JP6748474B2 - Decision support system and decision support method - Google Patents

Description

The present invention relates to a decision support system and a decision support method.

As a background art in this technical field, there is Patent Document 1. Patent Document 1 has an object of “supporting decision making of what kind of measures should be taken when an event of an organization or an individual's involvement target occurs”, and “a plurality of events of the engagement target are predetermined. Means for evaluating what evaluation value corresponds to on the evaluation axis of, and means for determining whether any of the evaluation values of the involved event evaluated on a plurality of evaluation axes is outside the allowable range , If any of the evaluation values on multiple evaluation axes is out of the allowable range, select the important evaluation axis and extract the measures defined corresponding to the important evaluation axes. Select an arbitrary number of policies, and if the selected policy is implemented, a means for inferring the change in the evaluation value on the important evaluation axis of the event by an inference model for each policy, and a target for participation based on the inference result And means for selecting the optimal strategy to correct the event to an acceptable state."

Further, in Patent Document 1, "means for evaluating what evaluation value an event of interest corresponds to on a plurality of predetermined evaluation axes, and evaluation values of each event of interest for evaluation evaluated on a plurality of evaluation axes" If any of the evaluation values on multiple evaluation axes are out of the acceptable range, the important evaluation axis is selected and the important evaluation axis is selected. The method of extracting the policy defined in accordance with the above, and the change in the evaluation value on the important evaluation axis of the involved event when selecting an arbitrary number of the extracted policies and implementing the selected policy "Decision support simulation system" is provided with means for inferring each of the measures by an inference model for each policy, and means for selecting an optimum policy for correcting the involved event to a state of an allowable range based on the inference result. This decision support simulation system supports inference and evaluation of a policy by an inference model in the case of occurrence of an event in which an organization or an individual is involved, and supports decision making on what policy to take.

JP, 2004-110469, A

However, in Patent Document 1, as a model for estimating an event that occurs due to decision making, a model first assembled by a user is used, and there is no method for updating the model in response to a continuously changing actual situation, There is a problem that it becomes a model that deviates from the reality and it becomes difficult to support decision making. Especially when the situation, information, knowledge and know-how change from moment to moment, the model tends to deviate from reality.

In addition, even when updating the model based on the situation and information that changes sequentially, if many additions and updates that have little impact on decision support (small meaning and effect) are included, the model will become huge and the quality will deteriorate. There is a problem that the performance (speed, accuracy) of the prediction simulation is reduced. In addition, the wider the range of events to be predicted, the larger the number of information sources for updating, and the lower the updating speed.

Therefore, in order to solve the above problems, it is an object of the present invention to provide a technique for effectively updating a model for estimating a phenomenon that occurs for decision support by simulation.

Decision support system of the present invention for achieving the above object, a causal relationship model management unit, and data collection and analysis unit, a decision support system comprising a causal relationship model updating unit, wherein the causal relationship model The management unit stores a plurality of events that occur in society or nature and the relationship of occurrence between the events as a causal relationship model, and the data collection/analysis unit acquires data related to the events. A causal relationship model update unit that has a function and a data analysis function that analyzes acquired information about the event, and the causal relationship model updating unit is configured to perform the causal relationship based on the causal relationship model and an analysis result by the data analysis function. An update information generation function for generating update data for updating the model, a model update evaluation function for evaluating the effectiveness of the update by the update data, and an interest event storage unit for storing an event of interest to the user, And the model update evaluation function evaluates the effectiveness of the update based on the update data and the event stored in the event-of-interest storage unit, and the causal relationship model management unit includes the model update evaluation function. The model registration/update function for updating the causal relationship model with the update data when the evaluation result by the above is above a certain level.

The decision support method of the present invention is a decision support method executed by a decision support system including a causal relationship model management unit, a data collection/analysis unit, and a causal relationship model updating unit , Acquired information on the events that occur in society or nature, analyzed the acquired information about the events, and composed of the plurality of events that occur in society or nature and the relationship of occurrence between the events. Based on the causal relationship model and the analysis result of the information, update data for updating the causal relationship model is generated, the effectiveness of the update by the update data is evaluated, and the evaluation result of the effectiveness of the update is generated. Is greater than or equal to a certain value, the causal relationship model is updated with the update data.

According to the present invention, it is possible to provide a technique for effectively updating a model for estimating an event that occurs for decision support by simulation.

It is a figure showing an example of the whole block diagram of a decision support system. It is a figure which shows an example of the functional block diagram of a decision support system. It is a figure which shows an example of a decision support sequence. It is a figure which shows an example of a causal relationship model DB. It is a figure which shows an example of reaction prediction result DB. It is a figure which shows an example of a decision support screen (initial event setting, prediction result display). It is a figure showing an example of a decision support screen (occurrence event change, prediction result display). It is a figure which shows an example of an interest event registration sequence. It is a figure which shows an example of interest event management DB. It is a flow chart which shows an example of a processing procedure (interesting event setting to a model) of an interested event registration function. It is a flow chart which shows an example of a processing procedure (interesting event setting to a prediction result) of an interested event registration function. It is a figure which shows an example of an interest event registration screen (interest event setting with respect to a model). It is a figure which shows an example of an interest event registration screen (interest event setting with respect to a prediction result). It is a figure which shows an example of a causal relationship model update sequence. It is a figure which shows an example of acquisition/analysis management DB. It is a figure which shows an example of acquisition information DB. It is a flow chart which shows an example of the processing procedure of a data acquisition function. It is a figure which shows an example of (a) analysis result management DB and (b) occurrence event management DB. It is a flow chart which shows an example of a processing procedure of a data analysis function. It is a figure which shows an example of temporary update model DB. 9 is a flowchart illustrating an example of a processing procedure of an update information generation function. It is a figure which shows an example of model update management DB. It is a flow chart which shows an example of a processing procedure of a model update evaluation function and a model registration and update function. It is a flow chart which shows an example of a processing procedure of an update log registration function and an update log analysis function. It is a figure which shows an example of a causal relationship model update sequence when a user intervenes. It is a figure which shows an example of a causal relationship model update screen.

Hereinafter, a decision support system 101 and a decision support method according to an embodiment of the present invention will be described with reference to the drawings. In the present embodiment, description will be made by taking as an example the support for decision making in a national strategy of predicting the actions of the country and the organization and examining measures, but the present embodiment can be applied to other fields.

In the present embodiment, a decision maker who makes a decision on a national strategy, or a person who supports the decision making, determines whether or not an event affecting the strategy (for example, military action of another country) occurs before or at the time of occurrence. Alternatively, a decision support method and a decision support system 101 that provide information for supporting the decision of the own organization will be described.

The present system 101 has a relationship between a social occurrence event such as a natural disaster and a stock price crash, and a behavior of each country including each organization and each organization (when an event occurs, which event occurs. Influence) is stored as a causal relationship model, and the causal relationship model is used to predict the occurrence of an event and the response to it, and the next event to occur by simulation, and the decision maker (or intention) Decision supporter). In addition, the causal relationship model includes information from information sources that the system 101 sequentially acquires, such as public information such as social media and press information, and non-report information such as survey report information and coping information that the user of the system 101 holds. Since it is updated sequentially according to the analysis result of the public internal information, it is possible to always present the prediction of the event according to the latest information. The present system 101 is used to examine or determine the response of the own country or its organization before or at the time of occurrence of an event (for example, military action of another country) that is executed at any time and affects the strategy. Alternatively, it may be used for diagram training such as assuming that an event that has not occurred at present has occurred, inputting/executing this system, and examining what kind of correspondence should be taken.

FIG. 1 is a diagram showing an example of the overall configuration of the decision support method and system according to this embodiment.

As shown in FIG. 1, a decision support system 101 according to this embodiment includes a data collection/analysis unit 102, a causal relationship model updating unit 103, an update evaluation management unit 104, a causal relationship model management unit 105, The reaction prediction unit 106 includes a user management unit 107 and an interface unit 108.

The data acquisition/analysis unit 102 collects the information source 109 via the network 110 such as the Internet or a local network, analyzes the collected information, and generates information for updating the causal relationship model. The information source 109 includes public information such as social media and news information, and non-public internal information such as survey report information and coping information held by the user of the system 101.

The causal relationship model updating unit 103 uses the information generated by the data acquisition/analysis unit 102 to update the causal relationship model managed by the causal relationship model management unit 105, and the effectiveness of the update (for the user). Evaluate whether a meaningful update has taken place.

The update evaluation management unit 104 manages the content of the update performed by the causal relationship model updating unit 103 and the evaluation result thereof.

The causal relationship model management unit 105 determines a social occurrence event such as a natural disaster or a stock price crash, and a relationship such as the behavior of each country/organization including its own country/organization (what event occurs when a certain event occurs). Information that indicates whether it affects the occurrence of) is managed as a causal relationship model.

The reaction prediction unit 106 uses a causal relationship model managed by the causal relationship model management unit 105 to predict the occurrence of a certain event, the result of the response to the event, and the next occurring event by simulation.

The user management unit 107 manages users who use the decision support system 101. For example, the user (model manager) 111 and the user (decision maker) 112 perform access authentication when using the system 101. .. The user (model manager) 111 is a user who is in charge of updating and evaluating the causal relationship model, and the user (decision maker) 112 is a user who uses the system 101 to predict events that will occur in the future. is there.

The interface unit 108 receives input of information from the terminals used by the user (model manager) 111 and the user (decision maker) 112, and processes the terminals used by the user in each unit of the decision support system 101. Output the result. Here, the decision support system 101 and each user's terminal may be directly connected, or may be connected via a network such as a local network or the Internet.

FIG. 2 is a diagram showing an example of a functional configuration diagram of the decision support system 101 according to the present embodiment.

As shown in FIG. 2, in the decision support system 101 according to the present embodiment, the data collection/analysis unit 102 provides a data acquisition function 201 and a data analysis function 202.

The data acquisition function 201 is held by the user of the system 101 and public information such as social media and news information based on the data (acquisition source data) about the information source and the search condition registered in the data acquisition/analysis management DB 204. It acquires non-public internal information such as survey report information and countermeasure information to be acquired, and registers the acquired information in the acquired information DB 203.

The data analysis function 202 performs natural language analysis processing, co-occurrence analysis processing, and the like on the data registered in the acquisition information DB 203, generates information for updating the causal relationship model, and registers it in the analysis result management DB 205. To do.

