Illegal case handling event semantic understanding and behavior judgment method based on multi-chain perception

Abstract

The embodiment of the invention provides an illegal case handling event semantic understanding and behavior judgment method based on multi-chain perception, and belongs to the technical field of natural language processing, and the method specifically comprises the steps of 1, obtaining a to-be-processed original text data set; 2, constructing a grammar graph and a semantic graph for the case handling event text; 3, generating an adjacent matrix for the syntax graph and the semantic graph; 4, performing multi-step propagation by a multi-chain mechanism to capture a long-distance dependency relationship; step 5, dynamically distributing attention by a sub-graph attention mechanism; step 6, introducing a gating mechanism to control information flow; step 7, mapping the high-dimensional features to a probability distribution space by the multi-layer perceptron; 8, the rule engine adjusts the prediction score to accord with the legal practice; and 9, selecting a behavior judgment result with the highest probability. Through the scheme disclosed by the invention, key information and deep semantics in a case are accurately captured, illegal behaviors are judged, and the conditions of misjudgment and missed judgment are reduced, so that judicial fairness and efficiency are improved.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the technical field of natural language processing, and in particular to a method for semantic understanding and behavior determination of illegal case handling events based on multi-chain perception. Background Art

[0002] With the continuous accumulation of judicial data and the rapid development of artificial intelligence technology, using intelligent technology to improve judicial efficiency and fairness has become a research hotspot. In the handling of illegal case handling events, accurately understanding the semantics of cases and determining the illegality of behaviors is crucial. However, there are many deficiencies in the existing technologies.

[0003] In terms of semantic understanding, traditional methods mainly rely on simple text feature extraction technologies such as the bag-of-words model and TF-IDF, and it is difficult to capture the deep semantics and complex logical relationships in the text. With the development of deep learning, methods such as convolutional neural networks (CNNs), recurrent neural networks (RNNs) and their variants such as long short-term memory networks (LSTMs) and pre-trained BERT have been applied to semantic understanding. These methods can capture the deep meaning of the text by learning the complex features of the text and have improved performance. However, these models still have limitations when dealing with long texts and complex syntactic structures. For example, when dealing with sentences containing multiple nested clauses and modifiers, existing methods often only focus on the semantic information of the text and it is difficult to accurately parse the semantics and syntactic structures of the sentences. Therefore, syntactic understanding can avoid misjudgments caused by semantic biases, thereby deeply understanding the complex logical relationships in the case and improving the accuracy of determining illegal behaviors.

[0004] In terms of behavior determination, existing technologies often rely on simple rule matching or shallow machine learning models and cannot accurately judge complex illegal behaviors. For example, when judging illegal case handling events, only a single factor is considered while ignoring other relevant factors in the case and their mutual relationships. At the same time, existing methods perform poorly when dealing with fuzzy and uncertain information and are prone to misjudgments and missed judgments. In addition, existing technologies lack effective utilization of judicial domain knowledge and cannot integrate legal knowledge and judicial practice experience into the behavior determination process, resulting in a deviation between the determination result and the actual judicial situation.

[0005] In summary, there are obvious deficiencies in the existing technologies in the semantic understanding and behavior determination of illegal case handling events and they cannot meet the requirements of the judicial domain for intelligence and precision. Therefore, a new method is needed to solve these problems and improve the processing efficiency and accuracy of illegal case handling events. Summary of the Invention

[0006] In view of this, embodiments of the present disclosure provide a method for semantic understanding and behavior determination of illegal case handling events based on multi-chain perception, which at least partially solves the problems of poor determination efficiency and accuracy in the prior art.

