A method and system for analyzing case-based information of coal mine text information

Abstract

The application provides a case text information-based accident logic graph analysis method and system. The method comprises the following steps: based on a template extraction algorithm, performing scene event extraction on a pre-acquired accident report sample to construct an initial accident logic graph K1 of a mine outburst accident; based on a reinforcement learning algorithm, optimizing the accident logic relationship in the initial accident logic graph K1 to obtain an application accident logic graph of the mine outburst accident; and based on a preset layout algorithm, performing layout display on the graph data of the application accident logic graph to generate an accident main graph solution of the mine outburst accident. In this way, the accident graph solution is automatically generated from the accident report sample, the automatically generated accident graph solution can help professional personnel quickly understand the accident causes, improve the reading speed of the accident report, reduce the analysis threshold of non-professional personnel, and improve the efficiency of accident guidance analysis.

Description

Technical Field

[0001] This application relates to the field of coal mine safety technology, and in particular to a method and system for analyzing the reasoning graph based on textual information from coal mine cases. Background Technology

[0002] Coal plays an irreplaceable role in economic and social development and occupies an important position in the energy consumption system, but mine accidents during coal mining have caused huge casualties and losses.

[0003] Studying past accident cases and analyzing accidents using accident causation theories are fundamental to preventing mine outbursts. Traditional case causation analysis involves extensive manual reading, relying on past knowledge and experience to identify causal information in descriptive texts. This process is tedious, time-consuming, and the results may be affected by the reader's psychological and physiological fatigue, as well as their knowledge and experience, often leading to inconsistent judgment standards and consequently, errors in the work.

[0004] Therefore, there is an urgent need to provide a technical solution that addresses the shortcomings of the existing technology. Summary of the Invention

[0005] The purpose of this application is to provide a reasoning graph analysis method and system based on coal mine case text information to solve or alleviate the problems existing in the above-mentioned prior art.

[0006] To achieve the above objectives, this application provides the following technical solution:

[0007] This application provides a method for analyzing the contextual graph of coal mine case text information, including: step S101, extracting scenario events from pre-acquired accident report samples based on a template extraction algorithm to construct an initial contextual graph K1 for a mine outburst accident; step S102, optimizing the accident logic relationships in the initial contextual graph K1 based on a reinforcement learning algorithm to obtain an applied contextual graph of the mine outburst accident; and step S103, displaying the graph data of the applied contextual graph based on a preset layout algorithm to generate a main accident diagram of the mine outburst accident.

[0008] Preferably, in step S101, the step of extracting contextual events from the pre-acquired accident report samples based on the template extraction algorithm includes: constructing an accident corpus of the mine outburst accident based on the acquired accident report samples; wherein, the accident corpus includes: a keyword library and a dictionary library; extracting key sentences from the accident report samples based on the keyword library; and standardizing the extracted key sentences from the coal mine case text based on the dictionary library to obtain standardized expressions of the key sentences in the accident report samples.

[0009] Preferably, in step S101, constructing the initial event graph K1 of the mine outburst accident includes: inputting the standardized expression of the key sentences of the obtained accident report sample into a pre-trained language representation model to vectorize the key sentences; inputting the vectorized expression of the key sentences as node features into a pre-constructed discrimination matrix, calculating the similarity between each word in the key sentences that make up the standardized expression, and performing a reduction operation on the multi-association nodes to obtain all words in the key sentences that make up the standardized expression that meet a preset similarity threshold; wherein, the multi-association nodes represent related words in the key sentences that make up the standardized expression; and performing a clustering operation on all words in the key sentences that make up the standardized expression that meet the preset similarity threshold based on an agglomerative hierarchical clustering algorithm to obtain multiple node clusters of words in the key sentences that meet the preset similarity threshold, and the multiple node clusters forming the initial event graph K1.

