the second plurality of persons based on communication history information 300 and attribute information 303 related to the second plurality of persons (persons for which the unknown relationship between the persons is to be estimated), and the estimation model 31.
When estimating existence of the unknown relationship between the persons, the estimation unit 32 extracts the feature of the time-series change regarding the communication and the attribute of the person in the SNS from the communication history information 300 and the attribute information 303, as in the estimation unit 14 according to the first example embodiment. At this time, the estimation unit 32 can use the predetermined algorithm (TGFN, STAR, Netwalk, etc.) described in the first example embodiment.
The SNS analysis system 30 according to the present example embodiment can improve accuracy with which existence of an unknown relationship between a plurality of persons in the SNS is estimated. This is because the SNS analysis system 30 estimates existence of the unknown relationship between the plurality of persons based on the estimation model 31 generated by using the result of extracting the feature of the time-series change from the information related to the communication between the plurality of persons via the SNS.
<Hardware Configuration Example>
Each unit in the SNS analysis system 10 illustrated in
The division of each unit illustrated in these drawings is a configuration for convenience of description, and various configurations can be assumed at the time of implementation. An example of a hardware environment in this case will be described with reference to
The information processing system 900 illustrated in
That is, the information processing system 900 including the above-described components is a general computer to which these components are connected via the bus 906. The information processing system 900 may include a plurality of CPUs 901 or may include a CPU 901 configured by a plurality of cores. The information processing system 900 may include a GPU (Graphical_Processing_Unit) (not illustrated) in addition to the CPU 901.
Then, the present invention described using the above-described example embodiment as an example supplies a computer program capable of achieving the following functions to the information processing system 900 illustrated in
In the above case, a general procedure can be used at present as a method of supplying the computer program into the hardware. Examples of the procedure include a method of installing the program in the apparatus via various recording media 907 such as a CD-ROM, a method of downloading the program from the outside via a communication line such as the Internet, and the like. In such a case, the present invention can be understood to be configured by a code constituting the computer program or the recording medium 907 storing the code.
The present invention is described above using the above-described example embodiments as exemplary examples. However, the present invention is not limited to the above-described example embodiments. That is, the present invention can have various aspects that can be understood by those skilled in the art within the scope of the present invention.
Note that part or all of each example embodiments described above can also be described as the following Supplementary Notes. However, the present invention exemplarily described by the above-described example embodiments is not limited to the following.
(Supplementary Note 1)
An SNS analysis system including
an estimation means configured to estimate, based on an estimation model representing a relationship between communication history information and attribute information related to a first plurality of persons and presence or absence of a relationship existing between the first plurality of persons and the communication history information and the attribute information related to a second plurality of persons, existence of an unknown relationship between the second plurality of persons, wherein
the communication history information represents a time-series change in at least any of exchange of information between the first plurality of persons or the second plurality of persons via an SNS and transmission of information related to each other by the first plurality of persons or the second plurality of persons via the SNS, and
the attribute information indicates a time-series change in attributes of the first plurality of persons or the second plurality of persons.
(Supplementary Note 2)
The SNS analysis system according to Supplementary Note 1, further including
a display control means configured to control a display device to display a reason for estimating existence of an unknown relationship between the second plurality of persons.
(Supplementary Note 3)
The SNS analysis system according to Supplementary Note 2, wherein
the communication history information indicates a follow result of an SNS between the first plurality of persons or the second plurality of persons.
(Supplementary Note 4)
The SNS analysis system according to Supplementary Note 2 or 3, wherein
the communication history information includes information posted on an SNS by the first plurality of persons or the second plurality of persons.
(Supplementary Note 5)
The SNS analysis system according to Supplementary Note 4, wherein
the posted information includes at least any of a text, a voice, and an image.
(Supplementary Note 6)
The SNS analysis system according to any one of Supplementary Notes 2 to 5, wherein
the communication history information indicates locations where the first plurality of persons or the second plurality of persons has performed communication by operating terminal devices.
(Supplementary Note 7)
The SNS analysis system according to any one of Supplementary Notes 2 to 6, wherein
the attribute information represents at least any of a criminal record of each of the first plurality of persons or the second plurality of persons and a situation in which the each person belongs to an organization.
(Supplementary Note 8)
The SNS analysis system according to any one of Supplementary Notes 2 to 7, further including
a graph generation means configured to generate a graph representing the communication history information.
