JPWO2016088230A1 - Causal relationship analysis apparatus and causal relationship analysis method - Google Patents

Abstract

The causal relationship analysis apparatus 10 analyzes the cause of the change in the time series data 205 of the nodes constituting the causal relationship model (causal loop diagram). The causal relationship analysis apparatus 10 is a node that is specified by tracing back the causal relationship between nodes in order starting from a node of interest that is a node having an inflection point in its time-series data 205 among the nodes constituting the causal relationship model For nodes, the node having the inflection point that occurred in the past in the time series data 205 is searched in order to identify the node that has affected the change in the time series data 205 of the node of interest. When a loop is detected by sequentially tracing the causal relationship between nodes, the causal relationship analysis apparatus 10 searches for other nodes that are not included in the loop that are the cause nodes of the nodes that are included in the loop. Include in

Description

  The present invention relates to a causal relationship analysis apparatus and a causal relationship analysis method.

  Patent Document 1 includes “structure data indicating a causal relationship between a plurality of events and information indicating the strength of the causal relationship between any two events for any two events having a causal relationship. A time-series first causal model and second causal model are generated, and for each event that has a causal relationship with the first event, the causal relationship with the first event between the first and second causal models The amount of change in strength and the amount of change in the probability of occurrence of each event are calculated, and among the events that have a causal relationship with the first event, the amount of change in the strength of the causal relationship is greater than or equal to a predetermined threshold A change in the strength of the causal relationship between the first event and the first event, and a change in an event in which the amount of change in the probability is greater than or equal to a predetermined threshold among events related to the first event. Specified as the cause of the change in the first event between the two causal models. " There.

JP 2006-127059 PR

  In recent years, with the advancement of data analysis technology, social trends, industry trends and individual values that have become complicated by utilizing open data such as statistical information published by the government, news articles, SNS (Social Networking Service) statements, etc. Analyzing changes over time, etc., and actively researching and utilizing marketing. However, since social factors, industry trends, and individual values change in a complex manner, various factors are involved, so when you focus on changes in the time-series data of an element of interest, find the element that is the true cause of that change. That is not always easy.

  System dynamics is known as a technique for analyzing factors of time-series changes by focusing on the behavior of a system that fluctuates over time. In system dynamics, elements (nodes) that affect system fluctuations are identified, and causal relationships between the nodes are modeled. Then, by performing a simulation based on the created model, the characteristics of the system that change over time are analyzed.

  In system dynamics, a causal loop diagram is used to model a causal relationship between nodes. The causal loop diagram is composed of arrows indicating the causal relationship with each node. The arrow expresses a "positive correlation" in which if one increases (or decreases) the other increases (or decreases), and one increases (or decreases) the other decreases (or increases) There is something that expresses the “negative correlation”. By using a causal loop diagram, it is possible to express a node that takes time until an effect appears. Therefore, causal loop diagrams are suitable for modeling the characteristics of elements that change over time.

  Here, in Patent Document 1, the strength of the causal relationship between the first event and the event in which the amount of change in the strength of the causal relationship is greater than or equal to a predetermined threshold among the events having a causal relationship with the first event. Of the first event between the first and second causal models, and the change of the event whose probability variation is equal to or greater than a predetermined threshold among the events that are causally related to the first event. Identified as a cause of change.

  However, since Patent Document 1 does not analyze the time difference of the time-series data change for each event, for example, analyze the influence of changes that occurred in the event A on the current event B in the past. I can't. In addition, when the amount of change in the strength of the causal relationship is less than or equal to the threshold, it is not possible to detect the node that causes the change when the sign of the change amount changes from positive (negative) to negative (positive). . In addition, when a group of nodes constituting the causal relationship model forms a loop, that is, a cause of a change in a certain result node is traced back to a node that is in a causal relationship with the result node (tracing the cause node in order). In the case of searching, handling in the case of returning to the result node is not particularly mentioned in Patent Document 1.

  The present invention has been made in view of such a background, applying system dynamics to factor analysis of time series changes such as social changes, industry trends, personal values, etc. to create a causal relationship model, It is an object of the present invention to provide a causal relationship analysis apparatus and a causal relationship analysis method capable of analyzing the influence between nodes on changes in time-series data of nodes constituting a model.

  In order to achieve the above object, a causal relationship analysis apparatus of the present invention is an information processing apparatus that analyzes a cause of a change in time-series data of nodes constituting a causal relationship model, and between nodes constituting a causal relationship model Stores data indicating the causal relationship and time-series data of each node constituting the causal relationship model, and starts from a node of interest that is a node having an inflection point in the time-series data among the nodes constituting the causal relationship model As for the nodes specified by sequentially tracing the causal relationship between the nodes, the time series data of the node of interest is searched by sequentially searching for nodes having inflection points in the time series data that occurred at a past time point. Identify the node that affected the change in.

  In addition, the subject which this application discloses, and its solution method are clarified by the column of the form for inventing, and drawing.

