A situation context analysis method and system for combat event graph
By constructing a combat event map and identifying event relationships, the problem of causal dynamic evolution analysis of battlefield events has been solved, and an in-depth understanding and efficient analysis of the development context of battlefield situations has been achieved.
Patent Information
- Application Number
- CN202410238712.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-04
- Publication Date
- 2025-05-13
- Estimated Expiration
- 2044-03-04
AI Technical Summary
It is difficult for existing technology to effectively analyze the causal dynamic evolution process of battlefield events, making it difficult to study and understand the development context of battlefield situations.
By designing the combat event mode, conducting event extraction design and relationship identification, building a combat event map, identifying subordinates, co-references, timing and causal relationships, and then analyzing the situation context.
It realizes the causal logic mining of battlefield events and dynamic analysis of situation contexts, improves the efficiency of officers and fighters in obtaining event information, and provides auxiliary decision-making capabilities.
Smart Images
Figure CN119248903B_ABST
Abstract
Description
Technical Field
[0001] The present invention mainly relates to the technical field of event graphs, and specifically to a situation context analysis method and system for a combat event graph. Background Art
[0002] At present, artificial intelligence technology plays a very important role in battlefield situation assessment. With the continuous development of high-tech, we have entered the era of information-based big data. Knowledge discovery and reasoning technology are widely used in military intelligence data analysis, command automation and decision support, weapon engagement effect analysis, geographic data analysis and other fields, enabling us to extract hidden relationships and knowledge from massive data. These technologies help us understand the enemy and battlefield situation from multiple levels such as campaigns and tactics, and grasp the enemy's combat characteristics, so that we can infer its combat plans and future actions based on the enemy's historical information and current status. Among the many types of knowledge, events are also a very important type of knowledge, playing an important role in the fields of machine learning and artificial intelligence. Event graphs can achieve a formal description of tactical behavior knowledge to a certain extent and deepen the understanding of events.
[0003] For some armed military actions, if there is no event map, only the content related to this single event can be obtained. However, through the event map, some subsequent events can also be obtained. Based on the understanding of the event level, the content that originally focused on a single event can be upgraded to a presentation of an event context that can link the context, cause and effect of the event, thereby greatly improving the efficiency of commanders and fighters in obtaining event information. By describing the dynamic changes of the battlefield situation, tracking the progress of the situation, and quickly and accurately obtaining core key information, commanders and fighters can be provided with auxiliary decision-making capabilities. However, there is currently a lack of analysis of the dynamic evolution process of battlefield event causality, making it difficult to study the development context of the battlefield situation. Summary of the invention
[0004] The technical problem to be solved by the present invention is: in response to the technical problems existing in the prior art, the present invention provides a situation context analysis method and system for combat event maps which are simple and quick to construct and improve the efficiency of obtaining event information.
[0005] In order to solve the above technical problems, the technical solution proposed by the present invention is:
[0006] A method for constructing a combat event graph comprises the following steps:
[0007] 1) Design the combat event model, determine the combat event type, trigger words and parameter roles;
[0008] 2) Designing combat event extraction based on the designed combat event model; the combat event extraction design includes trigger classifier design and parameter classifier design; the trigger classifier is used to detect and identify combat event trigger words and assign predefined types to them, and the parameter classifier is used to identify combat event parameter roles;
[0009] 3) Identify the relationship between combat events based on pattern matching methods; the combat event relationships include subordinate relationships, co-reference relationships, temporal relationships, and causal relationships;
[0010] 4) Construct a combat event map based on combat event patterns, combat event extraction design and combat event relationships.
[0011] Preferably, in step 3), in the pattern matching-based method, a pattern is first designed, and then the input text is pattern matched to find the explicit relationship in the corresponding input text; on the one hand, causal patterns are defined using vocabulary, grammar, part of speech and syntactic features to identify causal relationships in the text; on the other hand, keyword features are used to further identify causal relationships.
