Strategic Intent Analysis Method Based on Open Source Intelligence

Abstract

The present invention relates to a strategic intent analysis method based on open source intelligence, which belongs to the field of deep learning technology. The method includes obtaining open source intelligence text and performing preprocessing; extracting multiple event elements and extracting core words of the event; extracting event dependency subgraphs containing meta-event pairs; obtaining annotation labels corresponding to each meta-event in the meta-event pairs; constructing a preliminary event graph; obtaining an optimized event graph; performing subsequent event prediction on the events in the optimized event graph through a strategic event prediction model, and obtaining the inducement of the events in the optimized event graph. The method provided by the present application mines information from unstructured text, integrates relevant data and resources, and expands open source intelligence text data sets, extracts event elements and event relationships in the open source intelligence text data sets, analyzes the enemy's strategic intent through the event graph, realizes the transformation of traditional intelligence work to informatization and intelligence, and provides auxiliary decision-making capabilities for intelligence analysis work.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a strategic intent analysis method based on open source intelligence. Background Art

[0002] Big data and artificial intelligence technologies are being widely applied across all industries and sectors, becoming a significant force driving the intelligent transformation of society. The in-depth application of these technologies in the military is becoming an irreversible trend. Compared to their civilian applications, military applications place higher demands on the complexity, security, and robustness of these technologies. How to leverage these new technologies to achieve rapid and high-quality development in my country's military intelligence has become a pressing issue.

[0003] With the rapid development of high-performance computing capabilities, particularly with the hardware design features of high-performance, high-concurrency GPUs, deep learning has evolved from an algorithmic concept into a vital driving force for the development of information technology in society. In the military, the realization of intelligent warfare relies on deeper mining, analysis, distribution, and utilization of data. Strategic intent analysis, with its powerful ability to integrate and correlate knowledge, is precisely what military strategic analysis requires.

[0004] In recent years, the development of event graph technology has provided a solution for strategic intent analysis. In existing technologies, event relationship extraction technology is used to extract the correlation between events, thereby providing assistance for the strategic intent analysis of military events. In existing technologies, machine learning methods are used to model the relationship patterns between entities for relationship extraction, and two mainstream supervised learning methods have been formed: 1) Feature-based methods, which extract features at the lexical and semantic levels and use classifiers such as support vector machines for classification; 2) Kernel function-based methods, which determine the relationship type based on the similarity of the kernel function.

[0005] However, existing supervised learning methods have obvious shortcomings. Corpus annotation requires a lot of time and manpower. Although semi-supervised and unsupervised methods effectively avoid large-scale data annotation, they fail to fully consider the deep semantics of the text, resulting in low accuracy of relationship extraction. Summary of the Invention

[0006] The present invention aims to provide a strategic intent analysis method based on open source intelligence to address the deficiencies in the prior art. The technical problems to be solved by the present invention are achieved through the following technical solutions.

[0007] The strategic intent analysis method based on open source intelligence provided by the present invention includes:

[0008] Obtain open source intelligence texts through various means and store them in the form of files on a file server;

[0009] Preprocess the open source intelligence text in the file server to obtain the open source intelligence dataset;

[0010] The event factor extraction model and the event relationship extraction model are trained using open source intelligence datasets to obtain their corresponding model parameters.

[0011] The event element extraction model is used to classify messages in the open source intelligence dataset by event type, and multiple event elements are extracted based on the classification, as well as the core words of the event;

[0012] The neural network identification module of the event-reason relationship extraction model vectorizes the messages in the open source intelligence dataset to obtain the semantic vectors of the messages. The syntactic analysis module obtains the relationship between the semantic vectors to establish an event dependency graph for each event, extracts the meta-event pairs consisting of the meta-events corresponding to each event, and extracts the event dependency subgraph containing the meta-event pairs.