In addition, the data acquisition function 201 registers the information input to the present system 101 in the occurrence event management DB 206 on the assumption that the user has generated an event that is not currently occurring. In addition, the data analysis function 202 registers the information registered in the analysis result management DB 205 as the latest information currently occurring in the incident event management DB 206.

The causal relationship model updating unit 103 provides an interest event registration function 211, an update information generation function 212, a model update evaluation function 213, a prediction result evaluation function 214, and an update information correction function 215.

The event of interest registration function 211 acquires an event of interest to the user (obtained via an event of interest setting screen 262 described later) and registers it in the event of interest management DB (event of interest storage unit) 217.

The update information generation function 212 uses the data registered in the analysis result management DB 205 to generate relationships (events) and events (nodes) newly registered in the causal relationship model and register them. Then, the causal relationship model after being updated is registered in the temporary update model DB 216 as a temporary update model.

The model update evaluation function 213 uses the event of interest (node) registered in the event of interest management DB 217 and the way of updating (for example, the relationship between events (nodes) not registered in the existing causal relationship model). Evaluate the updated contents based on new additions, etc.).

The prediction result evaluation function 214 acquires a simulation result using a model before and after updating (existing causal relationship model and temporary update model) (acquiring via a reaction prediction generation function 241 described later) and outputs it to the user (described later). It is presented via the model update screen 263).

The update information correction function 215 acquires the correction content for the model update content from the user (acquired via the model update screen 263 described later) and corrects the information registered in the temporary update model DB 216.

The update result management unit 104 provides an update log registration function 222 and an update log analysis function 221.

The update log registration function 222 registers the update content generated by the causal relationship model updating unit 103 in the model update management DB 223.

The update log analysis function 221 extracts the importance of the information source and the information to be searched from the data regarding the update content registered in the model update management DB 223 and the event of interest to the user registered in the interest event management DB 217, The data regarding the information source and the search condition registered in the data acquisition/management DB 204 is updated.

The causal relationship model management unit 105 provides a model registration/update function 231 and a model acquisition function 232.

Based on the evaluation result of the model update evaluation function 213 (effectiveness of model update), the model registration/update function 231 uses the data registered in the causal relationship model DB 233 as the past model in the case of a certain evaluation or more. The data stored in the model management DB 234 and registered in the temporary update model DB 216 is registered in the causal relationship model DB 233 (causal relationship model is updated).

The model acquisition function 232 acquires the data registered in the causal relationship model DB 233 and the past model management DB 234 and outputs it to another function.

The reaction prediction unit 106 provides a reaction prediction generation function 241 and a reaction prediction adjustment function 242.

The reaction prediction generation function 241 uses the data (causal relationship model) registered in the causal relationship model DB 233, the past model management DB 234, and the tentative update model DB 216, and then the result of occurrence of a certain event and the corresponding result, The phenomenon that will occur is predicted by simulation, and the result is registered in the reaction prediction result DB 243.

The reaction prediction adjustment function 242 acquires the correction content for the reaction prediction result from the user (acquired via the model update screen 263 described later), and the event registered in the temporary update model DB 216 so that the correction content becomes the prediction result. Correct the relationship between nodes (between nodes).

The user management unit 107 provides a user registration/authentication function 251.

The user registration/authentication function 251 registers information (user name, password, etc.) about a user who uses the decision support system 101 in the user management DB 252, and is responsible for updating and evaluating a causal relationship model (model manager). 111 and a user (decision maker) 112 who uses the system 101 to predict events that will occur in the future, etc., performs access authentication when using the system 101.

The interface unit 108 provides a menu selection screen 261, a phenomenon of interest setting screen 262, a model update screen 263, and a decision support screen 264 to the terminal used by the user.

The menu selection screen 261 is a screen for selecting a function to be used from the functions provided by the decision support system 101 for the user (model manager) 111 and the user (decision maker) 112, for example.

The event of interest setting screen 262 is a screen for each user to input an event of interest.

The model update screen 263 presents the information registered in the acquired information DB 203, inputs information from the user, presents the analysis result registered in the analysis result management DB 205, and registers the causal relationship model DB 233 and the temporary update model DB 216. It is a screen for presenting a causal relationship model and correction input from the user, and presenting the prediction result registered in the reaction prediction result DB 243 and correction input from the user.

The decision support screen 264 is a screen for selecting a causal relationship model to be used for prediction, executing a simulation, inputting corresponding contents from the user, and presenting a prediction result.

FIG. 3 is a diagram showing an example of a decision support sequence according to the present embodiment.

This sequence uses the causal relationship model in the decision support system 101 to predict the occurrence of a certain event and the response to it, and the next event to occur by simulation, and the decision maker (or decision supporter) It is a basic process for presenting to.

When the user (decision maker) 112 logs in to the decision support system 101 with the user ID and password registered in advance in the user management DB 252, the menu selection screen 261 is displayed. On the menu selection screen 261, interest event setting, model updating, and decision support can be selected. Here, when the user 112 selects decision support, a decision support screen 264 is displayed (details of the screen will be described later with reference to FIG. 6).

The user ID of the user 112 is passed to the model acquisition function 232 as a model list acquisition request, and the model acquisition function 232 uses the causal relationship model DB 233 (details will be described later with reference to FIG. 4) of the causal relationship model based on the user ID. The management table 410 is searched, a record (model ID, model name, update time) corresponding to the user ID is acquired as causal relationship model management data, and is output to the decision support screen 264 as a model name list.

When the user 112 selects a model to be used for this decision support from the model name list (causal relationship model management data) output on the decision support screen 264, the model ID of the selected model is the model acquisition function 232. Passed to. The model acquisition function 232 records the model ID passed from the causal relationship model table 400 of the causal relationship model DB 233 or the past model management DB 234 in which the past causal relationship model is stored based on the model ID passed. (Model ID, node ID, node name, subject, update time, link destination node, relationship degree, relationship update time) is acquired as causal relationship model data and output to the decision support screen 264.

On the decision support screen 264, when the user 112 selects an event (initial event) that occurs in step 0 from the causal relationship model data and executes prediction start, the node ID of the initial event is displayed in the reaction prediction generation function 241. Passed. The reaction prediction generation function 241 uses the causal relationship model data acquired by the model acquisition function 232 as the selected initial event as an event that occurs in step 0, and predicts an event that occurs after step 1 by simulation, The prediction result (subject, occurrence step, node ID, node name, probability) is output to the decision support screen 264 as prediction result data.

Also, the output prediction result data and the initial event are registered as a scenario in the reaction prediction result DB 243 in association with the causal relationship model data used in the simulation. When the user 112 executes the prediction start without selecting the initial event, the reaction prediction generation function 241 generates the latest event (subject, node name) registered in the occurrence event management DB 206 at step 0. As the event, the causal relationship model data acquired by the model acquisition function 232 is used to predict the event that occurs after step 1 by simulation.

On the decision support screen 264, when the user 112 changes an event (node) that occurs in a certain step from the prediction result data to another event, the step number and the node IDs of the event (node) before and after the change are It is passed to the reaction prediction generation function 241 as change event data. The reaction prediction generation function 241 replaces the portion of the output prediction result data corresponding to the event (node) before the change in the change event data with the event (node) after the change, and simulates the events that occur after the step number. And the prediction result data before the re-predicted step number, the initial event, and the prediction result (subject, occurrence step, node ID, node name, probability) are used as prediction result data in the decision support screen 264. Output to.

The prediction result data is registered as a scenario in the reaction prediction result DB 243 in association with the causal relationship model data used for the simulation. By repeatedly executing the event change by the user 112 and the re-prediction by the reaction prediction generation function 241, what is to occur next is predicted while changing the pattern in which various events occur, thereby supporting the user's decision making. Information can be provided. In addition, as a countermeasure against an event (prediction result) that occurs in a certain step, the user selects which event (behavior) to select in the next step and presents the event that occurs after that, thereby making a decision. Can also be trained.

FIG. 4 is a diagram showing an example of the causal relationship model DB 233 according to the present embodiment.

The causal relationship model DB 233 has a causal relationship model table 400 and a causal relationship model management table 410.

The causal relationship model table 400 includes a model ID 401, a node ID 402, a node name 403, a subject 404, an update time 405, a link destination node 406, a relationship degree 407, and a relationship update time 408. ..

The model ID 401 is for identifying the model. The node ID 402 is for identifying an event (node). The node name 403 indicates the content of the event. The subject 404 indicates the subject (subject) of the event. The update time 405 indicates the time when the content of the record is updated. The link destination node 406 is for registering a node ID of a node (node that may occur next) related to the node of the record. The degree of relationship 407 indicates the strength of the relationship (high probability of the next occurrence). The relationship update time 408 indicates the time when the related nodes and the strength of the relationship are updated.

After that, when there are a plurality of related nodes, the contents equivalent to the link destination node 406, the relation degree 407, and the relation update time 408 are registered in the link destination node (2) 409 and later. For example, in the record 420, when an event (node) "Country C regulates trade" occurs, an event (node) "Country B military action" with node ID "4" may occur. The relationship is shown.

The causal relationship model management table 410 includes a model ID 411, a model name 412, an update time 413, a user ID 414, and a storage location 415.

The model ID 411 is for identifying the model. The update time 413 indicates the time when the model is updated (newly added). The user ID 414 is for identifying the owner of the model. The storage location 415 indicates the location (local path, URL, etc.) where the model corresponding to the record is stored. If the storage location 415 is not registered, it is registered in the causal relationship model table 400 as the latest model.

FIG. 5 is a diagram showing an example of the reaction prediction result DB 243 according to the present embodiment.

The reaction prediction result DB 243 has a reaction prediction result table 500 and a reaction prediction result management table 510.

The reaction prediction result table 500 includes a scenario ID 501, a subject 502, an occurrence step 503, a node ID 504, a node name 505, and an occurrence probability 506.

The scenario ID 501 is for identifying a prediction result (scenario). The subject 502 indicates the subject (subject) of the event (node) that occurs. The occurrence step 503 indicates the timing of the event that occurs. The node ID 504 is for identifying an event (node) that occurs. The node name 505 indicates the content of the event. The occurrence probability 506 represents the probability that the event will occur. For example, the record 520 indicates that the event (node) “Country A protests” occurs at a probability of 30% at the timing of step 2.