[0012] The present invention also discloses a situation context analysis method of a combat event map, comprising the steps of:
[0013] S1. constructing a combat event map for data from different sources based on the combat event map construction method described above;
[0014] S2. Perform word segmentation and named entity recognition on the input combat events to obtain named entity recognition results;
[0015] S3, using the sentence-transformer model to calculate the similarity between the named entity recognition results and the entities in the preset prior knowledge base, and matching the prior knowledge related to the combat events;
[0016] S4, calculating the similarity between the input combat event and all events in the event graph, and matching all event nodes and relationship information related to the input combat event;
[0017] S5. Analyze the situation context of the combat event map based on prior knowledge related to the combat event, all related event nodes and relationship information to obtain analysis results.
[0018] Preferably, in step S2, the Jieba word segmentation method is used to perform word segmentation and named entity recognition on the input event; wherein a Jieba word segmentation dictionary is customized; and according to the defined combat event mode, the parameter roles in the combat event mode are stored in the customized Jieba word segmentation dictionary.
[0019] Preferably, the specific process of step S3 is:
[0020] S31, inputting the named entity recognition result into the sentence-transformer model, calculating the similarity between the named entity recognition result and the entity information in the preset prior knowledge base, and matching the prior entity node with the highest similarity;
[0021] S32. Match relevant prior knowledge in the prior knowledge base according to the prior entity node.
[0022] Preferably, the specific process of calculating event similarity in step S4 is:
[0023] Different events are input into two BERT models to generate sentence representation vectors for representing the input sentences; the two BERT models have the same neural network structure and parameters and are regarded as different instances of the same BERT model; the representation vector is a representation of the deep semantic information learned by the BERT model;
[0024] The average pooling strategy is adopted to average all the sentence representation vectors in the event obtained by the BERT model to obtain the mean vector, and then the mean vector is used as the sentence representation vector of the whole sentence. The sentence representation vector is then cosine-phase calculated to obtain the semantic similarity.
[0025] Preferably, the BERT model takes mean square error loss as the optimization objective function.
[0026] Preferably, in step S5, the analysis results are presented in the form of a graph.
[0027] The present invention also discloses a combat event graph construction system, comprising:
[0028] The first program module is used to design the combat event model and determine the combat event type, trigger words and parameter roles;
[0029] The second program module is used to perform combat event extraction design based on the designed combat event model; the combat event extraction design includes trigger classifier design and parameter classifier design; the trigger classifier is used to detect and identify combat event trigger words and assign predefined types to them, and the parameter classifier is used to identify combat event parameter roles;
[0030] The third program module is used to identify the relationship between combat events based on the pattern matching method; the combat event relationship includes a subordinate relationship, a co-referential relationship, a temporal relationship and a causal relationship;
[0031] The fourth program module is used to construct a combat event map based on combat event patterns, combat event extraction design and combat event relationships.
[0032] The present invention further discloses a situation context analysis system for a combat event map, comprising:
[0033] The first module is used to construct a combat event map for data from different sources based on the combat event map construction system as described above;
[0034] The second module is used to perform word segmentation and named entity recognition on the input combat events to obtain the named entity recognition results;
[0035] The third module is used to calculate the similarity between the named entity recognition results and the entities in the preset prior knowledge base through the sentence-transformer model, and match the prior knowledge related to the combat events;
[0036] The fourth module is used to calculate the similarity between the input combat event and all events in the event graph, and match all event nodes and relationship information related to the input combat event;
[0037] The fifth module is used to analyze the situation context of the combat event map based on the prior knowledge related to the combat event, all related event nodes and relationship information.
[0038] Compared with the prior art, the advantages of the present invention are:
[0039] The combat event map construction method of the present invention mines the event causal logic by designing combat event patterns, combat event extraction design and combat event relationships, and constructs a combat event map, which is convenient for subsequent deduction of the situation development context of the target combat event; the above-mentioned map construction method is simple, fast and has high accuracy.