[0013] The adjacency matrix corresponding to the obtained event dependency subgraph and the feature vector representation matrix corresponding to the adjacency matrix are input into the attention representation module of the event relationship extraction model to obtain the category of the relationship between each meta-event corresponding to each event. The related meta-event pairs and the feature vectors corresponding to the meta-event pairs are input into the CRF layer of the event relationship extraction model to obtain the label corresponding to each meta-event in the meta-event pair and output them;

[0014] Construct a preliminary event-related graph that includes the core words of the event, event elements, and meta-event relationships;

[0015] Improve the preliminary event graph through the event reasoning model to obtain an optimized event graph;

[0016] The strategic event prediction model is used to predict subsequent events in the optimized causal map, and the causes of the events in the optimized causal map are obtained based on the meta-event relationships in the causal map.

[0017] In the above scheme, the event types include exercises and training, target movements, military conflicts, displays of force, and reconnaissance patrols.

[0018] In the above scheme, the method of classifying messages in the open source intelligence dataset by event type using the event element extraction model, extracting multiple event elements based on the classification, and extracting core words of the event includes:

[0019] An unsupervised algorithm is used to classify messages in the open source intelligence dataset by event type;

[0020] Extract event sentences containing arguments from messages in the classified open source intelligence dataset;

[0021] Predictively annotate the extracted event sentences and splice the annotated events to achieve chapter-level event extraction;

[0022] Arguments are used to supplement the extracted chapter-level events to obtain multiple event elements, and the core words of the events are extracted.

[0023] In the above solution, the process of improving the preliminary event graph through the event reasoning model to obtain an optimized event graph includes:

[0024] Extract new relations and events and expand them to form triples (L, P, R);

[0025] The triples (L, P, R) in the preliminary event graph and the triples (L, P, R) formed by extracting new relations and events are divided into training sets and test sets according to a certain ratio;

[0026] The training set is input into the QuatE model for training to obtain the vector representation of events containing the previous context L and the following context R, and the vector representation of the relationship containing the predicate P;

[0027] Based on the meta-learning framework, we fine-tune the vector representations of entities and relations, and define a triplet score function based on the vector representations of events and relations.

[0028] Remove the context R of a triple (L, P, R) in the test set to obtain an incomplete triple. Replace the removed object O with any event e in the test set that contains both the context L and the context R to form multiple new triples.

[0029] Obtain the scores corresponding to the new triples through the triple score function and sort them;

[0030] Predict incomplete triples based on the sorting results;

[0031] Repeat the above steps for other triplets (L, P, R) in the test set to obtain the optimal model parameters of the event reasoning model;

[0032] The new inter-event relationship composed of the predicted triples is obtained through the event reasoning model, so as to improve the preliminary event graph and obtain the optimized event graph.

[0033] In the above scheme, the event reasoning model includes an entity relationship representation pre-training module and an entity representation learning module. The entity relationship representation pre-training module is used to train the training set and the test set to obtain the vector representation of the event containing the previous context L and the following context R, and the vector representation of the relationship containing the predicate P respectively; the entity representation learning module is used to re-fine-tune the vector representation of the event and the vector representation of the relationship obtained by the entity relationship representation pre-training module based on the meta-learning framework.

[0034] In the above scheme, the event reasoning model also includes a matching module, which is used to match incomplete triples with new events, predict incomplete triples, and improve the preliminary event map.

[0035] In the above scheme, the strategic event prediction model includes an event representation module and a long-range time series module. The event representation module is used to map events in the optimized event graph to a low-dimensional vector space, and use Transformer to learn the semantic associations within the events to obtain a comprehensive event representation; the long-range time series module is used to model the contextual events in the narrative event chain composed of the comprehensive event representation through the LSTM neural network to obtain an event representation that integrates time series information.

[0036] In the above scheme, the strategic event prediction model also includes a global event evolution module and a gated attention module. The global event evolution module is used to capture the event evolution pattern in the optimized event graph and obtain the event representation of the event evolution pattern through the GNN network; the gated attention module dynamically fuses the event representation obtained by the long-range time series module that integrates the time series information and the event representation obtained by the global event evolution module through the gated attention mechanism, and predicts subsequent events.

[0037] In the above scheme, the multiple ways of obtaining open source intelligence text include: obtaining open source intelligence text from various news websites and government websites through selenium web crawlers, timers, multi-threading, and resource pools.