The reaction prediction result management table 510 includes a scenario ID 511, a scenario name 512, a predicted time 513, a user ID 514, and a model ID 515.

Scenario ID 511 is for identifying a prediction result (scenario). Predicted time 513 indicates the time when the scenario of the record is generated (predicted). User ID 514 is for identifying the owner of the causal relationship model used in the prediction of the record. Model ID 515 is for identifying a causal relationship model.

FIG. 6 is a diagram showing an example of the decision support screen (initial event setting, prediction result display) 264 according to the present embodiment.

The decision support screen 264 includes a scenario name display unit 601, a scenario name change button 602, a reaction prediction result display unit 603, a model name display unit 604, a model read button 605, and a prediction start button 606. ..

When the user 112 presses the model name display unit 604, the model name list output by the model acquisition function 232 described above is displayed. When the user 112 selects a model to be used for decision making support this time from the displayed list and presses the model reading button 605, the causal relationship model data of the model selected by the model acquisition function 232 described above is used for decision making. It is output to the support screen. Next, when the user 112 selects an event (initial event) that occurs in step 0 and presses the prediction start button 606, the prediction result output by the reaction prediction generation function 241 described above is displayed as the reaction prediction result display unit 603. Is displayed in. For example, when the user 112 presses the event (action) 610 of country B in step 0 as an initial event, the events of country B that can be selected are displayed on the display unit 611, and "A country" is selected from the displayed events. When “Military invasion” is selected and the prediction start button 606 is pressed, the prediction result is displayed on the reaction prediction result display unit 603. As shown in the reaction prediction result 621, the events occurring in each step of each subject are displayed with a probability. For example, the occurrence event 622 in the reaction prediction result indicates that the probability that Country A protests in Step 1 is 70%.

Further, a scenario name (automatically generated by the decision support system) for identifying the reaction prediction result is displayed on the scenario name display unit 601. The scenario name can be freely changed by the user 112, and the scenario name can be changed by pressing the scenario name change button 602.

FIG. 7 is a diagram showing an example of a decision support screen (change of occurrence event, display of prediction result) 264 according to the present embodiment.

When the user 112 selects an event to be changed from the reaction prediction result 621 displayed on the reaction prediction result display unit 603 and presses the prediction start button 606, the prediction result output by the reaction prediction generation function 241 described above is displayed. , Is displayed on the reaction prediction result display unit 603. For example, when the event (behavior) 710 of country A in step 1 is pressed, selectable events of country A (other events that may occur in step 1) are displayed on the display unit 711. When the "ship dispatch" is selected and the prediction start button 606 is pressed, the prediction result obtained by re-estimating the steps after step 2 is displayed.

FIG. 8 is a diagram showing an example of an interesting event registration sequence according to the present embodiment.

As shown in FIG. 8, in the event-of-interest registration sequence, a causal relationship model and a reaction prediction result based on the causal relationship model are presented to the user, and the event (node) of interest is selected from the presented information by the user. Then, the interest value for each node of the causal relationship model is set and registered in the interest event management DB 217.

When the user (model manager) 111 logs in to the decision support system 101 with the user ID and password registered in the user management DB 252 in advance, the menu selection screen 261 is displayed. On the menu selection screen 261, interest event setting, model updating, and decision support can be selected. Here, when the user 111 selects the interested event setting, the interested event setting screen 262 is displayed (details of the screen will be described later with reference to FIGS. 12 and 13).

Next, the user ID of the user 111 is passed to the model acquisition function 232, and the model acquisition function 232 searches the causal relationship model management table 410 of the causal relationship model DB 233 based on the user ID and records corresponding to the user ID (model (ID, model name, update time) is acquired as causal relationship model management data and is output to the interested event setting screen 262.

When the user 111 selects the model to be used for the current interest event setting from the causal relationship model management data output to the interest event setting screen 262, the model ID of the selected model is passed to the model acquisition function 232. Based on the model ID, the model acquisition function 232 receives a record (model ID, node ID) corresponding to the model ID from the causal relationship model table 400 of the causal relationship model DB 233 or the past model management DB 234 in which the past causal relationship model is stored. , Node name, subject, update time, link destination node, degree of relationship, relationship update time) is acquired as causal relationship model data and is output to the interest event setting screen 262. At this time, the node name and subject whose interest value has already been registered in the interest event management DB 217 may be output together with the causal relationship model data.

From the causal relationship model data output on the interest event setting screen 262, the user 111 selects a node for which an interest value is to be set as an event of interest, and inputs the interest value (interest event setting for the model). Interest values may be input to multiple nodes.

Here, the user 111 can also register an interest value for the prediction result. The user 111 selects a means for registering the interest value with respect to the prediction result, and selects the initial event (event that occurs in step 0). As a result, the reaction prediction generation function 241 uses the causal relationship model data acquired by the model acquisition function 232 and the initial event (the latest event registered in the incident event management DB 206 when there is no input from the user). (Subject, node name)) is used to predict events that occur after step 1 by simulation, and the prediction results (scenario ID, subject, occurrence step, node ID, node name, probability) are used as prediction result data. Is output to the event of interest setting screen 262. The prediction result data is also registered in the reaction prediction result DB 243.

From the prediction result data output to the interest event setting screen 262, the user 111 selects an event (node) for which an interest value is to be set as an event of interest, and inputs the interest value (interest event setting for prediction result). ). Interest values may be input to multiple nodes.

When the user 111 performs interest event registration on the interest event setting screen 262, the interest event setting data (model ID, node ID, interest value) input in the interest event setting for the model described above, and the interest event setting for the prediction result The interest event setting data (scenario ID, node ID, interest value) input in (1) is passed to the interest event registration function 211 together with the user ID. The interest event registration function 211 registers the interest value for each node of the causal relationship model in the interest event management DB 217 according to the processing described later with reference to FIGS. 10 and 11.

FIG. 9 is a diagram showing an example of the interest event management DB 217 according to the present embodiment.

The interest event management DB 217 has an interest value management table 900. The interest value management table 900 includes a node ID 901, a node name 902, a subject 903, an update time 904, an interest value 905, a user ID 906, and a model ID 907.

The node ID 901 is for identifying an event (node). The node name 902 indicates the content of the event. The subject 903 indicates the subject (subject) of the event. The update time 904 is the time when the content of the record is updated. The interest value 905 is the interest value set in the node of the record. The user ID 906 is for identifying the user who has set the interest value. The model ID 907 is for identifying the model.

FIG. 10 is a flowchart showing an example of a processing procedure (interesting event setting for a model) of the interesting event registration function 211 according to the present embodiment.

The interest event registration function 211 receives interest event setting data (model ID, node ID, interest value) for the model and interest event setting data (scenario ID, node ID, interest value) for the prediction result from the interest event setting screen 262. Then, the interest value for each node of the causal relationship model is registered in the interest event management DB 217 according to the following processing.

First, in S1001, the interest event registration function 211 confirms whether interest event setting data (model ID, node ID, interest value) for the model is received, and if so, executes S1002 and thereafter. If not received, S1101 and subsequent steps in FIG. 11 are executed.

In S1002, the interested event registration function 211 selects one interested event setting data that has not been processed by the present processing procedure from the received interested event setting data (model ID, node ID, interest value) for one or more models. Extract.

In S1003, the interested event registration function 211 uses the model ID of one interested event setting data extracted in S1002, the node ID, the interested value, and the user ID received together with the interested event setting data as the interested event data (record). The current event (update time) is registered together with the event of interest management DB 217 (if a record with the same node ID already exists, the current data (record) is overwritten).

In S1004, the interesting event registration function 211 searches the item “link destination node” (406, 409) of the causal relationship model table 400 (FIG. 4) and finds the node ID of the interesting event data (record) registered in S1003. A record (factor record) that has the link destination node ID is extracted. For example, in FIG. 4, since the link destination node of the “military exercise” with the node ID “5” is “2”, the “military exercise” is the cause record of the “protest” with the node ID “2”.

In S1005, the interested event registration function 211 executes S1006 and subsequent steps when the factor record exists in S1004, and executes S1010 and subsequent steps when the factor record does not exist.

In S1006, the interest event registration function 211 extracts the degree of relation corresponding to the link destination node in the factor record extracted in S1004, and the interest value registered in the interest event management DB 217 in S1003 or S1008 indicates the degree of relation (%). The value multiplied by the value (rounded down to 10) is calculated as the interest value. In FIG. 4, the cause record “military exercise” of “protest” with the node ID “2” has a degree of relationship of 50 with “protest”, so that the user 111 sets an interest value of 50 to “protest”, for example. In that case, the interest value of “military exercise” is 50×50 (%)=20 (rounded down to the nearest whole number).

In S1007, the interest event registration function 211 executes S1008 and subsequent steps when there is a factor record in which the interest value calculated in S1006 is other than 0, and executes S1010 and subsequent steps when they do not exist (when all are 0). ..

In S1008, the interest event registration function 211 sets the node ID of the node for which the interest value is calculated in S1006 and the interest value (for example, the interest value “20” in the military exercise of the node ID “5”) in the interest event data (record). Is registered in the interest event management DB 217 (if the same node ID already exists, the interest value is added (upper limit 100)).

In S1009, a cause record having the node ID of the record registered in S1008 as a link destination node is extracted from the causal relationship model table 400, and S1005 and subsequent steps are executed. By repeating S1005 to S1009, it is possible to set the interest value while tracing the causal relationship of the nodes (the interest value becomes smaller as the trace goes back).

In S1010, if the interested event setting data (model ID, node ID, and interest value) for the received model remains that has not been processed by this processing procedure, S1002 and subsequent steps are executed. If there is no remaining, S1101 and subsequent steps in FIG. 11 are executed.

FIG. 11 is a flowchart showing an example of a processing procedure (interesting event setting for a prediction result) of the interested event registration function 211 according to the present embodiment.

When the interest event registration function 211 receives the interest event setting data (scenario ID, node ID, interest value) for the prediction result from the interest event setting screen 262, the interest value registration function 211 sets the interest value for each node of the causal relationship model according to the following processing. , Is registered in the interest event management DB 217.

First, in S1101, the interest event registration function 211 confirms whether interest event setting data (scenario ID, node ID, interest value) for the prediction result is received, and if so, executes S1102 and thereafter. If not received, the process ends.