  ADVANTAGE OF THE INVENTION According to this invention, the influence between nodes with respect to the change of the time series data of the node which comprises a causal relationship model can be analyzed.

1 is a diagram illustrating a schematic configuration of a causal relationship analysis system 1. FIG. It is a data flow figure explaining the function with which the causal relationship analysis apparatus 10 is provided, and the data which the causal relationship analysis apparatus 10 manages. It is an example of a causal relationship model (causal loop diagram). It is an example of causal relationship model data 201. It is an example of the social media data 203. FIG. 2 is an example of open data 204; It is an example of the time series data 205. It is a flowchart explaining time series data generation processing S800. It is a flowchart explaining causal relationship analysis processing S900. It is an example of the analysis result display screen 1000.

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

  FIG. 1 shows a schematic configuration of a causal relationship analysis system 1 described as an embodiment. As shown in the figure, the causal relationship analysis system 1 includes a causal relationship analysis device 10 and a server device 20. The causal relationship analysis device 10 and the server device 20 are communicably connected via the communication network 5. The communication network 5 is, for example, the Internet or a dedicated line.

  The server device 20 is, for example, an SNS server (SNS: Social Network Service), an open data server, a Web server, and the like, and provides information (data) to other devices accessed via the communication network 5. .

  The causal relationship analysis device 10 accesses the server device 20 via the communication network 5 and acquires information from the server device 20. The causal relationship analysis device 10 uses the acquired information and applies the system dynamics (System Dynamics) to the investigation of changes in social trends, industry trends, personal values, etc. (causal loop diagram (causal loop diagram ( Regarding CLD: Causal Loop Diagram)), the cause of the change in the time series data of the nodes constituting the above causal relationship model is analyzed.

  The causal relationship analysis apparatus 10 stores data indicating the causal relationship between nodes constituting the causal relationship model, and time-series data of each node constituting the causal relationship model. The causal relationship analysis apparatus 10 generates the time series data based on information acquired via the communication network 5, for example.

  The causal relationship analysis apparatus 10 uses the generated time series data, and among the nodes constituting the causal relationship model, the causal relationship between the nodes is sequentially determined starting from the node of interest that is a node having an inflection point in the time series data. For a node identified by going back, the node that has affected the change in the time-series data of the node of interest is identified by sequentially searching for nodes that have inflection points that occurred in the past in the time-series data. . More specifically, for example, the causal relationship analyzer 10 traces the causal relationship between the nodes in order, so that the node identified last as the node having the inflection point generated in the past time point in the time-series data. Are identified as nodes that have affected the change in the time-series data of the node of interest.

  In addition, the causal relationship analysis apparatus 10 appropriately uses a degree of association that is an index indicating the strength of connection between nodes in the above specification. For example, the causal relationship analysis apparatus 10 determines that the node that has been confirmed last is affected by the node of interest when the degree of association between the node identified last and the node of interest is greater than or equal to a preset strength. As specified.

  The inflection point is, for example, (1) the point where the sign of the slope of the rate of change of the time series data is inverted, and (2) the point where the absolute value of the slope of the rate of change of the time series data changes more than a preset threshold value. (3) The point where the sign of the slope of the rate of change of the time series data is inverted and the absolute value of the slope of the rate of change has changed more than a preset threshold value. By using such a point as an inflection point, it is possible to accurately identify the node that has affected the change in the time-series data of the node of interest.

  When the causal relationship analysis apparatus 10 detects a loop by sequentially tracing the causal relationships between the nodes, the causal relationship analyzing apparatus 10 also causes other nodes that do not constitute the loop that are the causal nodes of the nodes that constitute the loop. Include in search target. In this way, the causal relationship analysis apparatus 10 detects the presence of a loop and expands the search range outside the loop, so that it is possible to reliably identify a node that has affected the change in time-series data of the node of interest without omission. it can.

  As illustrated in FIG. 1, the causal relationship analysis device 10 is realized by an information processing device (computer) including a processor 101, a storage device 102, an input device 103, an output device 104, and a communication device 105. The processor 101 is configured using, for example, a CPU (Central Processing Unit) and an MPU (Micro Processing Unit). The storage device 102 is a device that stores programs and data. For example, ROM (Read Only Memory), RAM (Random Access Memory), NVRAM (Non Volatile RAM), hard disk drive, SSD (Solid State Drive), optical type A storage device or the like. The input device 103 is a user interface that receives input of information and instructions from the user, and is, for example, a keyboard, a mouse, a touch panel, or the like. The output device 104 is a user interface that provides information to the user, such as a graphic card or a liquid crystal monitor. The communication device 105 is a communication interface that communicates with other devices via the communication network 5, and is, for example, a NIC (Network Interface Card) or a wireless LAN interface. The processor 101, the storage device 102, the input device 103, the output device 104, and the communication device 105 are communicably connected via communication means such as a bus.