[0010] Preferably, according to the formula:

[0011]

[0012] Determine the distinction matrix M * Element a in (K1) * ; where u i u j Sim represents the i-th and j-th words in the key sentences that make up the standardized expression, respectively. BERT (u i u j ) for u i with u j BERT sentence vector similarity.

[0013] Preferably, step S102 includes: based on a reinforcement learning algorithm, adding, deleting, and modifying all node clusters and event logic relationships in the initial event logic graph K1 to generate the application event logic graph K2.

[0014] Preferably, step S102 further includes: performing clustering operations on the application event graph K2 based on agglomerative hierarchical clustering algorithm to obtain the application event graph K3 of the mine outburst accident.

[0015] Preferably, in step S103, based on the agglomerative clustering algorithm, the node clusters of the application event graph K2 are re-clustered according to the similarity between the node clusters, and the similarity between each cluster after the re-clustering is calculated until the similarity between each cluster is less than the preset inter-cluster similarity threshold. The clustering operation ends, and the application event graph K3 of the mine outburst accident is obtained.

[0016] Preferably, in step S103, based on the dynamic programming algorithm, the nodes in the application event graph K2 are arranged in chronological order, and the accident logic relationship between the nodes in the application event graph K2 is simulated to obtain the main accident diagram of the mine outburst accident; or, based on the dynamic programming algorithm, the nodes in the application event graph K3 are arranged in chronological order, and the accident logic relationship between the nodes in the application event graph K3 is simulated to obtain the main accident diagram of the mine outburst accident.

[0017] Preferably, it also includes: identifying node clusters in the event graph K3 based on a community detection algorithm, and displaying the identification results in a layout to obtain a detailed accident diagram of the mine outburst accident.

[0018] This application embodiment also provides a reasoning graph analysis system based on coal mine case text information, including: an initial graph construction unit configured to extract scenario events from pre-acquired accident report samples based on a template extraction algorithm to construct an initial reasoning graph K1 for a mine outburst accident; an application graph construction unit configured to optimize the accident logic relationships in the initial reasoning graph K1 based on a reinforcement learning algorithm to obtain an application reasoning graph for the mine outburst accident; and an accident diagram generation unit configured to lay out and display the graph data of the application reasoning graph based on a preset layout algorithm to generate a main accident diagram for the mine outburst accident.

[0019] Beneficial effects:

[0020] The principle graph analysis method based on coal mine case text information proposed in this application involves extracting scenario events from pre-acquired accident report samples using a template extraction algorithm to construct an initial principle graph K1 for mine outburst accidents. Then, based on a reinforcement learning algorithm, the logical relationships of accidents in the initial principle graph K1 are optimized to obtain an applied principle graph for mine outburst accidents. Finally, based on a preset layout algorithm, the graph data of the applied principle graph is laid out and displayed to generate a main accident diagram for the mine outburst accident. This method automatically generates accident diagrams from accident report samples. These automatically generated diagrams not only help professionals quickly understand the causes of accidents, improve the reading speed of accident reports, lower the analysis threshold for non-professionals, and enhance the efficiency of accident guidance analysis, but also, by conducting multi-dimensional statistical analysis of factors such as time characteristics, location characteristics, unsafe behaviors, unsafe physical states, and job types related to mine outburst accidents, the distribution characteristics of various factors can be analyzed in depth. This provides data support for coal mine safety management, assesses the safety production situation, and guides safety production and refined management. Attached Figure Description

[0021] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. Wherein:

[0022] Figure 1 This is a flowchart illustrating a reasoning graph analysis method based on textual information from coal mine cases, provided according to some embodiments of this application.