[0038] In the above scheme, the semantic vectors of the messages obtained by vectorizing the messages in the open source intelligence dataset through the neural network identification module of the relationship extraction model include:

[0039] The neural network identification module of the event-reason relationship extraction model is used to receive messages in the open source intelligence dataset, perform word segmentation on the received messages, and obtain the semantic vector of the message based on the word segmentation results.

[0040] The embodiments of the present invention include the following advantages:

[0041] The embodiment of the present invention provides a strategic intent analysis method based on open source intelligence. By collecting and preprocessing open source intelligence texts, extracting event elements and event relationships in the open source intelligence text data set, establishing a preliminary strategic causal map based on the extracted information, and improving the preliminary causal map to obtain an optimized causal map, strategic intent analysis is performed based on the optimized causal map. The method can mine information from unstructured texts, integrate relevant data and resources, and expand the open source intelligence text data set. By analyzing the enemy's strategic intentions through the causal map, the traditional intelligence work is transformed into information-based and intelligent work, and auxiliary decision-making capabilities are provided for intelligence analysis work. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a step flow chart of an embodiment of a strategic intent analysis method based on open source intelligence of the present invention.

[0043] Figure 2 It is a flowchart of the steps of extracting multiple event elements of the present invention.

[0044] Figure 3 It is a framework diagram of event element extraction of the present invention.

[0045] Figure 4 It is a flow chart of the steps for improving the preliminary event map of the present invention.

[0046] Figure 5 It is a structural diagram of the event reasoning model of the present invention. DETAILED DESCRIPTION

[0047] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0048] like Figure 1 As shown, the present invention provides a strategic intent analysis method based on open source intelligence, including:

[0049] Step S1: Obtain open source intelligence text through various means, and store the open source intelligence text in the form of files in a file server.

[0050] In this embodiment, open source intelligence texts are obtained from various news websites and government websites through selenium web crawlers, timers, multi-threading, resource pools, etc.

[0051] Step S2: Preprocess the open source intelligence text in the file server to obtain the open source intelligence dataset.

[0052] In this embodiment, preprocessing the open source intelligence text in the document server includes preliminarily processing the open source intelligence text into the data form required for subsequent steps.

[0053] Step S3: Train the event element extraction model and the event relationship extraction model using the open source intelligence dataset to obtain their corresponding model parameters.

[0054] Step S4: Classify the messages in the open source intelligence dataset by event type through the event element extraction model, extract multiple event elements based on the classification, and extract the core words of the event.

[0055] In this embodiment, the event types include exercises and training, target movements, military conflicts, displays of force, reconnaissance patrols, and the like.

[0056] like Figure 2 As shown, step S4 includes:

[0057] Step S41: using an unsupervised algorithm to classify messages in the open source intelligence dataset by event type;

[0058] Step S42: extracting event sentences containing arguments from the messages in the classified open source intelligence dataset;

[0059] Step S43: predictively annotate the extracted event sentences, and splice the annotated events to achieve chapter-level event extraction;

[0060] Step S44: using arguments to supplement the extracted chapter-level events to obtain multiple event elements, and extracting the core words of the events.

[0061] like Figure 3 As shown in the figure, the framework diagram of event element extraction shows that the event types including exercises, target movements, military conflicts, shows of force, and reconnaissance patrols in the open source intelligence dataset are identified through unsupervised algorithms. According to the identification results, event sentences containing arguments are extracted from the classified open source intelligence dataset, and the extracted sentences are predicted and labeled. The labeled events are spliced ​​to achieve paragraph-level event extraction.

[0062] Step S5: The open source intelligence message is vectorized by the neural network identification module of the event-reason relationship extraction model to obtain the semantic vector of the message, and the relationship between the semantic vectors is obtained by the syntactic analysis module to establish the event dependency graph of each event, and the meta-event pairs composed of the meta-events corresponding to each event are extracted, and the event dependency subgraph containing the meta-event pairs is extracted.