  FIG. 2 is a data flow diagram for explaining the functions of the causal relationship analysis device 10 and the data managed by the causal relationship analysis device 10. The causal relationship analysis apparatus 10 includes functions of a time-series data generation unit 206, a causal relationship analysis unit 207, and an analysis result output unit 208. These functions are realized by the processor 101 reading and executing a program stored in the storage device 102. Each function shown in the figure is merely set for convenience in order to facilitate understanding of the function of the causal relationship analyzer 10, and the classification method and name of each function are not limited to the modes shown here. . These functions may be realized by hardware (ASIC (Application Specific Integrated Circuit) or the like), for example. Further, for example, a configuration in which a plurality of these functions are realized by a single piece of hardware may be adopted. Further, for example, a configuration in which these functions are realized in a distributed or cooperative manner by a plurality of hardware may be employed.

  As shown in the figure, the causal relationship analysis apparatus 10 includes causal relationship model data 201, a natural language dictionary 202 (such as a synonym dictionary and a co-occurrence word dictionary), social media data 203, open data 204, and time-series data 205. to manage. The causal relationship analysis apparatus 10 manages these data as, for example, a database table provided by a DBMS (DataBase Management System).

  The causal relationship model data 201 is data representing relationships between nodes described in the causal relationship model with respect to changes in social trends, industry trends, individual values, and the like.

  FIG. 3 shows an example of a causal relationship model (causal loop diagram) expressed by the causal relationship model data 201. In the figure, a node 301 indicates a node constituting the social trend A, for example. The link 302 indicates that there is a causal relationship between the nodes. The start point of the arrow is connected to the “cause node”, and the end point of the arrow is connected to the “result node”. A link 302 indicated by a solid line indicates that there is a “positive correlation” between the node at the start point and the node at the end point to which the link 302 is connected, and the link 302 indicated by the dotted line is connected to the link 302. It shows that there is a “negative correlation” between the start node and the end node.

  The loop 303 has a structure of returning to the starting node as a result of tracing back the causal relationship between the nodes in order from a certain node (tracing the cause node in order). In this example, the cause node of node A is node D, the cause node of node D is node C, the cause node of node C is node B, and the cause node of node B is node A. A loop is formed from node A back to node A.

  FIG. 4 shows causal relationship model data 201 corresponding to the causal relationship model (causal loop diagram) shown in FIG. As shown in the figure, the causal relation model data 201 includes a node ID 401, a node name 402, a cause node ID 403, a cause node relation 404, a cause node relation 405, a result node ID 406, a result node relation 407, and a result. It is composed of a plurality of records having each item of the degree of relevance 408 between nodes.

  Among these, the node ID 401 is set with an identifier (a character string, a numerical value, etc.) of a node constituting the causality model. The node name 402 is set with information (such as a character string) indicating the name of the node, for example, information indicating the contents of the node.

  The cause node ID 403 is set with the node ID of the node corresponding to the cause node when the node identified by the node ID 401 is the result node. In the inter-cause node relationship 404, information indicating the relationship (“positive correlation” or “negative correlation”) between the node identified by the node ID 401 and the node identified by the cause node ID 403 is set. In the present embodiment, when there is a “positive correlation” between the node specified by the node ID 401 and the node specified by the cause node ID 403, “1” indicates that the node specified by the node ID 401 and the cause node ID 403. If there is a “negative correlation” with the node specified by “0”, “0” is set.

  The cause node relevance degree 405 is a relevance index (hereinafter also referred to as a cause node relevance degree) that indicates the strength of association between the node identified by the node ID 401 and the node identified by the cause node ID 403. ) Is set. The causal relationship analysis unit 207 includes, for example, terms (keywords) included in the social media data 203 (or open data 204) related to the node specified by the node ID 401 and social related to the node specified by the cause node ID 403. The degree of association is obtained based on the co-occurrence frequency between terms (keywords) included in the media data 203 (or open data 204). For example, when the node ID 401 is “node A” and the cause node ID 403 is “node D”, the causal relationship analysis unit 207 stores the social media data 203 that matches “keyword a” and “keyword b” of the cause node ID 403. The number of cases is used as a denominator, and in addition to “keyword a” and “keyword b” of cause node ID 403, the degree of relevance is obtained based on the ratio obtained by using the number of cases of social media data 203 including “keyword d” of node ID 401 as a numerator. . The degree of association between cause nodes may be obtained from the correlation between the time series data 205 of the node specified by the node ID 401 and the time series data 205 of the node specified by the cause node ID 403 (in this case, the time series data 205, the evaluation value may be normalized or the time lag on the horizontal axis may be corrected as appropriate.

  In the result node ID 406, the node ID of the node corresponding to the result node when the node specified by the node ID 401 is set as the cause node is set. In the inter-result node relationship 407, information indicating the relationship between the node identified by the node ID 401 and the node identified by the result node ID 406 (“positive correlation” or “negative correlation”) is set. In the present embodiment, when there is a “positive correlation” between the node identified by the node ID 401 and the node identified by the result node ID 406, “1” indicates that the node identified by the node ID 401 and the result node ID 406 If there is a “negative correlation” with the node specified by “0”, “0” is set.