[0023] Figure 2 This is an example schematic diagram of a main accident diagram of a mine outburst accident generated according to some embodiments of this application;

[0024] Figure 3 This is an example schematic diagram of a detailed accident breakdown diagram of a mine outburst accident provided according to some embodiments of this application;

[0025] Figure 4 This is a schematic diagram of the structure of a reasoning graph analysis system based on coal mine case text information, provided according to some embodiments of this application. Detailed Implementation

[0026] The present application will now be described in detail with reference to the accompanying drawings and embodiments. Various examples are provided by way of explanation and not by way of limitation. In fact, those skilled in the art will recognize that modifications and variations can be made to the present application without departing from the scope or spirit thereof. For example, a feature shown or described as part of one embodiment may be used in another embodiment to produce yet another embodiment. Therefore, it is desirable that the present application encompass such modifications and variations that fall within the scope of the appended claims and their equivalents.

[0027] The applicant's research found that accident analysis in coal mine safety production is gradually shifting towards the application of natural language processing (NLP) technologies. For example, machine learning-based statistical extraction, NLP-based extraction, and rule-based pattern extraction have enabled information extraction from coal mine case texts. However, machine learning-based statistical extraction is often used for classification and prediction, focusing on the distribution of keywords in the text; NLP-based information extraction is highly complex and not suitable for widespread application; and rule-based pattern extraction has high requirements for the domain format of coal mine case texts, making it difficult to implement. Furthermore, while machine learning-based statistical extraction, NLP-based information extraction, and rule-based pattern extraction all have a high degree of automation, the extracted information is merely a simple listing, which is unfriendly to decision-making, lacks readability, and requires manual matching to determine the accident outcome. If the manual matching is incorrect, the analysis results will inevitably be erroneous, and errors in manual matching cannot be detected and corrected in a timely manner.

[0028] Based on this, the applicant proposes a case study graph analysis method based on coal mine case text information. This method obtains keywords describing accidents from accident report texts through a pre-built keyword library, acquires accident information through word segmentation and a dictionary, and presents an accident diagram that can be understood by various personnel through the general implementation process of case study graphs. Using this automatically generated accident diagram, not only can professionals quickly understand the causes of accidents, improve the reading speed of accident reports, lower the analysis threshold for non-professionals, and enhance the efficiency of accident guidance analysis, but also, by conducting multi-dimensional statistical analysis of factors such as the time characteristics, location characteristics, unsafe behaviors, unsafe physical states, and job types of mine outburst accidents, it can deeply analyze the distribution characteristics of various factors, provide data support for coal mine safety management, assess the safety production situation, and guide safety production and refined management.

[0029] like Figure 1 As shown, this reasoning graph analysis method based on coal mine case text information includes:

[0030] Step S101: Based on the template extraction algorithm, scenario events are extracted from the pre-acquired accident report samples to construct the initial event graph K1 of the mine outburst accident.

[0031] In this application, firstly, accident case reports are extensively collected from various websites, case report text collections, and electronic or paper research papers as accident report samples. Then, based on the obtained accident report samples, a corpus of mine outburst accidents is constructed, namely, a keyword database and a dictionary database for mine outburst accidents. Specifically, manual annotation or natural language processing techniques, such as word frequency statistics, TF-IDF, and topic models, can be used to identify frequently occurring words and topics in the accident report samples. The identified keywords and dictionary terms that meet preset rules are then compiled into a list or file to complete the construction of the keyword database and dictionary database, fully satisfying the uniqueness and quality requirements of the applied domain. In a specific example, a partial example of the constructed keyword database is shown in Table 1.

[0032] Table 1. Examples of Keyword Databases

[0033]

[0034] Examples of constructed dictionary libraries are shown in Table 2:

[0035] Table 2 Examples of Dictionaries

[0036]

[0037]

[0038]

[0039]

[0040] After the accident corpus is constructed, contextual event extraction is performed on the acquired accident report samples based on the corpus. First, key sentences are extracted from the accident report samples based on a keyword database. Specifically, the accident report samples are preprocessed using natural language processing tools (such as NLTK, Stanford CoreNLP, etc.), performing operations such as word segmentation, stop word removal, and part-of-speech tagging. The preprocessed text is then matched with keywords in the keyword database using the Knut-Morris-Platform operation to identify sentences or phrases containing those keywords. Based on the matching results, sentences or phrases containing the keywords are extracted as key sentences.