[0040] The situation context analysis method of the combat event map of the present invention mines the causal logic of data from different sources based on the combat event map construction method described above, constructs a combat event map, and upgrades the content that originally only focuses on a single event to a presentation of an event context that can link the context, cause and effect of the event, thereby greatly improving the efficiency of commanders and fighters in obtaining event information. By describing the dynamic changes of the battlefield situation, tracking the progress of the situation, and quickly and accurately obtaining core key information, commanders and fighters are provided with auxiliary decision-making capabilities. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 It is a schematic diagram of the event extraction process of the present invention.
[0042] Figure 2 Schematic diagram of the event similarity calculation model of the present invention.
[0043] Figure 3 It is an architecture diagram of the situation context analysis system of the present invention.
[0044] Figure 4 It is a flow chart of the combat event map construction method in an embodiment of the present invention.
[0045] Figure 5 The figure is a flow chart of the situation context analysis method in an embodiment of the present invention. DETAILED DESCRIPTION
[0046] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.
[0047] like Figure 4 As shown, the combat event graph construction method provided by the embodiment of the present invention is applied to the military field, and includes the steps of:
[0048] 1) Design the combat event model, determine the combat event type, trigger words and parameter roles;
[0049] 2) Designing combat event extraction based on the designed combat event model; the combat event extraction design includes trigger classifier design and parameter classifier design; the trigger classifier is used to detect and identify combat event trigger words and assign predefined types to them, and the parameter classifier is used to identify combat event parameter roles;
[0050] 3) Identify the relationship between combat events based on pattern matching methods; the combat event relationships include subordinate relationships, co-reference relationships, temporal relationships, and causal relationships;
[0051] 4) Construct a combat event map based on combat event patterns, combat event extraction design and combat event relationships.
[0052] The combat event map construction method of the present invention mines the event causal logic by designing combat event patterns, combat event extraction design and combat event relationships, and constructs a combat event map, which is convenient for subsequent deduction of the situation development context of the target combat event; the above-mentioned map construction method is simple, fast and has high accuracy.
[0053] In a specific embodiment, in step 1), for data from different sources but belonging to the same specific field, a combat event model is designed. The event map is defined as:
[0054]
[0055] in, s Represents an event or entity, Indicates an event (i.e. a combat event), such as "the infantry squad launched a fire attack"; Indicates entities, such as "weapons and equipment, combat units", etc. po As a predicate, it is the trigger word of the event, such as "composition, attack, counterattack", etc.; o As an object, it indicates the object associated with the event, such as "fire strike action, cover action, retreat action", etc.; Indicates relationship, , , Respectively represent the relationship between events, events and entities, and entities and entities, thereby connecting events and entities;
[0056] Then, according to the combat mission, combat events are defined into five types: attack, defense, movement, reconnaissance, and composition. The trigger words and roles of different event types are defined. The details are shown in Table 1.
[0057]
[0058] In a specific embodiment, if Figure 1 As shown, in step 2), after the text information is input, it is passed to the feature learner (including local feature learner and global feature learner) to obtain local and global features. Event extraction is designed based on the designed event pattern, mainly including trigger classifier design and parameter classifier design. The trigger classifier detects and identifies the event trigger words and assigns them appropriate predefined types, and the parameter classifier identifies the event roles. Taking an armed military action as an example, the goal of event extraction is: first extract the event, identify the event type of the event as "conflict", and determine the attribute values of the event according to the role of the "conflict" type, including event type, name, type, etc.