  The inter-result node association degree 408 is an association degree that is an index indicating the strength of connection between the node identified by the node ID 401 and the node identified by the result node ID 406 (hereinafter also referred to as inter-result node association degree). ) Is set. The causal relationship analysis unit 207 obtains the degree of association between result nodes by the same method as the degree of association between cause nodes described above, for example. The degree of association between result nodes may be obtained from the correlation between the time series data 205 of the node specified by the node ID 401 and the time series data 205 of the node specified by the result node ID 406 (in this case, the time series data 205, the evaluation value may be normalized or the time lag on the horizontal axis may be corrected as appropriate.

  FIG. 5 shows an example of the social media data 203 acquired from the server device 20. As shown in the figure, the social media data 203 is composed of a plurality of records having items of a speaker ID 501, a speech ID 502, a speech date 503, and a speech content 504. The social media data 203 is data provided by, for example, a social networking service, a blog, a bulletin board, or the like.

  In the speaker ID 501, information (character string, numerical value, etc.) for specifying the speaker of the social media data 203 is set, for example, a user name is set. Information (character string, numerical value, etc.) for identifying each utterance (record) is set in the utterance ID 502. In the utterance date and time 503, information (character string, numerical value, etc.) indicating the date and time when the utterance of the record is posted (posted) on social media is set. In the comment content 504, information (character string, numerical value, etc.) indicating the content of the comment posted (posted) on social media is set.

  FIG. 6 shows an example of the open data 204 acquired from the server device 20. As shown in the figure, the open data 204 is composed of a plurality of records each having items of a list ID 601, a title 602, and link information 603.

  The open data 204 is, for example, public information released by public offices, public institutions, companies, and research institutions. The open data 204 is publicly available and highly reliable information such as news and weather information, which is disclosed on a medium such as TV. Further, the open data 204 includes, for example, attribute information of a publisher such as a publication date and a publication source. The open data 204 includes information such as “who”, “what”, “when”, “where”, “why”, “how”, and the like. The open data 204 includes, for example, information that does not use slang or intentional erroneous input / error conversion.

  In the list ID 601, an identifier (character string, numerical value, etc.) of each record of the open data 204 is set. In the title 602, information (character string, numerical value, etc.) corresponding to the title of the open data 204 of the record is set. In the link information 603, information indicating the location of the open data 204 of the record (for example, URL (Uniform Resource Locator)) is set.

  FIG. 7 is an example of time-series data 205 generated using social media data 203 or open data 204. As shown in the figure, the time-series data 205 is composed of a plurality of records having items of node ID 701, node name 702, time 703, actual data 704, and negative / positive ratio 705.

  In the node ID 701, a node identifier (character string, numerical value, etc.) is set. The identifier set for the node ID 701 is common to the node ID 401 of the causal relationship model data 201. In the node name 702, information (such as a character string) indicating the name of the node is set. For example, information indicating the specific contents of the node is set. Information set in the node name 702 is the same as the node name 402 of the causal relationship model data 201.

  In the time 703, information indicating a time corresponding to the actual data 704 (for example, a time when the actual data 704 is generated) is set. The time 703 is used as information on the horizontal axis when the time series data 205 is expressed in a graph. For example, the time when the actual data 704 is acquired is set as the time 703. Although the figure shows a case where the year is set for the time 703, for example, a month, a date, or the like can be set for the time 703. Data included in the social media data 203 and the open data 204 may be used as they are for the period and step size of the time 703, or the data may be appropriately processed by the user or the like.

  In the actual data 704, the contents of the time-series data 205 (numerical values indicating sizes, etc.) are set. For example, in the actual data 704, an evaluation value included in the social media data 203 or the open data 204 and a value obtained by processing the evaluation value (ratio, etc.) are set. The actual data 704 is used as information on the vertical axis when the time series data 205 is expressed in a graph.

  In the negative / positive ratio 705, the number of social media data 203 having negative (negative) content about the content of the node corresponding to the record (for example, the content of the node name 702) in the entire social media data 203 corresponding to the record. And the number of social media data 203 having a positive content (hereinafter also referred to as a negative / positive ratio). The causal relationship analysis apparatus 10 determines whether the social media data 203 is negative content or positive content with respect to the content of the node, for example, the content of the social media data 203, the content of the node of the record, And by comparing the natural language dictionary 202. The negative / positive ratio 705 is used as information on the vertical axis when the time-series data 205 is expressed in a graph.

  Then, the process which the causal relationship analysis apparatus 10 performs is demonstrated. In the following description, the letter “S” attached to the head of the code is an abbreviation for the processing step.

<Time-series data generation processing>
FIG. 8 is a flowchart for explaining processing performed by the time series data generation unit 206 (hereinafter referred to as time series data generation processing S800). Hereinafter, the time series data generation process S800 will be described with reference to FIG. The time series data generation process S800 is executed as needed.