[0041] Then, based on the dictionary, the key sentences in the extracted coal mine case text are standardized to obtain the standardized expression of the key sentences in the accident report sample. Specifically, through the Knut-Maurice-Platform operation, the key sentences are matched with the words in the dictionary to find the key sentences containing the "keys" in the dictionary; then, through the string replacement operation, based on the matching results, the "keys" in the key sentences are replaced with the corresponding "values" according to the dictionary, thus achieving the standardized expression of the key sentences in the accident report sample.

[0042] After standardizing the key statements of the accident report samples, an initial cause-effect graph K1 for the mine outburst accident is constructed based on these standardized key statements. First, the standardized expressions of the key statements from the accident report samples are input into a pre-trained language representation model to vectorize the key statements. Specifically, the character-level key statements representing the accident causes, obtained from a dictionary, are input into the pre-trained language representation model (Bidirectional Encoder Representation from Transformers, BERT). The output of the BERT model is the vectorized representation of the key statements.

[0043] Then, the vectorized representation of the key sentences is input as node features into the pre-constructed discrimination matrix. The similarity between each word in the key sentences that make up the normalized expression is calculated to reduce the multi-association nodes and obtain all words in the key sentences that make up the normalized expression that meet the preset similarity threshold. Among them, the multi-association nodes represent the related words in the key sentences that make up the normalized expression.

[0044] Here, the discrimination matrix M of the initial logic graph K1 * (K1) is a 2n×2n matrix (n is a positive integer), and follows the formula:

[0045]

[0046] Determine the distinguishing matrix M * Any element a in (K1) * That is, when i is greater than or equal to j, a * (u i u j When i is less than j, a = 0. * (u i u j ) = Sim BERT (u i u jIn matrix representation, u is defined as follows: elements on the diagonal and above it are all 0, and elements below it represent the corresponding similarity scores. i u j Sim represents the i-th and j-th words in the key sentences that make up the standardized expression, respectively. BERT (u i u j ) for u i with u j The BERT sentence vector similarity, i.e., the similarity between the i-th and j-th words u in the key sentences output by the language representation model. i with u j The sentence vector similarity. It should be noted that the i-th and j-th words in the key sentences... i u j It is also the i-th and j-th node features in the initial event graph K1.

[0047] According to the formula:

[0048]

[0049] Determine u i with u j The BERT sentence vector similarity. Where, u i u j = (x1, ..., x t , ..., x m ), where m is the length of the sentence vector output by BERT.

[0050] Finally, based on the agglomerative hierarchical clustering algorithm, clustering is performed on all words in the key sentences that constitute the standardized expression that meet the preset similarity threshold, resulting in multiple node clusters of words in the key sentences that constitute the standardized expression that meet the preset similarity threshold. These multiple node clusters form the initial logical graph K1.

[0051] By distinguishing matrix M * (K1) characterizes the semantic proximity between the features of each node in the initial event graph K1 of a mine outburst accident. If u i with u j If nodes have high similarity, they can be merged into a single node, thus reducing the number of nodes in the graph. In a node encoding network without fusion and generalization, nodes related to mine outburst accidents exhibit strong correlation and similarity, making them more likely to be clustered together. Simultaneously, due to low similarity between different clusters, the distances between them are large, resulting in inter-cluster separation. In other words, based on the semantic similarity of mine outburst accidents, a clustering network is formed where nodes are densely clustered within each cluster but dispersed between clusters.

[0052] In this application, an agglomerative hierarchical clustering algorithm is used to cluster all words in the key sentences that constitute the standardized expression and meet the preset similarity threshold. By sorting the individual words that are initially clustered separately according to the similarity standard, the two clusters with the highest similarity are merged to form a new cluster. This process is repeated until the similarity between the new clusters is less than the given inter-cluster similarity threshold (i.e., the preset similarity threshold). The clustering operation ends, and the node clusters that finally meet the inter-cluster similarity threshold constitute the nodes of the initial event graph K1.