[0059] In a specific embodiment, there are multiple relationships between events, including subordinate relationships, coreference relationships, temporal relationships, and causal relationships. In step 3), a pattern matching-based method is used to identify explicit relationships. In the pattern matching-based method, a pattern is first designed, and then the input text is pattern matched to find the explicit relationship in the text. On the one hand, causal patterns are defined using features such as vocabulary, grammar, part of speech, and syntax to identify causal relationships in the text. For example, pattern one is defined as "because of X, so Y"; pattern two is defined as "X causes Y", etc., to describe direct causal relationships; pattern three is defined as "X is an influencing factor of Y" to describe the diversity of causal relationships. On the other hand, keyword features are used to further identify causal relationships. Considering that the causal relationships in some texts are not presented in the form of patterns, some keywords, such as "because of", "therefore", "so", etc., are designed as signs of causal relationships, and these keyword features are used to define rules to extract causal relationships.
[0060] After obtaining the combat event pattern, combat event extraction design and combat event relationship, a combat event graph is constructed and stored in the graph database Neo4j.
[0061] The embodiment of the present invention also provides a combat event graph construction system, including:
[0062] The first program module is used to design the combat event model and determine the combat event type, trigger words and parameter roles;
[0063] The second program module is used to perform combat event extraction design based on the designed combat event model; the combat event extraction design includes trigger classifier design and parameter classifier design; the trigger classifier is used to detect and identify combat event trigger words and assign predefined types to them, and the parameter classifier is used to identify combat event parameter roles;
[0064] The third program module is used to identify the relationship between combat events based on the pattern matching method; the combat event relationship includes a subordinate relationship, a co-referential relationship, a temporal relationship and a causal relationship;
[0065] The fourth program module is used to construct a combat event map based on combat event patterns, combat event extraction design and combat event relationships.
[0066] The combat event map construction system of the present invention corresponds to the above-mentioned combat event map construction method, and also has the advantages described in the above-mentioned combat event map construction method.
[0067] like Figure 5 As shown, the embodiment of the present invention also provides a situation context analysis method of a combat event map, comprising the steps of:
[0068] S1. constructing a combat event map for data from different sources based on the combat event map construction method described above;
[0069] S2. Perform word segmentation and named entity recognition on the input combat events to obtain named entity recognition results;
[0070] S3, using the sentence-transformer model to calculate the similarity between the named entity recognition results and the entities in the preset prior knowledge base, and matching the prior knowledge related to the combat events;
[0071] S4, calculating the similarity between the input combat event and all events in the event graph, and matching all event nodes and relationship information related to the input combat event;
[0072] S5. Analyze the situation context of the combat event map based on prior knowledge related to the combat event, all related event nodes and relationship information to obtain analysis results.
[0073] The situation context analysis method of the combat event map of the present invention mines the causal logic of data from different sources based on the combat event map construction method described above, constructs a combat event map, and upgrades the content that originally only focuses on a single event to a presentation of an event context that can link the context, cause and effect of the event, thereby greatly improving the efficiency of commanders and fighters in obtaining event information. By describing the dynamic changes of the battlefield situation, tracking the progress of the situation, and quickly and accurately obtaining core key information, commanders and fighters are provided with auxiliary decision-making capabilities.
[0074] In a specific embodiment, in step S1, data from different sources are used to construct a knowledge graph for the military field based on the ontology model layer and stored in the graph database Neo4j, including event pattern design function, event extraction function and relationship extraction function.
[0075] In a specific embodiment, in step S2, the Jieba word segmentation method is used to segment the input event and recognize named entities. In addition, a Jieba word segmentation dictionary is customized; according to the defined event pattern, the parameter roles in the event pattern, mainly the subject (event or entity) and the object (the object associated with the event), are stored in the customized dictionary to improve the accuracy of word segmentation recognition.
[0076] In a specific embodiment, in step S3, taking into account the problems of multiple meanings and multiple words having the same meaning in the expression of entity information, the named entity recognition result is input into the sentence-transformer model, and the similarity between the named entity recognition result and the entity information in the prior knowledge base is calculated to match the prior entity node with the highest similarity; finally, the relevant prior information is matched in the prior knowledge base according to the prior entity node to obtain the prior entity subgraph.