In S1102, the interest event registration function 211 sets the interest event setting data (scenario ID, node ID, interest value) for the received one or more prediction results to the interest event setting data not yet processed by this processing procedure. Extract one.

In S1103, the interest event registration function 211 uses the scenario ID of one interest event setting data extracted in S1102, the node ID, the interest value, and the user ID received together with the interest event setting data as interest event data (record). As the current time (update time), it is registered in the interest event management DB 217 (if a record with the same node ID already exists, it is overwritten with the current data (record)). Regarding the scenario ID, the model ID is acquired by referring to the reaction prediction result management table 510, and the model ID is registered in the interest event management DB 217.

In S1104, the interesting event registration function 211 extracts a record of the node ID of the reaction prediction result table 500 (FIG. 5) based on the scenario ID and the node ID of the interesting event setting data extracted in S1102.

In S1105, the interested event registration function 211 responds to the record having the scenario ID of one interested event setting data extracted in S1102 and having the occurrence step value −1 of the record extracted in S1104 as the occurrence step value. Whether or not it exists in the prediction result table 500 is confirmed. If it exists, S1106 and subsequent steps are executed, and if it does not exist, S1110 and subsequent steps are executed. For example, if the record extracted in S1104 is the “military exercise” with the node ID “5” in the reaction prediction result table 500, the generation step is “3”, and thus the record with the generation step “2”. , There are three records with node IDs “2”, “3”, and “4”.

In S1106, the interested event registration function 211 extracts the node ID of the record having the occurrence step value -1 of the record extracted in S1104 or S1109 as the occurrence step value, and registers it in the interested event management DB 217 in S1103 or S1108. The interest value×½ value×the value calculated by the occurrence probability of the record extracted in the previous step (rounding down to one place) is calculated as the interest value. For example, if the record extracted in S1104 is “military exercise” with the node ID “5” in the reaction prediction result table 500, the generation step is “3”, and thus the record with the generation step “2”. , There are three records with node IDs “2”, “3”, and “4”. In this case, since the probability of occurrence of “military exercise” is 50%, for example, the interest value of “military action” of the node ID “4” is set by the user 111 to “military exercise”. In this case, 90×1/2×50(%)=20 (rounded down to one place).

In S1107, the interest event registration function 211 executes S1108 and subsequent steps for a record in which the calculated interest value is other than 0 when there is a record in which the interest value calculated in S1106 is not 0, and when all are 0, S1110. Do the following.

In S1108, the interest event registration function 211 sets the node ID and interest value (for example, the interest value “20” of the military action of the node ID “4”) of the record for which the interest value is calculated in S1106 as interest event data (record). Is registered in the interest event management DB 217 (if the same node ID already exists, the interest value is added (upper limit 100)).

In S1109, the interest event registration function 211 extracts the record of the node ID in the reaction prediction result table 500 for the node ID of the record in which the interest value is calculated in S1106, and executes S1105 and subsequent steps. By repeating S1105 to S1109, it is possible to set the interest value while tracing back the prediction result (the interest value becomes smaller as going back).

In S1110, the interest event registration function 211 determines whether there is any interest event setting data (scenario ID, node ID, interest value) for the received prediction result that has not been processed by this processing procedure. Executes S1102 and subsequent steps, and if there is no remaining, ends the processing.

FIG. 12 is a diagram showing an example of an interest event registration screen (interest event setting for a model) 262 according to the present embodiment.

The interest event registration screen 262 includes a menu return button 1201 for returning to the menu screen, a model name display section 1202, a model display button 1203, a prediction result display button 1204, a search input section 1205, and a search button 1206. An interest event setting unit 1207 and an interest value registration button 1208 are included.

When the user 111 presses the model name display unit 1202, the model name list output by the model acquisition function 232 described above is displayed. When the user 111 selects a model to be used for the current event registration from the displayed list and presses the model display button 1203, the causal relationship model data of the selected model is displayed by the model acquisition function 232. It is displayed on the event of interest setting unit 1207. Next, when the user 111 presses an event for setting an interest value from the events (nodes) displayed in the interest event setting unit 1207, the interest value registration unit 1209 is displayed and the user 111 is interested in the event. You can enter a value. When the user 111 completes the input and presses the interest value registration button 1208, the interest event registration function 211 described above calculates the interest value set for each event and the interest value of each related event. It is registered in the event management DB 217.

When the scale of the model is large, a search keyword is input to the search input unit 1205 and the search button 1206 is pressed, so that the event (node) searched based on the search keyword is displayed in the center of the interest event setting unit 1207. You can display and enter the interest value.

FIG. 13 is a diagram showing an example of an interesting event registration screen (interesting event setting for a prediction result) 262 according to the present embodiment.

When the user 111 presses the model name display unit 1202, the model name list output by the model acquisition function 232 described above is displayed. When the user 111 selects a model to be used for the current event registration from the displayed list and presses the prediction result display button 1204, the reaction prediction result data output by the reaction prediction generation function 241 described above is displayed. It is displayed on the event of interest setting unit 1207. Next, when the user 111 presses an event for setting an interest value from the events (nodes) displayed in the interest event setting unit 1207, the interest value registration unit 1301 is displayed and the user 111 is interested in the event. You can enter a value. When the user 111 completes the input and presses the interest value registration button 1208, the interest value registration function 211 described above causes the interest value set for each event and the interest value of each event that occurred before the occurrence of the event. It is calculated and registered in the interest event management DB 217.

For example, as shown in FIG. 13, when “military exercise” in step 2 is designated as an event of interest (for example, interest value 90), the interest value 90 is set to the node (military exercise) of the causal relationship model (interest event of interest). Registered in the management DB 217). For the prediction result "military action" in the previous step (step 1), the probability of military exercise is multiplied by 70%, and 90 x 1/2 x 0.7 = 30 (rounded down to the nearest 1) is set. To be done. For the node (trade control) at the immediately preceding step (step 0), the probability of occurrence of military action (interest value 30) is multiplied by 80%, and 30×1/2×0.8=10 is set. It

FIG. 14 is a diagram showing an example of a causal relationship model update sequence according to this embodiment.

As shown in FIG. 14, in the causal relationship model update sequence, data such as news articles are acquired from an information source on a regular basis or according to an instruction from a user, and an event (node) for model update and the node Performs data analysis such as relationship extraction, generates update information, evaluates the effectiveness of the update information, updates the model when the effectiveness is above a certain level, and suspends the update when the effectiveness is below a certain level. ..

The data acquisition function 201 periodically (for example, every three hours) or in response to a request from the user (model manager) 111, acquires information source data (information source ID) from the acquisition/analysis management DB 204 regarding the information source of the data acquisition target. , URL, etc.) and search data (search ID, keyword) for information search, and based on these data, information for updating the causal relationship model is obtained from the information source 109 described above. Regarding the information source that has acquired the information, the acquisition time is registered in the corresponding record of the acquisition/analysis management DB 204. Further, the acquired information is registered in the acquired information DB 203.

The data analysis function 202 performs noun extraction and subject predicate extraction by natural language analysis, co-occurrence number extraction by co-occurrence analysis, etc. on the acquired information registered in the acquired information DB 203, and registers the analysis result in the analysis result management DB 205. To do. The subject predicate extraction result is registered in the occurrence event management DB 206 as an event that is currently occurring.

The update information generation function 212 extracts the relationship between the events (nodes) by using the analysis result registered in the analysis result management DB 205 and the acquisition time of each piece of acquired information registered in the acquired information DB 203. The relationship between the event (node) to be updated and the event (node to node) is registered in the temporary update model DB 216 as update target data. Further, the causal relationship model data to be updated this time is acquired from the causal relationship model DB 233, and the updated temporary updated model data is generated based on the update target data and registered in the temporary updated model DB 216.

The model update evaluation function 213 uses the temporary update model data (update data) registered in the temporary update model DB 216 to update patterns of each event (node) (for example, between events not registered in the existing causal relationship model). The updated content is evaluated based on the relationship (between nodes) being newly added). In addition, the update content is evaluated depending on whether or not the event of interest (node) registered in the event of interest management DB 217 has been updated. The update effectiveness is calculated based on these evaluation results.

The model registration/updating function 231 uses the data registered in the causal relationship model DB 233 as a past model based on the update effectiveness calculated by the model update evaluation function 213 when the update effectiveness is a certain value or more. The data stored in the model management DB 234 and registered in the temporary update model DB 216 is registered in the causal relationship model DB 233 (causal relationship model is updated).

The update log registration function 222 registers the update content registered in the temporary update model DB 216 in the model update management DB 223.

The update log analysis function 221 extracts the importance of the information source and the information to be searched from the update content registered in the model update management DB 223 and the data related to the event of interest to the user registered in the interest event management DB 217, The data regarding the information source and the search condition registered in the acquisition/analysis management DB 204 is updated.

FIG. 15 is a diagram showing an example of the acquisition/analysis management DB 204 according to the present embodiment.

The acquisition/analysis management DB 204 has an information source management table 1500 and a search management table 1510.

The information source management table 1500 includes an information source ID 1501, an information source (URL) 1502, an update time 1503, a previous acquisition time 1504, an importance 1505, and a user ID 1506.

The information source ID 1501 is for identifying the information source. The information source (URL) 1502 is for identifying the location of the information source. The update time 1503 indicates the time when each record was updated. The last acquisition time 1504 indicates the time when the information was acquired from the information source of the record. The importance 1505 indicates the importance of each information source. The user ID 1506 is for identifying a user who uses information collection from the information source of the record. Here, in the information source (URL) 1502, a storage location on the user's local server (a folder in which articles and report documents held by the user are stored) may be registered.

The search management table 1510 includes a search ID 1511, a keyword 1512, an update time 1513, an importance 1514, and a user ID 1515.

The search ID 1511 is for identifying a search keyword. The keyword 1512 is for use in search. The update time 1513 indicates the time when each record was updated. The importance level 1514 indicates the importance of each keyword. The user ID 1515 is for identifying a user who uses information collection using the keyword of the record.

FIG. 16 is a diagram showing an example of the acquisition information DB 203 according to this embodiment.

The acquisition information DB 203 has an acquisition information table 1600.