[0053] In other words, after obtaining the distinction matrix M * After (K1), assume the distinguishing matrix M * The maximum value in (K1) is a * (u i u j ), a * (u i u j Characterization matrix M * The i-th and j-th words in (K1) i u j The semantic similarity, i.e., u i u j For nodes of the same type, cluster u according to the agglomerative hierarchical clustering algorithm. i u j By merging, a new cluster (u) is obtained. i u j New cluster (u) i u j The center of ) is u ij Next, the similarity between each new cluster is calculated. Specifically, according to the formula:

[0054]

[0055] Determine the similarity between the new clusters. Representing cluster u ij , cluster u t The number of nodes in the array.

[0056] Step S102: Based on the reinforcement learning algorithm, optimize the accident logic relationship in the initial event graph K1 to obtain the application event graph of the mine outburst accident.

[0057] Specifically, based on reinforcement learning algorithms, all node clusters and accident logic relationships in the initial event graph K1 are added, deleted, and modified to generate the application event graph K2. In other words, using a reinforcement learning model (e.g., a neural network model), each node in the initial event graph K1 and its accident logic relationships with other nodes are added, deleted, and modified. When the loss function of the reinforcement learning model converges, the node and its associated accident logic relationships are considered optimal. Then, the next node in the initial event graph K1 and its associated accident logic relationships are optimized until all nodes in the initial event graph K1 are optimized. The resulting event graph is the application event graph K2 for mine outburst accidents.

[0058] Furthermore, based on agglomerative hierarchical clustering, the application event graph K2 can be clustered to obtain the application event graph K3 for mine outburst accidents. Specifically, based on the agglomerative clustering algorithm, the node clusters of the application event graph K2 are further sub-clustered according to the similarity between the node clusters, and the similarity between each sub-cluster is calculated until the similarity between each sub-cluster is less than a preset inter-cluster similarity threshold. The clustering operation then ends, resulting in the application event graph K3 for mine outburst accidents. It should be noted that the steps and procedures for obtaining the application event graph K3 by performing agglomerative hierarchical clustering on the application event graph K2 are similar to those for constructing the initial event graph K1, and will not be elaborated upon here.

[0059] Step S103: Based on the preset layout algorithm, the map data of the application event map is laid out and displayed to generate the main accident diagram of the mine outburst accident.

[0060] After obtaining the application event diagram of a mine outburst accident, in order to further reduce the threshold for reading, understanding, and analyzing mine outburst accidents, this application uses a preset layout algorithm to display the nodes and logical relationships in the application event diagram in the form of an accident illustration, such as... Figure 2 , Figure 3 As shown. For example, the nodes and logical relationships in the application event diagram K2 or application event diagram K3 can be displayed in the form of an accident master diagram, so that the basic information of the mine outburst accident (such as time, location, casualties, and the type of operation that caused the accident), as well as the accident cause information (such as unsafe actions and unsafe physical states), the attribute values ​​of the accident cause information (node ​​attributes in the application event diagram), and emergency rescue measures can be summarized and displayed.

[0061] Specifically, based on the dynamic programming algorithm, the nodes in the application event graph K2 are arranged in chronological order, and the accident logic relationship between the nodes in the application event graph K2 is simulated to obtain the main accident diagram of the mine outburst accident; or, based on the dynamic programming algorithm, the nodes in the application event graph K3 are arranged in chronological order, and the accident logic relationship between the nodes in the application event graph K2 is simulated to obtain the main accident diagram of the mine outburst accident.