[0077] In a specific embodiment, in step S4, all nodes and relationship information related to the event are obtained by inputting the event. Specifically, the function of inputting events in the search box is implemented using Typeahead. When the user enters in the text box, an input event is triggered. JS captures this event in the browser and constructs an AJAX asynchronous HTTP query request. The server receives the request containing the keyword entered by the user, and queries the server through this keyword to find relevant data as much as possible and send it to the browser. The browser obtains the response data from the server, and dynamically renders a search result list on the browser and displays it. After the event is entered in the search box, the event is vectorized, the event similarity is calculated based on all event nodes in the event graph, and the event with the highest similarity to the input event is matched; finally, a query statement is generated to retrieve all information related to the event in the graph database.
[0078] The specific method for calculating event similarity is as follows: perform word segmentation on all situation events to obtain raw corpus data, use the sentence-transformer model to process all raw data, input different events into two BERT models, these two BERT models share parameters and can also be understood as the same BERT model, obtain the sentence representation vector of each sentence (these two BERT models have the same neural network structure and parameters, so they can be regarded as different instances of the same BERT model, and generate sentence representation vectors for representing the input sentences, where the representation vector is a representation of the deep semantic information learned by the BERT model); adopt the average pooling strategy, calculate the average of all word vectors in the event obtained by the BERT model, and finally use the mean vector as the sentence vector of the whole sentence to calculate the cosine phase of the sentence representation vector obtained to obtain the semantic similarity:
[0079]
[0080] in, , Represents the event vector. The model uses mean squared error (MAE) loss as the optimization objective function.
[0081] In addition, you can select one or more of "starting event, target event, relationship type" (such as "starting event", "starting event, relationship", "relationship, target event", "target event", "starting event, relationship, target event", etc.) to obtain the corresponding event relationship information. The query results are returned in the form of a graph. If there is no query result in the end, a prompt message is returned.
[0082] In the above overall analysis architecture, the backend includes: Neo4j, a graph database used to store network nodes and the relationships between nodes; Flask, a Python-based Web microframework; py2neo, a Python API package for neo4j; and Echarts, a frontend that displays nodes and relationships, for analysis and production of visualization graphics. Figure 3 shown.
[0083] The present invention also discloses a situation context analysis system for a combat event map, comprising:
[0084] The first module is used to construct a combat event map for data from different sources based on the combat event map construction system as described above;
[0085] The second module is used to perform word segmentation and named entity recognition on the input combat events to obtain the named entity recognition results;
[0086] The third module is used to calculate the similarity between the named entity recognition results and the entities in the preset prior knowledge base through the sentence-transformer model, and match the prior knowledge related to the combat events;
[0087] The fourth module is used to calculate the similarity between the input combat event and all events in the event graph, and match all event nodes and relationship information related to the input combat event;
[0088] The fifth module is used to analyze the situation context of the combat event map based on the prior knowledge related to the combat event, all related event nodes and relationship information.
[0089] The situation context analysis system of the combat event map of the present invention corresponds to the above-mentioned situation context analysis method and also has the advantages described in the above-mentioned analysis method.
[0090] The above are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All technical solutions under the concept of the present invention belong to the protection scope of the present invention. It should be pointed out that for ordinary technicians in this technical field, some improvements and modifications without departing from the principle of the present invention should be regarded as the protection scope of the present invention.