The acquisition information table 1600 includes an update ID 1601, an article ID 1602, an article content 1603, an information source (URL) 1604, an acquisition time 1605, a status 1606, and a user ID 1607.

The update ID 1601 is for identifying at which update the information was acquired. The article ID 1602 is for identifying each piece of acquired information. The article content 1603 is for registering the content of the acquired information. The information source (URL) 1604 indicates the information source that acquired each information. The acquisition time 1605 indicates the time when each piece of information was acquired. The status 1606 indicates whether each piece of information has been processed by the data analysis function 202. The user ID 1607 is for identifying the user who uses the information of the record.

FIG. 17 is a flowchart showing an example of the processing procedure of the data acquisition function 201 according to this embodiment.

The data acquisition function 201 receives data such as a news article from an information source with a certain degree of importance or higher at regular intervals (for example, every 3 hours) or in response to a request from the user (model manager) 111 according to the following processing. get. Also, information (articles and sentences) is acquired by performing a search on the Internet with keywords having a certain degree of importance or higher. An ID is issued for each acquired article and registered in the acquisition information DB 203.

When the data acquisition function 201 is activated, an update ID for this update is issued in step S1701 in order to identify the update.

In step S1702, the data acquisition function 201 determines that the importance 1505 is a certain value or more (for example, 30 or more) in the information source management table 1500 of the acquisition/analysis management DB 204 (FIG. 15) or a certain time or more from the last acquisition time 1504. Records of the progress (for example, 3 months or more) are extracted.

In step S1703, the data acquisition function 201 acquires information (article) from the information source (URL) 1502 on the Internet and the local network for the record extracted in step S1702. For the information source (URL) from which information has been acquired, the data acquisition function 201 updates the previous acquisition time 1504 of the record in the information source management table 1500 to the current time.

In S1704, the data acquisition function 201 issues an article ID for identification for each information (article) acquired in S1703, and becomes the update ID issued in S1701, the content of the acquired information (article content), and the acquisition source. When managing the information source (URL), the acquisition time, and the update timing and update contents for each user, the information is registered in the acquisition information table 1600 (FIG. 16) together with the user ID. Here, "unprocessed" is registered in each of the statuses 1606 of the acquisition information table 1600.

In step S1705, the data acquisition function 201 extracts, from the search management table 1510 of the acquisition/analysis management DB 204, a record whose importance level 1514 is a certain value or more (for example, 30 or more).

In step S1706, the data acquisition function 201 searches the Internet and local network for the record extracted in step S1705 based on the keyword 1512, and acquires information (article).

In S1707, the data acquisition function 201 issues an article ID for identification for each information (article) acquired in S1706, and indicates the update ID issued in S1701, the content of the acquired information (article content), and the acquisition source. When managing the information source (URL), the acquisition time, the update timing and the update content for each user, the information is registered in the acquisition information table 1600 together with the user ID. Here, "unprocessed" is registered in each of the statuses 1606 of the acquisition information table 1600.

In step S1708, the data acquisition function 201 determines whether it is the update timing (for example, whether one month has passed since the previous update (issue of update ID)). If it is the update timing, the process is ended, and if it is not the update timing, S1702 and the subsequent steps are executed after a certain period of time (for example, 3 hours). The user (model manager) 111 may set the update timing and the timing of returning to S1702.

For example, if the data acquisition function 201 is started immediately after the series of processes of the causal relationship model update sequence is finished and the update timing is set to 1 month and the acquisition time is set to 3 hours, the latest information is acquired every 3 hours. That is, the causal relationship model update process is performed every month (that is, the processes of S1702 to S1708 are performed 240 times during the update).

FIG. 18 is a diagram showing an example of (a) analysis result management DB 205 and (b) occurrence event management DB 206 according to this embodiment.

The analysis result management DB 205 has an analysis result management table 1800 and a co-occurrence result management table 1810.

The analysis result management table 1800 includes an update ID 1801, an article ID 1802, a noun 1803, and a subject predicate set 1804.

The update ID 1801 is for identifying at which update the analysis result is obtained. The article ID 1802 is for identifying each piece of information that is the basis of analysis. The noun 1803 is for registering a noun extracted from each information (article). The subject predicate set 1804 is for registering a set of subject and predicate extracted from each information (article).

The co-occurrence result management table 1810 includes an update ID 1811, a subject predicate set (1) 1812, a subject predicate set (2) 1813, and a co-occurrence count 1814.

The update ID 1811 is for identifying at which update the analysis result is obtained.
The subject predicate set (1) 1812 and the subject predicate set (2) 1813 are for registering subject predicate sets having a co-occurrence relationship (appearing in the same article). The co-occurrence count 1814 is for registering the co-occurrence count.

The event management DB 206 has an event management table 1820.

The incident event management table 1820 includes an article ID 1821, a subject predicate set 1822, and an acquisition time 1823.

The article ID 1821 is for identifying the original information (article) in which the subject predicate set was described. The subject predicate set 1822 is for registering a set of subjects and predicates (events occurring) extracted from each information (article). The acquisition time 1823 indicates the time when the original information (article) in which the subject predicate set was described was acquired.

FIG. 19 is a flowchart showing an example of the processing procedure of the data analysis function 202 according to this embodiment.

The data analysis function 202 extracts nouns and subject predicates by natural language analysis and co-occurrence number extraction by co-occurrence analysis for the acquired information (articles and sentences) registered in the acquired information DB 203 (FIG. 16) according to the following processing. And so on. Further, the subject predicate extraction result is registered in the occurrence event management DB 206 (FIG. 18) as an event currently occurring.

When the data analysis function 202 is activated, in step S1901, the acquisition information table 1600 is searched and one record whose status 1606 is “unprocessed” is extracted.

In S1902, the data analysis function 202 acquires the article content 1603 for the record extracted in S1901.

In S1903, the data analysis function 202 extracts a noun from the article content acquired in S1902 by natural language analysis.

In step S1904, the data analysis function 202 extracts a set of subjects and predicates (for example, "Country A" protested") from the article content acquired in step S1902 by natural language analysis.

In step S1905, the data analysis function 202 analyzes the noun and subject predicate set extracted in steps S1903 and S1904, along with the article ID of the extracted original information (article) and the update ID indicating the current update, and the analysis result management table 1800 (see FIG. Register in 18). Further, in the acquisition information table 1600, the status of the record extracted in S1901 is updated to "processed". Further, the registered subject predicate set is registered in the generated event management table 1820 of the generated event management DB 206 together with the article ID and the acquisition time of the extracted original information (article) as the currently occurring event.

In S1906, the data analysis function 202 refers to the acquisition information table 1600 and determines whether or not there is a record whose status 1606 is “unprocessed”. If the data analysis function 202 remains, the processing after S1901 is executed, and if not, the processing after S1907 is executed.

In S1907, co-occurrence analysis is performed on all the article contents acquired in S1902, and the number of times (co-occurrence number) that appears at the same time is extracted for all the subject predicate sets extracted in S1904.

In S1908, the data analysis function 202 uses the co-occurrence information (two or more subject predicate sets (for example, (Country B, military invasion) and (Country A, protest)) and co-occurrence count) extracted in S1907 as co-occurrence information. It registers in the origination result management table 1810. For example, the co-occurrence information to be registered may be narrowed down, such as registering only a set of subject predicates having a certain number of co-occurrences (eg, 5 or more).

FIG. 20 is a diagram showing an example of the temporary update model DB 216 according to this embodiment.

The temporary update model DB 216 has an update target data table 2000 and a temporary update model data table 2010.

The update target data table 2000 includes a temporary node ID 2001, a node name 2002, a subject 2003, an update status 2004, an article ID 2005, a link destination temporary node 2006, and a degree of relation 2007.

The temporary node ID 2001 is for identifying the event (node) to be updated this time. The node name 2002 indicates the content of the event. The subject 2003 indicates the subject (subject) of the event. The update status 2004 indicates whether to update the event (node) of the record. The article ID 2005 is for identifying each piece of information that is the source of the event (node) of the record. The link destination temporary node 2006 is for registering the temporary node ID of a node (node that may occur next) related to the node of the record. The degree of relationship 2007 and the strength of the relationship (high probability of next occurrence) are shown. After that, when there are a plurality of related nodes, the contents equivalent to the link destination temporary node 2006 and the relation degree 2007 are registered in the link destination temporary node (2) 2008 and later.

The temporary update model data table 2010 includes a node ID 2011, a node name 2012, a subject 2013, an update pattern 2014, a link destination node 2015, and a degree of relationship 2016.

The node ID 2011 is for identifying an event (node). The node name 2012 indicates the content of the event. The subject 2013 indicates the subject (subject) of the event. The update pattern 2014 indicates an update pattern of the event (node) of the record (for example, a relationship between events (nodes) that has not been registered in the existing causal relationship model is newly added). The link destination node 2015 is for registering the node ID of a node (node that may occur next) related to the node of the record. The degree of relationship 2016 indicates the strength of the relationship (high probability of the next occurrence). After that, when there are a plurality of related nodes, the contents equivalent to the link destination node 2015 and the relation degree 2016 are registered in the link destination node (2) 2017 and thereafter.

FIG. 21 is a flowchart showing an example of the processing procedure of the update information generation function according to this embodiment.

The update information generation function 212 extracts the set of subjects and nodes (subject predicate set) obtained from the information source this time and their relationship from the analysis result management DB 205 (FIG. 18) according to the following processing (information (article)). (Extracted from the order in which they appear, the time relationship between articles, and the number of co-occurrences in all articles) to generate update target data (update target data table 2000), and the existing causal relationship model and update target. The temporary update model data (temporary update model data table 2010) is generated from the following data.

When the update information generation function 212 is activated, the subject predicate set 1804 and the article ID 1802 of each record are extracted from the analysis result management table 1800 in S2101. Here, the update ID of the record whose update status 2217 of the update content management table 2210 (FIG. 22) described later is “pending” and the record of the analysis result management table 1800 corresponding to the article ID are also extracted as the analysis target. You may do it.

In S2102, the update information generation function 212 issues a temporary node ID for each of the subject predicate sets extracted in S2101 as update target nodes (sets of subject and node).