[0062] In the distance-based example, a time-based layout algorithm—dynamic programming—is used to arrange the nodes in the event graph in chronological order and simulate the evolution of the mine outburst event by the movement (i.e., logical relationships) between the nodes. Specifically, first, the nodes are randomly arranged; then, for each pair of nodes, the attraction and repulsion forces are calculated based on the distance and connectivity between the nodes. Attraction brings adjacent nodes closer, while repulsion separates non-adjacent nodes; next, the velocity of each node is calculated based on the attraction and repulsion forces between the nodes, and the node position is updated; this process is repeated until all nodes reach their target positions, completing the main event graph.

[0063] Furthermore, detailed accident diagrams can be generated for each node cluster in the application event graph K3 as needed, such as... Figure 3 As shown, based on the community detection algorithm, node clusters in the event graph K3 are identified, and the identification results are displayed in a layout to obtain a detailed accident diagram of a mine outburst accident. It should be noted that the specific operations and processes for generating detailed accident diagrams for each node cluster in the event graph K3 are similar to those for generating the main accident diagram, and will not be elaborated upon here.

[0064] In addition to displaying mine outburst accidents in the form of accident diagrams, this application can also supplement the display of mine outburst accidents by using statistical tables and other information display methods.

[0065] Therefore, through the general implementation process of the incident diagram, an accident diagram that can be understood by various personnel is presented. Using this automatically generated accident diagram, not only can professionals quickly understand the causes of accidents, improve the reading speed of accident reports, lower the analysis threshold for non-professionals, and improve the efficiency of accident guidance analysis, but also, by conducting multi-dimensional statistics on factors such as time characteristics, location characteristics, unsafe behaviors, unsafe physical states, and types of work in mine outburst accidents, the distribution characteristics of various factors can be analyzed in depth. This provides data support for coal mine safety management, assesses the safety production situation, and guides safety production and refined management.

[0066] This application also provides a reasoning graph analysis system based on textual information from coal mine cases, such as... Figure 4As shown, the system includes: an initial map construction unit 401, an application map construction unit 402, and an accident diagram generation unit 403.

[0067] The initial graph construction unit 401 is configured to extract scenario events from pre-acquired accident report samples based on a template extraction algorithm to construct an initial event graph K1 for a mine outburst accident. The application graph construction unit 402 is configured to optimize the accident logic relationships in the initial event graph K1 based on a reinforcement learning algorithm to obtain an application event graph for a mine outburst accident. The accident diagram generation unit 403 is configured to lay out and display the graph data of the application event graph based on a preset layout algorithm to generate the main accident diagram for a mine outburst accident.

[0068] The principle graph analysis system based on coal mine case text information provided in this application can implement the steps and processes of any of the above-mentioned principle graph analysis methods based on coal mine case text information and achieve the same technical effect, which will not be described in detail here.

[0069] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for analyzing the logical graph of coal mine case text information, characterized in that, include: Step S101: Based on the obtained accident report samples, construct an accident corpus of mine outburst accidents; wherein, the accident corpus includes: a keyword database and a dictionary database; Based on the keyword library, key phrases were extracted from the accident report samples; Based on the dictionary, the key sentences in the extracted coal mine case texts are standardized to obtain the standardized expressions of the key sentences in the accident report samples; The standardized expression of the key sentences in the obtained accident report sample is input into a pre-trained language representation model to vectorize the key sentences; The vectorized representation of the key statement is input as node feature into a pre-constructed discrimination matrix. The similarity between each word in the key statement that makes up the normalized expression is calculated to perform a reduction operation on multiple associated nodes, obtaining all words in the key statement that meet the preset similarity threshold; wherein, according to the formula: ； Determine the distinction matrix elements in ;in, These represent the first and second key statements that make up the standardized expression. The, the One word, for and of Sentence vector similarity; The multiple associated nodes represent key sentences that contain related words that constitute the standardized expression; Based on a cohesive hierarchical clustering algorithm, clustering is performed on all words in the key sentences that constitute the standardized expression and meet the preset similarity threshold. This results in multiple node clusters of words in the key sentences that meet the preset similarity threshold, and these node clusters form an initial logical graph. ; Step S102: Based on the reinforcement learning algorithm, process the initial event graph. The accident logic relationships in the data are optimized to obtain the application logic diagram of the mine outburst accident. Step S103: Based on the preset layout algorithm, the map data of the application event map is laid out and displayed to generate the main accident diagram of the mine outburst accident.