[0063] In this embodiment, the neural network identification module of the matter-reason relationship extraction model is used to receive messages in the open source intelligence data set, perform word segmentation processing on the received messages, and obtain the semantic vector of the message based on the word segmentation processing result.

[0064] Step S6: Input the adjacency matrix corresponding to the obtained event dependency subgraph and the eigenvector representation matrix corresponding to the adjacency matrix into the attention representation module of the causal relationship extraction model to obtain the category of the relationship between each meta-event corresponding to each event, input the related meta-event pairs and the eigenvectors corresponding to the meta-event pairs into the CRF layer of the causal relationship extraction model to obtain the annotation labels corresponding to each meta-event in the meta-event pairs, and output them.

[0065] In this embodiment, the event-sense relationship extraction model includes a neural network identification module, a syntactic analysis module and an attention representation module. The neural network identification module is used to vectorize the open source intelligence message to obtain the semantic vector of the message; the syntactic analysis module is used to obtain the relationship between the semantic vectors to establish an event dependency graph for each event, and to extract the meta-event pairs composed of the meta-events corresponding to each event, and to extract the event dependency subgraph containing the meta-event pairs; the attention representation module is used to process the adjacency matrix corresponding to the obtained event dependency subgraph and the eigenvector representation matrix corresponding to the adjacency matrix to obtain the category of the relationship between each meta-event corresponding to each event.

[0066] Step S7: Construct a preliminary event graph including the core words of the event, event elements and meta-event relationships.

[0067] Step S8: Improve the preliminary event graph through the event reasoning model to obtain an optimized event graph.

[0068] like Figure 4 As shown, step S8 includes:

[0069] Step S81: extract new relations and events and expand them to form triples (L, P, R);

[0070] Step S82: The triples (L, P, R) in the preliminary event graph and the triples (L, P, R) formed by extracting new relations and events for expansion are divided into a training set and a test set according to a certain ratio;

[0071] Step S83: Input the training set into the QuatE model for training to obtain vector representations of events containing the preceding context L and the following context R, and vector representations of relations containing the predicate P;

[0072] Step S84: re-fine-tune the vector representation of entities and the vector representation of relationships based on the meta-learning framework, and define a triplet score function based on the vector representation of events and the vector representation of relationships;

[0073] Step S85: Remove the context R of a triple (L, P, R) in the test set to obtain an incomplete triple, and replace the removed object O with any event e in the test set that contains the context L and the context R to form multiple new triples;

[0074] Step S86: Obtain scores corresponding to the new triples using the triple score function and sort them;

[0075] Step S87: predicting incomplete triples based on the sorting results;

[0076] Step S88: Repeat the above steps for other triples (L, P, R) in the test set to obtain the optimal model parameters of the event reasoning model;

[0077] Step S89: Obtain the new inter-event relationship composed of the predicted triples through the event reasoning model, thereby improving the preliminary event graph and obtaining an optimized event graph.

[0078] In this embodiment, the scoring function is f(l,p,e)=||l+pe||, where l is the vector representation of L above, p is the vector representation of the relationship, and e is the vector representation of the event. The smaller the score, the greater the possibility that the new triple is established, and vice versa, the less likely the new triple is established.

[0079] In this embodiment, the first N events e in the ranking are used to replace the removed following R respectively, where N can be selected as needed.

[0080] like Figure 5 As shown, the event reasoning model includes an entity relationship representation pre-training module, an entity representation learning module and a matching module. The entity relationship representation pre-training module is used to train the training set and the test set to obtain the vector representation of the event containing the upper context L and the lower context R, and the vector representation of the relationship containing the predicate P respectively; the entity representation learning module is used to re-fine-tune the vector representation of the event and the vector representation of the relationship obtained by the entity relationship representation pre-training module based on the meta-learning framework; the matching module is used to match incomplete triples with new events, predict incomplete triples, and improve the preliminary event map.

[0081] In this embodiment, by predicting incomplete triples through the above process, multiple events similar to the missing contexts in the incomplete triples can be obtained, thereby completing and improving the preliminary event graph.