In step S2103, the update information generation function 212 extracts the relationship between the update target node set in step S2102 and the subject predicate set having the same article ID in the analysis result management table 1800 (for example, the relationship between the context in the article The subject predicate set that comes out from is set as the link destination of the subject predicate set that comes out earlier), and the provisional node ID and the link destination provisional node ID are set. For example, in the record 1805 of article ID “1”, when “Country A, protest” appears after “Country B, military invasion”, “Country A, protest” is “Country B, military.” Link to "Invade".

In S2104, the update information generation function 212 extracts a record (related record) including a certain number or more of the same nouns from the nouns 1803 of each record of the analysis result management table 1800. For example, the record 1805 with the article ID “1” and the record 1806 with the article ID “2” have three or more same nouns (Z territory, country A, country B), and are therefore determined to be related.

In S2105, the update information generation function 212 extracts the relationship between the subject predicate sets of the related record determined in S2104 (refer to the acquisition time 1605 of the acquisition information table 1600 from the article ID of each record, and temporally before and after). The relationship is extracted), and the temporary node ID and the link destination temporary node ID are set. For example, “Country C, mediate” in the record 1806 with the article ID “2” is the link destination of “Country A, protest” in the record 1805 with the article ID “1”.

In S2106, the update information generation function 212 extracts the strength of the relationship between the subject predicate sets from the co-occurrence result management table 1810 (the higher the number of co-occurrence, the stronger the relationship), and the temporary node ID and the link destination temporary. The degree of relation of node IDs is set (for example, the number of co-occurrence=degree of relation).

In S2107, the update information generation function 212 updates the subject and node set (subject predicate set) extracted in S2102 to S2106, the temporary node ID and its relationship (link destination temporary node ID, relationship degree), and the article ID. Register in the data table 2000.

In step S2108, the update information generation function 212 integrates the duplicated records (having the same node name as the subject) in the update target data table 2000 into one (when there are different link destination nodes, the links are linked. Connect to the destination node (2) and later). For example, since the nodes extracted from the article IDs “1” and “3” in the analysis result management table 1800 are duplicated, they are combined into one (delete one), and the update target data is set as the record 2021 and the record 2022. Registered in the table 2000. Here, prepare a dictionary file that determines duplicate vocabulary (a file that describes rules for guessing that the vocabulary is the same), and for example, determine that military invasion and territorial invasion are the same, By a similar process, the record with the node name “territorial invasion” may be integrated with the record with the node name “military invasion”.

In step S2109, the update information generation function 212 sets the update status to “pending” for the record that has no link destination node or is not a link destination node of any record in the update target data table 2000. As a result, nodes that are not connected to any of the nodes (have no causal relationship) are meaningless nodes that do not affect the prediction result, so exclude them from the update target and prevent the size of the updated model from increasing. ..

In step S2110, the update information generation function 212 sets the update status of the record whose update status 2004 is not “pending” to “update” in the update target data table 2000. Furthermore, the update information generation function 212 compares the current causal relationship model table 400 (FIG. 4, hereinafter, existing model) and performs the following processing to register the temporary updated model data.
(1) First, the contents of all records except the model ID 401 of the existing model are directly registered in the temporary update model data table 2010. For the update pattern 2014, all "0" (no update) is registered.
(2) Among the set of the subject and the node name of the update target data table 2000, those that do not exist in the existing model issue a new node ID and register it in the temporary update model data table 2010. In this newly registered record, if the link destination node (subject thereof, node name) does not exist in the existing model (new node ID is issued), the link destination node and degree of relation are also added. Then, “1” is registered in the update pattern 2014. That is, if the newly registered record and its link destination node (subject thereof, node name) do not exist in the existing model, “1” is registered in the update pattern 2014. On the other hand, if the newly registered record does not exist in the existing model, but the link destination node (main body, node name) of which exists in the existing model, as the link destination node, the node ID of the existing model, The degree of relationship extracted this time is added, and “2” is registered in the update pattern 2014. Also, although the newly registered record does not have a link destination node, that record is a link destination node of the existing model or another record of the update target data table 2000 (node name, only temporary update model of main body). For those registered in the data table 2010), “3” is registered in the update pattern 2014.
(3) Of the set of subject and node name in the update target data table 2000, it exists in the existing model (registered in the temporary update model data table 2010 in (1) above), and (the subject of the link destination node , Node names) are also the same (existing in the existing model), the degree of relation of the record in the temporary update model data table 2010 is updated (for example, the degree of relation is +10), and the update pattern 2014 shows 4” is registered. Further, of the set of the subject and the node name of the update target data table 2000, it exists in the existing model, but the link destination node (the subject of the link, the node name) does not exist in the existing model (a new node ID is issued. ), the link destination node and the degree of relation are added, and “5” is registered in the update pattern 2014.

FIG. 22 is a diagram showing an example of the model update management DB 223 according to this embodiment.

The model update management DB 223 has an update effectiveness management table 2200 and an update content management table 2210.

The update effectiveness management table 2200 is composed of an update ID 2201, a model ID 2202, an update time 2203, an update effectiveness 2204, and a user ID 2205.

The update ID 2201 is for identifying which update time is the result of effectiveness evaluation. The model ID 2202 is for identifying the updated model. The update time 2203 indicates the time when the model is updated. The update effectiveness 2204 is for registering the effectiveness evaluation result. The user ID 2205 is for identifying the owner of the model.

The update content management table 2210 includes an update ID 2211, a node ID 2212, a node name 2213, a subject 2214, an update pattern 2215, an article ID 2216, a link destination node 2217, and a degree of relationship 2218.

The update ID 2211 is for identifying at which update time it was updated. The node ID 2212 is for identifying a node. The node name 2213 indicates the content of the event. The subject 2214 indicates the subject (subject) of the event. The update pattern 2215 shows the update pattern of the event (node) of the record. It is for identifying the article ID 2216 and each piece of information that is the source of the event (node) of the record. The update status 2217 indicates whether or not the event (node) of the record has been updated. The link destination node 2218 is for registering the node ID of a node (node that may occur next) related to the node of the record. The degree of relationship 2219 indicates the strength of the relationship (high probability of the next occurrence). After that, when there are a plurality of related nodes, the contents equivalent to the link destination node 2218 and the relation degree 2219 are registered in the link destination node (2) 2220 and thereafter.

FIG. 23 is a flowchart showing an example of the processing procedure of the model update evaluation function 213 and the model registration/update function 231 according to this embodiment.

The model update evaluation function 213 calculates the update effectiveness for the update content of this time according to the following processing.

When the model update evaluation function 213 is activated, in S2301, the update pattern 2014 of the temporary update model data table 2010 of the temporary update model DB 216 (FIG. 20) described above is searched, and the record whose update pattern is other than “0” (no update) To extract.

In S2302, the model update evaluation function 213 confirms whether or not there is a record extracted in S2301, and if there is, executes the processing of S2303 and thereafter, and if not, executes S2307 and thereafter.

In S2303, the model update evaluation function 213 acquires an interest value from the interest value management table 900 of the interest event management DB 217 (FIG. 9) for each record extracted in S2301 based on the node ID.

In S2304, the model update evaluation function 213 calculates the sum of interest values of each record acquired in S2303, and calculates the update effectiveness level. For example, the sum of interest values×0.1 is set as the update effectiveness.

In S2305, the model update evaluation function 213 acquires an update pattern for each record extracted in S2301 and calculates the number of records corresponding to each of the update patterns 1 to 5.

In S2306, the model update evaluation function 213 weights each update pattern based on the number calculated in S2305 to calculate the update effectiveness. For example, (the number of update patterns “2”×5×0.1)+(the number of update patterns “5”×3×0.1)+(the number of update patterns “1”×2×0.1)+ Let (the number of update patterns “3”×1×0.1)+(the number of update patterns “4”×0.5×0.1) be the update validity. As a result, when a new causal relationship is added, it is possible to make an evaluation according to the way of updating, such as the effectiveness of updating being high.

In S2307, the model update evaluation function 213 adds the respective update effectiveness calculated in S2304 and S2306 to calculate the update effectiveness for this update. If no record is extracted in S2301, the update validity is set to "0". Further, the update ID of this time and the calculated update effectiveness are registered in the update effectiveness management table 2200 of the model update management DB 223 (FIG. 22).

Next, the model registration/update function 231 updates the causal relationship model based on the update effectiveness calculated by the model update evaluation function 213 when the update effectiveness is equal to or higher than a certain value according to the following processing.

In S2308, the model registration/updating function 231 confirms whether or not the update validity calculated in S2307 is a certain value or more, and if it is more than a certain value (for example, 30 or more), executes S2309 and thereafter, and otherwise. In the case of, S2310 and subsequent steps are executed.

In S2309, the model registration/update function 231 issues a new model ID and saves the data registered in the causal relationship model table 400 of the causal relationship model DB 233 (FIG. 4) in the past model management DB 234 as a past model. Then, the data registered in the temporary update model data table 2010 of the temporary update model DB 216 (FIG. 20) is registered in the causal relationship model table 400 of the causal relationship model DB 233 (causal relationship model is updated). In addition, the newly issued model ID is registered in the causal relationship model management table 410 together with the current time (update time) (the model name is arbitrarily created by the system or registered according to the naming rule defined by the user 111).

In S2310, the model registration/update function 231 registers the model ID issued in S2309 and the update time in the record corresponding to the current update ID in the update effectiveness management table 2200 of the model update management DB 223 (FIG. 22). .. If S2309 is not executed, "NULL" is registered in the model ID 2202 and the current time is registered in the update time 2203.

FIG. 24 is a flowchart showing an example of the processing procedure of the update log registration function 222 and the update log analysis function 221 according to this embodiment.

The update log registration function 222 registers the current update content registered in the temporary update model DB 216 (FIG. 20) in the model update management DB 223 (FIG. 22) according to the following process.

When the update log registration function 222 is activated, in S2401, records with update patterns other than 0 (no update) are extracted from the temporary update model data table 2010. Also, the article ID that is the source of the update of the record is extracted from the update target data table 2000.

In S2402, the update log registration function 222 confirms whether or not there is a record extracted in S2401, and if there is, executes S2403 and subsequent steps, and if not, executes S2404 and subsequent steps.

In S2403, the record extracted in S2401 is added to the update content management table 2210 together with the article ID and the current update ID.

In S2404, the update log registration function 222 extracts from the update target data table 2000 a record whose update status is “pending”.