  First, the time series data generation unit 206 selects one node (record, node ID 401) of the causal relationship model data 201 (S801).

  Subsequently, the time series data generation unit 206 generates a keyword used for generating the time series data 205 based on the node name 402 of the selected node. For example, when the character string “node A” is set in the node name 402, the time-series data generation unit 206 generates “keyword a” and “keyword b” based on the character string. The keyword may be generated using a word extracted from the node name 402 as it is, or a synonym or co-occurrence word acquired by comparing the word extracted from the node name 402 with the natural language dictionary 202, etc. (S802).

  Subsequently, the time-series data generation unit 206 determines whether there is open data 204 that matches the keyword generated in S802 (S803). If there is open data 204 that matches the generated keyword (S803: YES), the process proceeds to S804. If there is no open data 204 that matches the generated keyword (S803: NO), the process proceeds to S805.

  In S804, the time-series data generation unit 206 acquires the open data 204 using the keyword generated in S802, and generates the time-series data 205 using the acquired open data 204. Thereafter, the process proceeds to S806.

  On the other hand, in S805, the time-series data generation unit 206 acquires the social media data 203 using the keyword generated in S802, and generates the time-series data 205 using the acquired social media data 203. In this case, the data on the vertical axis of the time-series data 205 may be actual data 704 or a negative / positive ratio 705. Thereafter, the process proceeds to S806.

  In step S806, the time series data generation unit 206 determines whether all nodes have been selected in step S801. When all the nodes have been selected (S806: YES), the time series data generation unit 206 ends the time series data generation process S800. If an unselected node remains (S806: NO), the time-series data generation unit 206 selects one unselected node in S801 and executes the processing from S802.

  By executing the time series data generation process S800 described above, time series data 205 based on the social media data 203 or the open data 204 is generated.

<Causality analysis process>
FIG. 9 is a flowchart for explaining a process (hereinafter referred to as a causal relation analysis process S900) performed by the causal relation analysis unit 207. By executing the causal relationship analysis process S900, it is possible to identify a node (hereinafter, also referred to as a factor node) that has affected the change in the time-series data 205 of the node of interest in the causal relationship model. Hereinafter, the causal relationship analysis process S900 will be described with reference to FIG. The causal relationship analysis process S900 is executed after the time-series data generation process is executed.

  First, the causal relationship analysis unit 207 determines the relevance level (cause node relevance 405, result node relevance) of each node (each node (record, node ID 401) of the causal relationship model data 201) of the causal relationship model (causal loop diagram). Degree 408) is calculated (S901).

  Subsequently, the causal relationship analysis unit 207 calculates the rate of change of the time series data 205 for each node (each node (record, node ID 401) of the causal relationship model data 201) of the causal relationship model (causal loop diagram). For example, the causal relationship analysis unit 207 calculates the rate of change of the time series data 205 for each year based on the difference between the previous year's evaluation value and the current year's evaluation value (S902).

  Subsequently, the causal relationship analysis unit 207 determines whether or not there is a node having an inflection point in the time series data 205 for each node constituting the causal relationship model (causal loop diagram) (S903). . When such a node exists (S903: YES), the causal relationship analysis unit 207 performs the processing from S904. On the other hand, when such a node does not exist (S903: NO), the causal relationship analysis unit 207 ends the causal relationship analysis process S900.

  In S904, the causal relationship analysis unit 207 selects one node as a target node from nodes having inflection points in the time-series data 205 (S904). For example, the selection may be made from among the nodes having inflection points in the time-series data 205, for example, the nodes where the cause node relation 404 or the result node relation 407 is broken before and after the occurrence of the inflection point (for example, the correct node This is performed by selecting a node that has changed from a negative correlation to a negative correlation or vice versa. Further, for example, the user may make the selection.

  Subsequently, the causal relationship analysis unit 207 determines whether or not the time-series data 205 of the cause node of the node of interest has an inflection point that occurred in the past from the time t1 when the inflection point of the node of interest occurred. Determination is made (S905). If the time-series data 205 of the cause node of the node of interest has an inflection point that occurred in the past from the time t1 at which the inflection point of the node of interest occurred (S905: YES), the process proceeds to S906. On the other hand, if the time-series data 205 of the cause node of the node of interest does not have an inflection point that occurred in the past from the time t1 at which the inflection point of the node of interest occurred (S905: NO), the process proceeds to S907. move on.

  In S906, the causal relationship analysis unit 207 determines that the node that matches the condition in S905 (the time series data 205 has an inflection point that occurred in the past from the time t1 when the inflection point of the node of interest occurred). Node) is set as a terminal node. Note that when there are a plurality of nodes that match the condition of S905, the causal relationship analysis unit 207 selects, for example, the cause node having the maximum value of the relationship degree 405 between cause nodes. Thereafter, the process proceeds to S908.