[0082] Step S9: Use the strategic event prediction model to predict subsequent events for the events in the optimized causal map, and obtain the causes of the events in the optimized causal map based on the meta-event relationships in the causal map.

[0083] In this embodiment, the strategic event prediction model includes an event representation module, a long-range temporal sequence module, a global event evolution module and a gated attention module. The event representation module is used to map events in the optimized causal graph to a low-dimensional vector space, and use Transformer to learn the semantic associations within the events to obtain a comprehensive event representation; the long-range temporal sequence module is used to model the contextual events in the narrative event chain composed of the comprehensive event representation through the LSTM neural network to obtain an event representation that integrates temporal information; the global event evolution module is used to capture the event evolution pattern in the optimized causal graph, and obtain the event representation of the event evolution pattern through the GNN network; the gated attention module dynamically fuses the event representation integrated with temporal information obtained by the long-range temporal sequence module and the event representation obtained by the global event evolution module through the gated attention mechanism, and predicts subsequent events.

[0084] It should be noted that the above detailed description is exemplary and is intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the art to which this application belongs.

[0085] It should be noted that the terms used herein are intended only to describe specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0086] It should be noted that the terms "first," "second," and the like in the specification and claims of this application and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, such that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein.

[0087] In addition, the terms "comprises" and "comprising" and any variations thereof are intended to cover a non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, product, or apparatus.

[0088] For ease of description, spatially relative terms such as "above", "above", "on the upper surface of", "above", etc. may be used herein to describe the spatial positional relationship of a device or feature to other devices or features as shown in the figures. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figures. For example, if the device in the drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be subsequently positioned as "below other devices or structures" or "below other devices or structures". Thus, the exemplary term "above" can include both "above" and "below". The device can also be positioned in other different ways, such as rotated 90 degrees or in other orientations, and the spatially relative descriptions used herein are interpreted accordingly.

[0089] In the above detailed description, reference is made to the accompanying drawings, which form a part hereof. In the drawings, similar symbols typically identify similar components, unless the context dictates otherwise. The illustrated embodiments described in the detailed description, drawings, and claims are not meant to be limiting. Other embodiments may be used, and other changes may be made, without departing from the spirit or scope of the subject matter presented herein.

[0090] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A strategic intent analysis method based on open source intelligence, characterized by: The method comprises: Obtain open source intelligence texts through various means and store them in the form of files on a file server; Preprocess the open source intelligence text in the file server to obtain the open source intelligence dataset; The event factor extraction model and the event relationship extraction model are trained using open source intelligence datasets to obtain their corresponding model parameters. The event element extraction model is used to classify messages in the open source intelligence dataset by event type, and multiple event elements are extracted based on the classification, as well as the core words of the event; The neural network identification module of the event-reason relationship extraction model vectorizes the messages in the open source intelligence dataset to obtain the semantic vectors of the messages. The syntactic analysis module then extracts the relationships between the semantic vectors to build an event dependency graph for each event. Meta-event pairs consisting of meta-events corresponding to each event are then extracted, and event dependency subgraphs containing these meta-event pairs are then extracted. The adjacency matrix corresponding to the obtained event dependency subgraph and the feature vector representation matrix corresponding to the adjacency matrix are input into the attention representation module of the event relationship extraction model to obtain the category of the relationship between each meta-event corresponding to each event. The related meta-event pairs and the feature vectors corresponding to the meta-event pairs are input into the CRF layer of the event relationship extraction model to obtain the label corresponding to each meta-event in the meta-event pair and output them; Construct a preliminary event-related graph that includes the core words of the event, event elements, and meta-event relationships; Improve the preliminary event graph through the event reasoning model to obtain an optimized event graph; The strategic event prediction model is used to predict subsequent events in the optimized causal map, and the causes of the events in the optimized causal map are obtained based on the meta-event relationships in the causal map.

2. The strategic intent analysis method based on open source intelligence according to claim 1 is characterized in that: The types of events described include exercises and training, target movements, military conflicts, shows of force, and reconnaissance patrols.