In step S2405, the update log registration function 222 confirms whether or not there is a record extracted in step S2404. If there is the record, step S2406 or later is executed, and if not, step S2407 or later is executed.

In S2406, the update log registration function 222 adds the record extracted in S1204 to the update content management table 2210 together with the article ID of this record and the update ID of this time.

In S2407, the update log registration function 222 confirms whether or not there is a record added to the update content management table 2210 in S2403 and S2406, and if there is, executes S2408 and subsequent steps; To execute.

Next, the update log analysis function 221 extracts the importance of the information source and the information to be searched from the data regarding the update content and the event of interest to the user according to the following processing, and acquires and analyzes the management DB 204 (FIG. 15). ) Is updated. Further, the update log analysis function 221 extracts and deletes unnecessary nodes in the causal relationship model DB 233 (FIG. 4).

In S2408, the update log analysis function 221 extracts a record in which the update status of the update content management table 2210 (FIG. 22) is "update" and the update pattern is "1" or "5". Based on the article ID of each extracted record, the importance of the information source corresponding to the article ID is incremented by +1 from the acquired information table 1600 (FIG. 16) and the information source management table 1500 (FIG. 15).

In step S2409, if the node ID of the record extracted in step S2403 (the record added to the update content management table 2210) is registered in the interest value management table 900 (FIG. 9), the update log analysis function 221 determines that record. Based on the article ID, the importance of the information source corresponding to the article ID is incremented by +10 from the acquired information table 1600 and the information source management table 1500.

In S2410, the update log analysis function 221 newly adds the node name of the record for which the update status of the update content management table 2210 is “pending” to the search management table 1510 (FIG. 15) as a keyword. (If a record with the same node name is already registered, the importance of the record is incremented by +1). Further, for the record registered in the causal relationship model table 400, the node name of the record is newly added to the search management table 1510 as a keyword (the same node name is already registered in the record of the search management table 1510). In that case, the importance of the record is incremented by 1). Further, the importance of the record of the search management table 1510 having the same node name as the node name of the record newly added/updated to the interest value management table 900 (FIG. 9) (the record updated between the last update time and the current update). Increase by 10.

In S2411, the update log analysis function 221 does not update each record of the causal relationship model table 400 for both the update time 405 and the relationship update time 408 for a certain period (for example, half a year or more), and the node ID of the record. The relationship update time 408 of the record registered in the link destination node 406 has not been updated for a certain period of time (for example, half a year or more), and is in the subject predicate set 1804 of the record having the current update ID in the analysis result management table 1800. If there is no match with the node name 403 and the subject 404 of the record, the record is deleted from the causal relationship model table 400, and the update content management table 2210 stores the content of the record with the update ID of this time. Register with update status "Delete".

FIG. 25 is a diagram showing an example of a causal relationship model update sequence when a user intervenes according to the present embodiment.

In the causal relationship model update sequence when a user intervenes, each function of the causal relationship model update sequence in FIG. 14 is activated by information and instructions from the user (model manager) 111 input through the model update screen 263. Then, data such as news articles is acquired from the information source, data is analyzed such as model update events (nodes) and relationships between those nodes, and update information is generated. Evaluate. Next, together with the effectiveness, the causal relationship model before and after the update and the reaction prediction result using the causal relationship model before and after the update are presented to the user 111 via the model update screen 263, and the user corrects the updated content or causes the causal effect. The execution instruction of the relational model update is issued, and the causal relation model is updated by the model registration/update function 231 in the same procedure as the causal relation model update sequence of FIG.

The model update screen 263 acquires the temporary update model data (the updated causal relationship model) from the temporary update model DB 216, and the interest value of each event (node) from the interest event management DB 217. In addition, the model acquisition function 232 is used to acquire causal relationship model data (causal relationship model before update) from the causal relationship model DB 233 and present it to the user 111.

When the user 111 performs correction (correction of relationship between events (nodes), deletion of event (node), change of event (node) name, etc.) to the presented temporary update model data, the update information correction function 215 Then, the correction content is acquired, and each item of the corresponding record among the records registered in the temporary update model DB 216 is corrected.

The reaction prediction generation function 241 acquires the causal relationship model data before and after the update from the causal relationship model DB 233 and the temporary update model DB 216, and performs the same processing as the processing described in FIG. Of the causal relationship model and the temporary update model), the reaction prediction result is generated by simulation and registered in the reaction prediction result DB 243.

The prediction result evaluation function 214 acquires the reaction prediction result using the generated model before and after the update (existing causal relationship model and temporary update model), and outputs it to the model update screen 263.

When the user 111 corrects the reaction prediction result (the prediction result using the temporary update model) output on the model update screen 263, the reaction prediction adjustment function 242 acquires the correction content and executes the simulation, The relationship between the events (nodes) registered in the temporary update model DB 216 is corrected so that the correction content becomes the prediction result.

The model registration/updating function 231 updates the causal relationship model based on the temporary update model DB 216 modified by the user 111 by the process equivalent to the process described in FIG.

FIG. 26 is a diagram showing an example of the model update screen 263 according to the present embodiment.

The model update screen 263 is presented with the information registered in the acquired information DB 203, the information input from the user, the analysis result registered with the analysis result management DB 205, the causal relationship model DB 233 and the temporary update model DB 216. An interface is provided for presenting a causal relationship model and correction input from the user, presenting the prediction result registered in the reaction prediction result DB 243, and making correction input from the user.

The model update screen 263 includes a button 2601, a data selection unit 2602, a manual input button 2603, an automatic collection button 2604, a data collection result display unit 2605, an analysis execution button 2606, a data analysis result display unit 2607, Temporary update execution button 2608, update result display unit 2609, pre-update causal relationship model display unit 2610, post-update causal relationship model display unit 2611, update effectiveness degree display unit 2612, and unrestore button 2613. A correction reflection button 2614, an update execution button 2615, a prediction result display unit 2616, a pre-update prediction result display unit 2617, a post-update prediction result display unit 2618, and a prediction execution button 2619. ..

The button 2601 is a button for returning to the menu for returning to the menu screen. The data selection unit 2602 is a button for registering data used for model update. The manual input button 2603 is a button for registering the data designated by the data selection unit 2602. The automatic collection button 2604 is a button for automatically collecting data by this system.

When the user (model administrator) 111 presses the data selection unit 2602, a list of data used for model update (a list of files stored in a specific folder in advance) is displayed. Then, when the user (model manager) 111 selects data and presses the manual input button 2603, the selected data is registered in the acquisition information DB 203 as data to be used for model update, and the contents of the acquisition information DB 203 are data. It is displayed on the collection result display unit 2605. When the automatic collection button 2604 is pressed, the data acquisition function 201 automatically collects information by the same processing as in FIG. 17, and the data registered in the acquired information DB 203 is displayed on the data collection result display unit 2605.

When the user 111 presses the analysis execution button 2606, the data analysis function 202 performs analysis by the same processing as in FIG. 19, and the data registered in the analysis result management DB 205 is displayed on the data analysis result display unit 2607. ..

When the user 111 presses the temporary update execution button 2608, the update information generation function 212 creates temporary update model data by the same process as in FIG. 21, and the current contents of the causal relationship model DB 233 show the causal relationship before updating. The contents of the temporary update model DB 216 are displayed on the model display unit 2610 on the updated causal relationship model display unit 2611. Further, the update effectiveness calculated by the same process as in FIG. 23 is displayed on the update effectiveness display unit 2612 by the model update evaluation function 213.

Here, the user 111 corrects the temporary update model data displayed in the updated causal relationship model display unit 2611 (correction of relationship between events (nodes), deletion of event (node), and event (node) name change). When the modification reflection button 2614 is pressed, the temporary update model data registered in the temporary update model DB 216 is corrected by the update information correction function 215, and the corrected content causes the causal relationship after the update. It is displayed on the model display portion 2611. When the restore before modification button 2613 is pressed, the state after the temporary update execution button 2608 is pressed can be restored. When the update execution button 2615 is pressed, the causal relationship model is updated by the model registration/update function 231 based on the temporary update model DB 216 modified by the user 111 by the process similar to the process described in FIG. ..

When the user presses the prediction execution button 2619, the reaction prediction generation function 241 acquires the causal relationship model data before and after the update from the causal relationship model DB 233 and the temporary update model DB 216, which is equivalent to the processing described in FIG. By the processing of (1), the reaction prediction result by the simulation using the models before and after the update (the existing causal relationship model and the temporary update model) is generated. The prediction result using the current causal relationship model DB 233 is displayed on the prediction result display unit 2617 before updating, and the prediction result using the temporary update model DB 216 is displayed on the updated prediction result display unit 2618. Here, when the user 111 corrects the reaction prediction result (the prediction result using the temporary update model) displayed on the updated prediction result display unit 2618, the reaction prediction adjustment function 242 executes the simulation, The relationship between the events (nodes) registered in the temporary update model DB 216 is corrected so that the correction content becomes the prediction result. The corrected content is displayed on the updated causal relationship model display unit 2611.

According to the decision support system 101 described above, the causal relationship model updating unit 103 generates temporary update model data for updating the causal relationship model based on the causal relationship model and the analysis result by the data analysis function 202. It has an update information generation function 202 and a model update evaluation function 213 that evaluates the effectiveness of the update by the temporary update model data, and the causal relationship model management unit 105 determines that the evaluation result by the model update evaluation function 213 is above a certain level. In this case, it has a model registration/update function 231 for updating the causal relationship model with the provisionally updated model data.

According to such a configuration, when updating the model so as to correspond to the actual situation that changes sequentially, the updating of the causal relationship model meaningful to the user is executed, so that the causal relationship model of the decision support system 101 It is possible to prevent quality deterioration.

The model update evaluation function 213 further includes an interest event management DB 217 that stores an event of interest to the user, and the model update evaluation function 213 evaluates the effectiveness of the update based on the temporary update model data and the event stored in the interest event management DB 217. To do.

According to such a configuration, the effectiveness of the update is evaluated based on an event that is of interest to the user, so that it is possible to prevent the update of the causal relationship model that is meaningless to the user. It is possible to prevent the quality deterioration and enlargement of the causal relationship model of the system 101.

In addition, the reaction prediction unit 106 predicts, by simulation, a plurality of events that occur subsequent to a certain event when the certain event occurs, displays the prediction result on the terminal, and updates the causal relationship model based on the causal relationship model. The unit 103 has an interest event registration function 211 that acquires an event of interest to the user, which is input to the terminal by the user, and stores it in the interest event management DB 217.