Claims
1. A situation context analysis method for a combat event map, characterized in that: Includes steps: S1. Based on the combat event graph construction method, a combat event graph is constructed using data from different sources; S2. Perform word segmentation and named entity recognition on the input combat events to obtain named entity recognition results; S3, using the sentence-transformer model to calculate the similarity between the named entity recognition results and the entities in the preset prior knowledge base, and matching the prior knowledge related to the combat events; S4, calculating the similarity between the input combat event and all events in the event graph, and matching all event nodes and relationship information related to the input combat event; S5. Analyze the situation context of the combat event map based on prior knowledge related to the combat event, all related event nodes and relationship information, and obtain analysis results; In step S1, the method for constructing a combat event map includes the following steps: 1) Design the combat event model, determine the combat event type, trigger words and parameter roles; 2) Design combat event extraction based on the designed combat event model; The combat event extraction design includes trigger classifier design and parameter classifier design; The trigger classifier is used to detect and identify combat event trigger words and assign predefined types to them, and the parameter classifier is used to identify combat event parameter roles; 3) Identify the relationship between combat events based on pattern matching methods; the combat event relationships include subordinate relationships, co-reference relationships, temporal relationships, and causal relationships; 4) Construct a combat event map based on combat event patterns, combat event extraction design, and combat event relationships; The specific process of step S3 is: S31, inputting the named entity recognition result into the sentence-transformer model, calculating the similarity between the named entity recognition result and the entity information in the preset prior knowledge base, and matching the prior entity node with the highest similarity; S32, matching relevant prior knowledge in the prior knowledge base according to the prior entity node; The specific process of calculating event similarity in step S4 is: Different events are input into two BERT models to generate sentence representation vectors for representing the input sentences; these two BERT models have the same neural network structure and parameters and are regarded as different instances of the same BERT model; The representation vector is a representation of the deep semantic information learned by the BERT model; The average pooling strategy is adopted to average all the sentence representation vectors in the event obtained by the BERT model to obtain the mean vector, and then the mean vector is used as the sentence representation vector of the whole sentence. The sentence representation vector is then cosine phase calculated to obtain the semantic similarity, which is specifically: in, , Represents an event vector.
2. The situation context analysis method of the combat event map according to claim 1 is characterized in that: In step S2, the Jieba word segmentation method is used to segment the input event and perform named entity recognition; wherein a custom Jieba word segmentation dictionary is provided; and according to the defined combat event mode, the parameter roles in the combat event mode are stored in the custom Jieba word segmentation dictionary.
3. The situation context analysis method of the combat event map according to claim 2 is characterized in that: The BERT model uses mean square error loss as the optimization objective function.
4. The situation context analysis method of the combat event map according to any one of claims 1 to 3, characterized in that: In step S5, the analysis results are displayed in the form of a graph.
5. The situation context analysis method of the combat event map according to claim 1 is characterized in that: In step 3), in the pattern matching-based method, firstly, a pattern is designed, and then the input text is pattern matched to find the explicit relationship in the corresponding input text; on the one hand, the causal pattern is defined using vocabulary, grammar, part of speech and syntactic features to identify the causal relationship in the text; on the other hand, the keyword features are used to further identify the causal relationship.
6. A situation context analysis system for a combat event map, used to execute the steps of the situation context analysis method for a combat event map as claimed in any one of claims 1 to 5, characterized in that: include: The first module is used to construct a combat event map based on the combat event map construction system to construct a combat event map based on data from different sources; The second module is used to perform word segmentation and named entity recognition on the input combat events to obtain the named entity recognition results; The third module is used to calculate the similarity between the named entity recognition results and the entities in the preset prior knowledge base through the sentence-transformer model, and match the prior knowledge related to the combat events; The fourth module is used to calculate the similarity between the input combat event and all events in the event graph, and match all event nodes and relationship information related to the input combat event; The fifth module is used to analyze the situation context of the combat event map based on the prior knowledge related to the combat event, all related event nodes and relationship information; The combat event graph construction system includes: The first program module is used to design the combat event model and determine the combat event type, trigger words and parameter roles; The second program module is used to perform combat event extraction design based on the designed combat event model; the combat event extraction design includes trigger classifier design and parameter classifier design; the trigger classifier is used to detect and identify combat event trigger words and assign predefined types to them, and the parameter classifier is used to identify combat event parameter roles; The third program module is used to identify the relationship between combat events based on the pattern matching method; the combat event relationship includes a subordinate relationship, a co-referential relationship, a temporal relationship and a causal relationship; The fourth program module is used to construct a combat event map based on combat event patterns, combat event extraction design and combat event relationships.