3. The method for analyzing strategic intent based on open source intelligence according to claim 2, characterized in that: The method of classifying messages in the open source intelligence dataset by event type through the event element extraction model, extracting multiple event elements based on the classification, and extracting core words of the event includes: An unsupervised algorithm is used to classify messages in the open source intelligence dataset by event type; Extract event sentences containing arguments from messages in the classified open source intelligence dataset; Predictively annotate the extracted event sentences and splice the annotated events to achieve chapter-level event extraction; Arguments are used to supplement the extracted chapter-level events to obtain multiple event elements, and the core words of the events are extracted.

4. The method for strategic intent analysis based on open source intelligence according to claim 1, characterized in that: The method of improving the preliminary event graph by the event reasoning model to obtain an optimized event graph includes: Extract new relations and events to expand and form triples (L, P, R); The triples (L, P, R) in the preliminary event graph and the triples (L, P, R) formed by extracting new relations and events are divided into training sets and test sets according to a certain ratio; The training set is input into the QuatE model for training to obtain the vector representation of events containing the previous context L and the following context R, and the vector representation of the relationship containing the predicate P; Based on the meta-learning framework, we fine-tune the vector representations of entities and relations, and define a triplet score function based on the vector representations of events and relations. Remove the context R of a triple (L, P, R) in the test set to obtain an incomplete triple. Replace the removed object O with any event e in the test set that contains both the context L and the context R to form multiple new triples. Obtain the scores corresponding to the new triples through the triple score function and sort them; Predict incomplete triples based on the sorting results; Repeat the above steps for other triplets (L, P, R) in the test set to obtain the optimal model parameters of the event reasoning model; The new inter-event relationship composed of the predicted triples is obtained through the event reasoning model, so as to improve the preliminary event graph and obtain the optimized event graph.

5. The method for analyzing strategic intent based on open source intelligence according to claim 4, characterized in that: The event reasoning model includes an entity relationship representation pre-training module and an entity representation learning module. The entity relationship representation pre-training module is used to train the training set and the test set to obtain the vector representation of the event containing the previous context L and the following context R, and the vector representation of the relationship containing the predicate P; the entity representation learning module is used to re-fine-tune the vector representation of the event and the vector representation of the relationship obtained by the entity relationship representation pre-training module based on the meta-learning framework.

6. The method for analyzing strategic intent based on open source intelligence according to claim 5, characterized in that: The event reasoning model also includes a matching module, which is used to match incomplete triples with new events, predict incomplete triples, and improve the preliminary event graph.

7. The method for strategic intent analysis based on open source intelligence according to claim 1, characterized in that: The strategic event prediction model includes an event representation module and a long-range time series module. The event representation module is used to map events in the optimized event graph to a low-dimensional vector space, and use Transformer to learn the semantic associations within the events to obtain a comprehensive event representation; the long-range time series module is used to model the contextual events in the narrative event chain composed of the comprehensive event representation through the LSTM neural network to obtain an event representation that integrates time series information.

8. The method for strategic intent analysis based on open source intelligence according to claim 7, characterized in that: The strategic event prediction model also includes a global event evolution module and a gated attention module. The global event evolution module is used to capture the event evolution pattern in the optimized event graph and obtain the event representation of the event evolution pattern through the GNN network; the gated attention module dynamically fuses the event representation obtained by the long-range timing module that integrates timing information and the event representation obtained by the global event evolution module through the gated attention mechanism, and predicts subsequent events.

9. The method for strategic intent analysis based on open source intelligence according to claim 1, characterized in that: The multiple methods of obtaining open source intelligence texts include: obtaining open source intelligence texts from various news websites and government websites through selenium web crawlers, timers, multi-threading, and resource pools.

10. The method for strategic intent analysis based on open source intelligence according to claim 1, characterized in that: The semantic vectors of the messages obtained by vectorizing the messages in the open source intelligence dataset through the neural network identification module of the event-reason relationship extraction model include: The neural network identification module of the event-reason relationship extraction model is used to receive messages in the open source intelligence dataset, perform word segmentation on the received messages, and obtain the semantic vector of the message based on the word segmentation results.

Images