According to such a configuration, an event of interest to the user, which is input to the prediction result, is registered, and the effectiveness of the update is evaluated based on the registered event. It is possible to prevent the updating, and it is possible to prevent the quality and the enlargement of the causal relationship model of the decision support system 101.

The update information generation function 212 causes the terminal to display the updated causal relationship model based on the provisionally updated model data, and the causal relationship model updating unit 103 acquires a correction to the updated causal relationship model input by the user to the terminal. Then, it has an update information correction function 215 for correcting the provisionally updated model data.

According to such a configuration, since the causal relationship model desired by the user is created, a causal relationship model meaningful to the user can be provided. In addition, addition to a causal relationship model that is meaningless to the user is suppressed, so that the causal relationship model can be prevented from becoming huge.

The model update evaluation function 213 causes the terminal to display the evaluation result of the update effectiveness, and the model registration/update function 231 acquires the update instruction of the causal relationship model input by the user to the terminal, and temporarily updates the model data. To update the causal relationship model.

With such a configuration, it is possible to provide a causal relationship model meaningful to the user. In addition, addition to a causal relationship model that is meaningless to the user is suppressed, so that the causal relationship model can be prevented from becoming huge.

Further, the reaction prediction unit 106 predicts, by simulation, a plurality of events that occur subsequent to a certain event when a certain event occurs, based on the causal relationship model and the updated causal relationship model based on the provisionally updated model data. Then, the reaction prediction generation function 241 for displaying the prediction result on the terminal and the correction for the prediction result based on the updated causal relationship model by the user input to the terminal for displaying the prediction result based on the updated causal relationship model are acquired. And a reaction prediction adjustment function 242 that corrects the provisionally updated model data.

According to such a configuration, since the causal relationship model desired by the user is created, a causal relationship model meaningful to the user can be provided. In addition, addition to a causal relationship model that is meaningless to the user is suppressed, so that the causal relationship model can be prevented from becoming huge.

The update information generation function 212 deletes one data when the same event is included in the update target data table 2000.

With such a configuration, it is possible to prevent the causal relationship model of the decision support system 101 from becoming huge.

In addition, when the event updated by the temporary update model data among the plurality of events included in the updated causal relationship model by the model registration/update function 231 is stored in the interested event management DB 217, the updated event The data acquisition function 201 includes the update evaluation management unit 104 that increases the degree of importance corresponding to the data related to the information source or the search condition used to acquire the event, and the data acquisition function 201 acquires the information about the event based on the importance.

According to this configuration, the importance of the information source or the like from which the event of interest to the user is acquired is increased, so that the information source or the like can be narrowed down. Then, by acquiring information from an information source or the like having a certain degree of importance or higher, it is possible to update the causal relationship model that is meaningful to the user, and it is possible to prevent deterioration of update performance.

In addition, the model registration/update function 231 allows the information source or the search condition used to acquire the updated event for the event updated by the provisionally updated model data among the plurality of events included in the updated causal relationship model. The data acquisition function 201 includes the update evaluation management unit 104 that increases the degree of importance corresponding to the data regarding, and the data acquisition function 201 acquires the information regarding the event based on the degree of importance.

With this configuration, it is possible to narrow down the information sources and the like. Then, by acquiring information from an information source or the like having a certain degree of importance or higher, it is possible to update the causal relationship model that is meaningful to the user, and it is possible to prevent deterioration of update performance.

An update evaluation management unit 104 is provided for deleting, from the causal relationship model, an event of the causal relationship model that has not been updated for a certain period of time and has not been updated for a certain period of time in relation to other events.

With such a configuration, it is possible to prevent the causal relationship model of the decision support system 101 from becoming huge.

As a result of the analysis by the data analysis function 202, for one event, if there is no event that follows the one event, or if the one event does not occur following the remaining other events, update The information generation function 212 excludes the one event from the events that update the causality model. That is, in the update target data table 2000, the update status is set to “pending” for a record that has no link destination node or is not a link destination node of any record.

With such a configuration, it is possible to prevent the size of the updated causal relationship model from increasing.

It should be noted that the present invention is not limited to the above-described embodiments, but includes various modifications. For example, the above-described embodiments have been described in detail in order to explain the present invention in an easy-to-understand manner, and are not necessarily limited to those having all the configurations described.

For example, in the update target data table 2000, the update information generation function 212 sets the update status to “pending” for a record that does not have a link destination node or is not a link destination node of any record, and sets the record concerned. Although the record is excluded from the update target, two records in which the update pattern 2014 has a relationship of “1” and “3” in the temporary update model data table 2010 may be excluded from the update target. Since such a record does not significantly affect the prediction result, it can be excluded from the update target and the size of the causal relationship model after the update can be prevented from becoming huge.

Further, each of the configurations, functions, processing units, processing means, and the like of the above-described embodiments may be realized by hardware by designing a part or all of them with, for example, an integrated circuit. Further, the above-described respective configurations, functions and the like may be realized by software by a processor interpreting and executing a program that realizes each function. Information such as a program, a table, and a file that realizes each function can be placed in a recording device such as a memory, a hard disk, an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, or a DVD.

Further, the control lines and information lines are shown to be necessary for explanation, and not all the control lines and information lines are shown in the product. In reality, it may be considered that almost all the configurations are connected to each other.

101 Decision Support System 102 Data Collection/Analysis Unit 103 Causal Relationship Model Updating Unit 104 Update Evaluation Management Unit 105 Causal Relationship Model Management Unit 106 Reaction Prediction Unit 201 Data Acquisition Function 202 Data Analysis Function 204 Acquisition/Analysis Management DB
211 Interest event registration function 212 Update information generation function 213 Model update evaluation function 215 Update information correction function 217 Interest event management DB
231 Model registration/update function 241 Reaction prediction generation function 242 Reaction prediction adjustment function

Claims (11)

A decision support system comprising a causal relationship model management unit, a data collection/analysis unit, and a causal relationship model updating unit,
The causal relationship model management unit stores a plurality of events that occur in society or nature, and the relationship of occurrence of the events as a causal relationship model,
The data collection/analysis unit is
A data acquisition function for acquiring information on the event,
And a data analysis function for analyzing the acquired information about the event,
The causal relationship model updating unit,
An update information generation function that generates update data for updating the causal relationship model based on the causal relationship model and the analysis result by the data analysis function,
A model update evaluation function for evaluating the effectiveness of the update by the update data,
An interest event storage unit that stores an event that is of interest to the user,
The model update evaluation function evaluates the effectiveness of update based on the update data and the event stored in the interested event storage unit,
The causal relationship model management unit has a model registration/update function for updating the causal relationship model with the update data when the evaluation result by the model update evaluation function is a certain value or more,
Decision support system characterized by. Based on the causal relationship model, a plurality of events occurring subsequent to the certain event when a certain event occurs is predicted by simulation, further comprising a reaction prediction unit for displaying the prediction result on the terminal,
The causal relationship model updating unit has an event-of-interest registration function of acquiring an event of interest to the user input to the prediction result of the terminal by the user and storing the event in the event-of-interest storage unit.
The decision support system according to claim 1. The update information generation function causes the terminal to display the updated causal relationship model based on the update data,
The causal relationship model updating unit has an update information correction function of acquiring a correction to the updated causal relationship model input to the terminal by the user and correcting the update data.
The decision support system according to claim 1. The model update evaluation function causes the terminal to display an evaluation result of update effectiveness,
The model registration/update function updates the causal relationship model with the update data when an instruction to update the causal relationship model input by the user to the terminal is acquired,
The decision support system according to claim 3. Based on the causal relationship model after the causal relationship model and the updated causal relationship model based on the update data, a plurality of events that occur subsequent to the certain event when a certain event occurs are respectively predicted by simulation, and the prediction result is displayed on the terminal. Reaction prediction generation function to be displayed,
A reaction prediction adjustment that acquires a correction to the prediction result based on the updated causal relationship model by the user input to the terminal that displays the prediction result based on the updated causal relationship model, and corrects the updated data. Function,
Further comprising a reaction prediction unit having
The decision support system according to claim 1. The update information generation function deletes one data when the same event is included in the update data,
The decision support system according to any one of claims 1 to 5. The data collection/analysis unit stores a plurality of acquisition source data set in advance for the data acquisition function to acquire information regarding the event and an importance indicating the importance of each acquisition source data. Has an analysis storage unit,
When the event updated by the update data among the plurality of events included in the causal relationship model by the model registration/update function is stored in the interest event storage unit, acquisition of the updated event Further comprising an update evaluation management unit that increases the importance corresponding to the acquisition source data used in
The data acquisition function acquires information about the event based on the importance,
The decision support system according to any one of claims 1 to 6. The data collection/analysis unit stores a plurality of acquisition source data preset for the data acquisition function to acquire information on the event and an importance indicating the importance of each acquisition source data. Has an analysis storage unit,
With the model registration/update function, for the event updated by the update data among the plurality of events included in the causal relationship model, the event corresponding to the acquisition source data used to acquire the updated event It further comprises an update evaluation management unit that increases the degree of importance,
The data acquisition function acquires information about the event based on the importance,
The decision support system according to any one of claims 1 to 6. Regarding the event of the causal relationship model, there is no update for a certain period or more, and further comprising an update evaluation management unit that deletes an event that has not been updated for a certain period or more in relation to other events from the causal relationship model,
The decision support system according to any one of claims 1 to 8. As a result of the analysis by the data analysis function, for one event, if there is no event that occurs after the one event, or if the one event does not occur after the remaining other events, The update information generation function excludes the one event from the event that updates the causality model,
The decision support system according to any one of claims 1 to 9. A decision support method executed by a decision support system comprising a causal relationship model management unit, a data collection/analysis unit, and a causal relationship model updating unit ,
Obtain information about events that occur in society or nature,
Analyze the acquired information about the event,
To update the causal relationship model based on a plurality of the events that occur in society or nature and a causal relationship model configured by the relationship of occurrence of the events and an analysis result of the information. Generate update data,
Evaluating the effectiveness of the update by the update data,
Updating the causal relationship model with the update data when the evaluation result of the effectiveness of the update is more than a certain value,
Decision support method characterized by.

Images