  In S907, the causal relationship analysis unit 207 selects the end node (the node of interest when there is no node that matches the condition in S905; otherwise, the node set in S906) in the time-series data 205 of the node of interest. It is determined that the node has influenced the change, that is, the factor node, and the causal relationship analysis process S900 is ended. In S907, when the causal relationship analysis unit 207 determines that the node that has detected the terminal node is the factor node, the terminal node is also conditional on the degree of association between the terminal node and the node of interest being greater than or equal to a preset strength. May be determined as a factor node.

  In S908, the causal relationship analysis unit 207 determines the presence or absence of a loop including the node of interest by tracing back the nodes having the causal relationship starting from the node of interest based on the causal relationship model data 201 (tracing the cause node in order). When the causal relationship analysis unit 207 detects a loop, the causal relationship analysis unit 207 includes a change that occurred in the past from the time t2 when the inflection point occurred in the time series data 205 of the terminal node in the time series data 205 during the loop It is determined whether there is a node having a point. When a loop exists and a node having an inflection point that occurred in the past from time t2 at which an inflection point occurred in the time series data 205 of the end node in the time series data 205 in the loop (S908) : YES), the process proceeds to S909.

  On the other hand, when there is no loop, or there is no node having an inflection point that occurred in the past from the time t2 when the inflection point occurred in the time series data 205 of the terminal node in the time series data 205 (S908: NO), the process proceeds to S907. In S907, the causal relationship analysis unit 207 determines that the terminal node is a factor node, and ends the process.

  In S909, the causal relationship analysis unit 207 sets the node detected in S908 as a terminal node.

  In S910, the causal relationship analysis unit 207 traces the causal relationship of the nodes constituting the loop detected in S908 (following the cause nodes in order), thereby causing the cause nodes of the nodes included in the loop detected in S908. It is determined whether or not there is a node outside the loop that has an inflection point that occurred in the past from time t3 when the time series data 205 of the end node occurred. If such a node exists (S910: YES), the process proceeds to S911. On the other hand, when such a node does not exist (S910: NO), the process proceeds to S912.

  In S911, the causal relationship analysis unit 207 determines that the node detected in S910 is a factor node, and ends the causal relationship analysis process S900. In S911, when the causal relationship analysis unit 207 determines that the node detected in S910 is a factor node, it is also a condition that the degree of association between the node detected in S910 and the node of interest is greater than a preset strength. The node detected in S910 may be determined as the factor node.

  On the other hand, in S912, the causal relationship analysis unit 207 determines that the node set as the terminal node in S909 is a factor node, and ends the causal relationship analysis process S900. In S912, when the causal relationship analysis unit 207 determines that the terminal node is the factor node, the terminal node is determined to be the factor node on the condition that the degree of association between the terminal node and the node of interest is equal to or higher than a preset strength. May be determined.

  According to the causal relationship analysis process S900 described above, it is possible to identify a factor node that has affected the time-series data 205 of the node of interest.

  In addition, when a causal relationship analysis process S900 cannot identify a factor node, for example, the causal relationship analysis unit 207 configures character information (for example, social media data 203 and open data 204) related to the node of interest. From the term that makes up the character information related to the node that has a causal relationship with the node of interest (for example, social media data 203 or open data 204), out of the character information related to the node of interest after the removal, You may make it output the term which cooccurs with the keyword (for example, the keyword produced | generated from the node name 702 of an attention node) about an attention node. As a result, information serving as a clue for specifying the factor node can be provided to the user.

<Processing example>
Next, a specific processing example of the causal relationship analysis processing S900 will be described in the case where the causal relationship model has the contents of FIG. 3 and the node whose node name 402 is “node A” is selected as the node of interest in S904.

  First, in step S905, the causal relationship analysis unit 207 identifies “node D” or “node G” that is the cause node of the node of interest “node A”, and the time series data 205 indicates the inflection point of the node of interest. It is determined whether or not it has an inflection point that occurred in the past from the time t1 that occurred. Here, the causal relationship analysis unit 207 determines that the time-series data 205 of “node D” has an inflection point that occurred in the past from time t1 (S905: YES), and in S906, the causality The relationship analysis unit 207 sets “node D” as a terminal node.

  Subsequently, in S908, the causal relationship analysis unit 207 detects the loop 303 including the “node A” by sequentially tracing back the nodes having the causal relationship from the “node A” that is the node of interest. Then, the causal relationship analysis unit 207 determines whether there is a node in the loop 303 that has an inflection point that occurred in the past from the time t2 when the inflection point occurred in the time series data 205 of the terminal node in the time series data 205. Determine whether or not. Here, it is assumed that the causal relationship analysis unit 207 detects “node C”, which is a node having an inflection point that occurred at time t3 past time t2 in the time series data 205.

  In step S909, the causal relationship analysis unit 207 sets “node C” detected in step S908 as a terminal node.

  In S910, the causal relationship analysis unit 207 traces back the causal relationship of the nodes constituting the loop detected in S908, so that the cause node of the node included in the loop detected in S908 (the node outside the loop) It is determined whether or not there is data having an inflection point that occurred in the past from the time t3 when the time series data 205 of the end node is generated. Here, it is assumed that such a node is not detected, and the process proceeds to S912.