(Supplementary Note 9)
The SNS analysis system according to Supplementary Note 8, wherein
the graph includes a node representing each of the first plurality of persons or the second plurality of persons and an edge representing each relationship between the first plurality of persons or the second plurality of persons via an SNS.
(Supplementary Note 10)
The SNS analysis system according to Supplementary Note 9, further including
a model generation means configured to generate the estimation model based on communication history information and attribute information related to the first plurality of persons in a predetermined period, and presence or absence of a relationship, existing between the first plurality of persons, that is unknown in the predetermined period but is known after the predetermined period.
(Supplementary Note 11)
The SNS analysis system according to Supplementary Note 10, wherein
the model generation means extracts a feature of a time-series change in a relationship between the first plurality of persons via an SNS using a predetermined algorithm from the graph to which presence or absence of a relationship, existing between the first plurality of persons, that is unknown in the predetermined period is assigned as a label, and then determines an explanatory variable of existence of an unknown relationship between the first plurality of persons based on a result of the extraction to generate the estimation model including the explanatory variable.
(Supplementary Note 12)
The SNS analysis system according to Supplementary Note 11, wherein
the model generation means generates the estimation model excluding a node in which an influence on a relationship between the communication history information and the attribute information related to the first plurality of persons and presence or absence of a relationship existing between the first plurality of persons is smaller than a reference.
(Supplementary Note 13)
The SNS analysis system according to Supplementary Note 11 or 12, wherein
the graph generation means generates the graph including the attribute information, and
the model generation means determines, from the graph, the explanatory variable related to an attribute of each of the first plurality of persons.
(Supplementary Note 14)
The SNS analysis system according to any one of Supplementary Notes 11 to 13, wherein
the model generation means determines a degree of importance on estimation of existence of the unknown relationship for each of a plurality of the explanatory variables, and
the estimation means estimates existence of the unknown relationship based on the degree of importance.
(Supplementary Note 15)
The SNS analysis system according to Supplementary Note 14, wherein
the model generation means determines the degree of importance different for each of the first plurality of persons for the same explanatory variable.
(Supplementary Note 16)
The SNS analysis system according to Supplementary Note 14 or 15, wherein
the display control means causes the display device to display names of the explanatory variables side by side in descending order of the degree of importance and display the estimation reason in a mode of displaying values of the explanatory variables.
(Supplementary Note 17)
An SNS analysis device including
an estimation means configured to estimate, based on an estimation model representing a relationship between communication history information and attribute information related to a first plurality of persons and presence or absence of a relationship existing between the first plurality of persons and the communication history information and the attribute information related to a second plurality of persons, existence of an unknown relationship between the second plurality of persons, wherein
the communication history information represents a time-series change in at least any of exchange of information between the first plurality of persons or the second plurality of persons via an SNS and transmission of information related to each other by the first plurality of persons or the second plurality of persons via the SNS, and
the attribute information indicates a time-series change in attributes of the first plurality of persons or the second plurality of persons.
(Supplementary Note 18)
An SNS analysis method including
an information processing system estimating, based on an estimation model representing a relationship between communication history information and attribute information related to a first plurality of persons and presence or absence of a relationship existing between the first plurality of persons and the communication history information and the attribute information related to a second plurality of persons, existence of an unknown relationship between the second plurality of persons, wherein
the communication history information represents a time-series change in at least any of exchange of information between the first plurality of persons or the second plurality of persons via an SNS and transmission of information related to each other by the first plurality of persons or the second plurality of persons via the SNS, and
the attribute information indicates a time-series change in attributes of the first plurality of persons or the second plurality of persons.
(Supplementary Note 19)
A recording medium storing an SNS analysis program for causing a computer to execute
an estimation process of estimating, based on an estimation model representing a relationship between communication history information and attribute information related to a first plurality of persons and presence or absence of a relationship existing between the first plurality of persons and the communication history information and the attribute information related to a second plurality of persons, existence of an unknown relationship between the second plurality of persons, wherein
the communication history information represents a time-series change in at least any of exchange of information between the first plurality of persons or the second plurality of persons via an SNS and transmission of information related to each other by the first plurality of persons or the second plurality of persons via the SNS, and
the attribute information indicates a time-series change in attributes of the first plurality of persons or the second plurality of persons.
The present invention can be used for estimation of any case event that can occur through an SNS, for example, estimation of a special fraud group, estimation of an assailant or a victim of an abduction case, estimation of a person requiring attention such as a terrorist, a crime premonitor, or a suicide wanna-be, and transaction of illegal drugs.
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/JP2020/014061 | 3/27/2020 | WO |