2. The reasoning graph analysis method based on coal mine case text information according to claim 1, characterized in that, Based on reinforcement learning algorithms, the initial event graph is... All node clusters and event logic relationships in the graph are added, deleted, and modified to generate the application event graph. .

3. The reasoning graph analysis method based on coal mine case text information according to claim 2, characterized in that, Also includes: Based on the agglomerative hierarchical clustering algorithm, the application logic graph is analyzed. Clustering operations were performed to obtain the application causal graph of the mine outburst accident. ; Based on the dynamic programming algorithm, the application event graph is analyzed. The nodes are arranged in chronological order, simulating the application's event graph. By analyzing the logical relationships between the nodes in the diagram, a main diagram of the mine outburst accident can be obtained.

4. The reasoning graph analysis method based on coal mine case text information according to claim 3, characterized in that, In step S103, Based on the agglomerative clustering algorithm, and according to the application logic graph... The similarity between node clusters is used to evaluate the application logic graph. The node clusters are further sub-clustered, and the similarity between each sub-cluster is calculated until the similarity between each sub-cluster is less than a preset inter-cluster similarity threshold. The clustering operation ends, and the application causal graph of the mine outburst accident is obtained. .

5. The reasoning graph analysis method based on coal mine case text information according to claim 2, characterized in that, In step S103, Based on the dynamic programming algorithm, the application event graph is analyzed. The nodes in the graph are arranged in chronological order, simulating the application's event graph. By analyzing the logical relationships between the nodes in the diagram, a main diagram of the mine outburst accident can be obtained.

6. The reasoning graph analysis method based on coal mine case text information according to claim 1, characterized in that, Also includes: Based on the community detection algorithm, the event graph is analyzed. The node clusters in the diagram are identified, and the identification results are displayed in a layout to obtain a detailed diagram of the mine outburst accident.

7. A reasoning graph analysis system based on textual information from coal mine cases, characterized in that, include: The initial graph construction unit is configured to construct an accident corpus of mine outburst accidents based on the acquired accident report samples; wherein, the accident corpus includes: a keyword library and a dictionary library; Based on the keyword library, key phrases were extracted from the accident report samples; Based on the dictionary, the key sentences extracted from the coal mine case text are standardized to obtain the standardized expression of the key sentences in the accident report sample; The standardized expression of the key sentences in the obtained accident report sample is input into a pre-trained language representation model to vectorize the key sentences; The vectorized representation of the key statement is input as node feature into a pre-constructed discrimination matrix. The similarity between each word in the key statement that makes up the normalized expression is calculated to perform a reduction operation on multiple associated nodes, obtaining all words in the key statement that meet the preset similarity threshold; wherein, according to the formula: ； Determine the distinction matrix elements in ;in, These represent the first and second key statements that make up the standardized expression. The, the One word, for and of Sentence vector similarity; The multiple associated nodes represent key sentences that contain related words that constitute the standardized expression; Based on a cohesive hierarchical clustering algorithm, clustering is performed on all words in the key sentences that constitute the standardized expression and meet the preset similarity threshold. This results in multiple node clusters of words in the key sentences that meet the preset similarity threshold, and these node clusters form an initial logical graph. ; The application graph construction unit is configured to use a reinforcement learning algorithm to process the initial event graph. The accident logic relationships in the data are optimized to obtain the application logic diagram of the mine outburst accident. The accident diagram generation unit is configured to display the diagram data of the application event diagram based on a preset layout algorithm, and generate the main accident diagram of the mine outburst accident.

Images