  In step S912, the causal relationship analysis unit 207 determines that “node C” set as the terminal node in step S909 is a factor node.

  As described above, according to the causal relationship analysis system 1 of the present embodiment, it is possible to analyze the influence between nodes on the change of the time series data 205 of the nodes constituting the causal relationship model. Therefore, for example, it is possible to analyze the influence of changes that have occurred in the event A in the past on the current event B. In addition, since the causal relationship analysis device 10 searches for a node using the time when the inflection point occurs as a criterion, it can accurately identify the factor node. Moreover, since the causal relationship analysis apparatus 10 searches for a factor node while considering a loop, the factor node can be reliably searched without omission.

<Screen example>
FIG. 10 is an example of a screen (hereinafter referred to as an analysis result display screen 1000) that the analysis result output unit 208 outputs (displays) to the output device 104. As shown in the figure, the analysis result display screen 1000 has a causal relationship model display column 1010, a time series data (attention node) display column 1020, and a time series data (detection node) display column 1030. The causal relationship model display field 1010 is provided with an inflection point detection time display field 1011.

  In the causal relationship model display column 1010, a causal loop diagram of the causal relationship model as an analysis target is displayed. In the displayed cause and effect loop diagram, a design that highlights the node of interest (in the same diagram, a design in which the symbol “!” Is drawn in a solid circle), a design that highlights the cause node (in the diagram, A design in which a symbol “*” is drawn in a solid circle), and a design in which a node having an inflection point is highlighted in the time series data 205 (in the same drawing, a symbol “!” Is drawn in a broken circle). Displayed). The inflection point detection time display field 1011 displays the time at which the inflection point of the time series data 205 of the node of interest has occurred.

  The user can easily grasp the attention node and the factor node by referring to the causal relationship model display field 1010. Further, by referring to the causal relationship model display field 1010, it is possible to easily grasp the node having the inflection point in the time series data 205, and the user can easily grasp the node that is a candidate of the node of interest. it can. Also, the user refers to the inflection point detection time display column 1011, and the time when the inflection point of the time-series data 205 of the node of interest occurs (which inflection point is used as the starting point to search for the factor node) Can be easily grasped. Note that the node of interest to be analyzed may be specified by the user operating the user interface to specify the node displayed in the causal relationship model display field 1010.

  In the time series data (attention node) display field 1020, the inflection point of the time series data 205 corresponding to the attention node is displayed. The user can easily confirm the time series data 205 of the node of interest by referring to the time series data (node of interest) display column 1020. In order to facilitate the correspondence with the display contents of the causal relationship model display field 1010, the time series data (attention node) display field 1020 has a design for emphasizing the attention node (in FIG. Is displayed on the screen).

  In the time-series data (detection node) display column 1030, the time-series data 205 of the node detected by tracing the cause node starting from the node of interest and the time when the inflection point occurred in each time-series data 205 are displayed. The In the time-series data (detection node) display field 1030, the time-series data 205 of the node specified as the factor node is highlighted (in the figure, a thick frame line and an emphasis design (in the solid line circle in the figure)). The symbol “*” is highlighted by the design drawn.) The user refers to the time series data (attention node) display field 1020 and the time series data (detection node) display field 1030 to indicate It is possible to visually grasp the relationship between the time series data 205 and the time series data 205 of the factor node.

  By the way, this invention is not limited to above-described embodiment, Various modifications are included. For example, the above-described embodiment has been described in detail for easy understanding of the present invention, and is not necessarily limited to one having all the configurations described. Further, a part of the configuration of an embodiment can be replaced with the configuration of another embodiment, and the configuration of another embodiment can be added to the configuration of an embodiment. In addition, it is possible to add, delete, and replace other configurations for a part of the configuration of each embodiment.

  Each of the above-described configurations, functions, processing units, processing means, and the like may be realized by hardware by designing a part or all of them with, for example, an integrated circuit. In addition, each of the above-described configurations, functions, and the like may be realized by software by interpreting and executing a program that realizes each function by the processor. Information such as programs, tables, and files for realizing each function can be stored in a recording device such as a memory, a hard disk, or an SSD, or a recording medium such as an IC card, an SD card, or a DVD.

  The control lines and information lines indicate what is considered necessary for the explanation, and not all the control lines and information lines on the product are necessarily shown. Actually, it may be considered that almost all the components are connected to each other.

1 causal relationship analysis system, 5 communication network, 10 causal relationship analysis device, 20 server device, 101 processor, 102 storage device, 103 input device, 104 output device, 105 communication device, 201 causal relationship model data, 202 natural language dictionary, 203 social media data, 204 open data, 205 time series data, 206 time series data generation unit, 207 causal relationship analysis unit, 208 analysis result output unit, S800 time series data generation processing, S900 causal relationship analysis processing, 1000 analysis result display Screen, 1010 Causal relation model display field, 1020 Time series data (attention node) display field, 1030 Time series data (detection node) display field

Claims (15)

An information processing apparatus for analyzing a factor of change in time series data of nodes constituting a causal relationship model,
Storing data indicating the causal relationship between nodes constituting the causal relationship model, and time-series data of each node constituting the causal relationship model;
Among the nodes that make up the causal relationship model, nodes that are identified by tracing back the causal relationship between the nodes starting from the node of interest that is the node having the inflection point in the time-series data. A causal relationship analysis apparatus that identifies a node that has affected a change in time-series data of the node of interest by sequentially searching for nodes having the generated inflection points in the time-series data. The causal relationship analyzer according to claim 1,
The inflection point is the point where the sign of the slope of the rate of change of the time series data is inverted, the point where the absolute value of the slope of the rate of change of the time series data has changed by more than a threshold, and the slope of the rate of change of the time series data. A causal relationship analysis apparatus, wherein the sign is inverted and the absolute value of the gradient of the change rate changes at least a threshold value. The causal relationship analyzer according to claim 1,
When a loop is detected by sequentially tracing the causal relationship between the nodes, the other nodes that do not constitute the loop and that are the cause nodes of the nodes that constitute the loop are included in the search target. Relationship analysis device. The causal relationship analyzer according to claim 1,
By tracing back the causal relationship between the nodes in order, the node identified last as the node having the inflection point in the time point in the past is affected to the change in the time series data of the node of interest. A causal relationship analysis device that identifies a given node. The causal relationship analyzer according to claim 4,
Storing a degree of association that is an index indicating the strength of connection between the nodes;
When the degree of association between the last identified node and the node of interest is greater than or equal to a preset strength, the last identified node is identified as a node that has affected the node of interest. Relationship analysis device. The causal relationship analyzer according to claim 5,
Storing character information related to the nodes constituting the causal relationship model;
The causality analysis apparatus which calculates | requires the said association degree based on the co-occurrence frequency of the term contained in the said character information of the said node specified last, and the term contained in the said character information of the said attention node. The causal relationship analyzer according to claim 1,
Storing character information related to the nodes constituting the causal relationship model;
If it is not possible to identify the node that has affected the time-series data change of the node of interest by the search, from the terms constituting the character information related to the node of interest, to the node that has a causal relationship with the node of interest A causal relationship analysis apparatus that removes terms constituting related character information and outputs a term that co-occurs with a keyword for the node of interest among character information associated with the node of interest after the removal. The causal relationship analyzer according to claim 1,
A causal relationship analysis device that generates the time-series data based on at least one of open data and social media data. The causal relationship analyzer according to claim 1,
The information indicating the node identified as the node that has affected the change in the time-series data of the node of interest, the information indicating the change of the time-series data of the node of interest, and the change in the time-series data of the node of interest A causal relationship analysis apparatus that generates a screen including information indicating a change in the time-series data of a given node. A method for analyzing a factor of change in time series data of nodes constituting a causality model,
Information processing device
Storing data indicating the causal relationship between nodes constituting the causal relationship model, and time-series data of each node constituting the causal relationship model;
Among the nodes that make up the causal relationship model, nodes that are identified by tracing back the causal relationship between the nodes starting from the node of interest that is the node having the inflection point in the time-series data. A causal relationship analysis method comprising: a step of sequentially searching for nodes having the generated inflection points in the time-series data to identify a node that has affected the change in the time-series data of the node of interest. The causal relationship analysis method according to claim 10,
The inflection point is the point where the sign of the slope of the rate of change of the time series data is inverted, the point where the absolute value of the slope of the rate of change of the time series data has changed by more than a threshold, and the slope of the rate of change of the time series data. A causal relationship analysis method, wherein the sign is inverted and the absolute value of the slope of the change rate changes at least a threshold value. The causal relationship analysis method according to claim 10,
When a loop is detected by sequentially tracing the causal relationship between the nodes, the information processing apparatus causes other nodes that do not constitute the loop to be the cause nodes of the nodes that constitute the loop. Causal analysis method to be included in the search target. The causal relationship analysis method according to claim 10,
The information processing apparatus traces back the causal relationship between the nodes in order, so that the node identified last as a node having inflection points in the time series in the time series data is the time series of the node of interest. A causal analysis method that identifies nodes that have affected data changes. The causal relationship analysis method according to claim 13,
The information processing apparatus is
Storing a degree of association that is an index indicating the strength of connection between the nodes;
Identifying the last identified node as the node that has affected the target node when the degree of association between the last identified node and the target node is greater than or equal to a preset strength A causal relationship analysis method further including and. The causal relationship analysis method according to claim 14,
The information processing apparatus is
Storing character information related to nodes constituting the causal relationship model;
A causal relationship analysis further comprising: determining the relevance based on a co-occurrence frequency of a term included in the character information of the node identified last and a term included in the character information of the node of interest. Method.

Images