[0007] Embodiments of the present disclosure provide a method for semantic understanding and behavior determination of illegal case handling events based on multi-chain perception, including:

[0008] Step 1, obtain the original text dataset to be processed;

[0009] Step 2, construct a syntax graph and a semantic graph for the case handling event text;

[0010] Step 3, generate an adjacency matrix for the syntax graph and the semantic graph;

[0011] Step 4, use a multi-chain mechanism to perform multi-step propagation to capture long-distance dependencies;

[0012] Step 5, use a subgraph attention mechanism to dynamically allocate attention;

[0013] Step 6, introduce a gating mechanism to control information flow;

[0014] Step 7, use a multi-layer perceptron to map high-dimensional features to a probability distribution space;

[0015] Step 8, use a rule engine to adjust the prediction score to conform to legal practice;

[0016] Step 9, select the behavior determination result with the highest probability.

[0017] According to a specific implementation manner of the embodiments of the present disclosure, the specific steps of Step 2 include:

[0018] Step 2.1, perform dependency syntactic analysis on the text to construct a syntax graph G syn =(V syn , E syn );

[0019] Step 2.2, perform semantic dependency analysis on the text to construct a semantic graph G sem =(V sem , E sem );

[0020] According to a specific implementation manner of the embodiments of the present disclosure, the specific steps of Step 3 include:

[0021] Step 3.1, generate an adjacency matrix A syn for the syntax graph G syn ;

[0022] Step 3.2, generate an adjacency matrix A sem for the semantic graph Gsem ;

[0023] According to a specific implementation manner of an embodiment of the present disclosure, step 4 specifically includes:

[0024] Step 4.1, initialize the text embedding vector through a pre-trained Bert model;

[0025] Step 4.2, utilize the multi-chain mechanism to perform multi-step propagation in the chain structure of the adjacency matrix to capture the long-distance dependency relationships between words;

[0026] According to a specific implementation manner of an embodiment of the present disclosure, step 5 specifically includes:

[0027] Step 5.1, introduce the subgraph attention mechanism to calculate the attention scores for the internal nodes of each subgraph with different numbers of chains;

[0028] Step 5.2, normalize to obtain the attention coefficients, and update the node features by weighted summation;

[0029] According to a specific implementation manner of an embodiment of the present disclosure, step 6 specifically includes:

[0030] Introduce the gating mechanism, calculate the gating signal for each node, and adjust the node representation through the gating signal to obtain the final node representation.

[0031] According to a specific implementation manner of an embodiment of the present disclosure, step 7 specifically includes:

[0032] Step 71, use the multi-layer perceptron MLP to map the final node representation from the high-dimensional features to the probability distribution space of the violation event categories;

[0033] Step 72, obtain the category prediction score vector through the fully connected layer, and normalize the prediction score vector using the softmax function to obtain the prediction probability distribution of each crime name;

[0034] According to a specific implementation manner of an embodiment of the present disclosure, step 8 specifically includes:

[0035] Introduce a rule engine, set the threshold θ for specific laws and regulations, and adjust the predicted scores to conform to the actual legal judgment logic.

[0036] According to a specific implementation manner of an embodiment of the present disclosure, step 9 specifically includes:

[0037] Based on the predicted scores adjusted by the rule engine, select the behavior determination with the highest probability as the final prediction result.

[0038] The semantic understanding and behavior determination solution for illegal case handling events based on multi-chain perception in the embodiments of the present disclosure includes: Step 1, obtaining the original text dataset to be processed; Step 2, constructing a syntax graph and a semantic graph for the case handling event text; Step 3, generating an adjacency matrix for the syntax graph and the semantic graph; Step 4, using a multi-chain mechanism to perform multi-step propagation to capture long-distance dependencies; Step 5, using a subgraph attention mechanism to dynamically allocate attention; Step 6, introducing a gating mechanism to control information flow; Step 7, using a multi-layer perceptron to map high-dimensional features to a probability distribution space; Step 8, using a rule engine to adjust the prediction score to conform to legal practice; Step 9, selecting the behavior determination result with the highest probability.

[0039] The beneficial effects of the embodiments of the present disclosure are as follows: Through the semantic understanding and behavior determination method for illegal case handling events based on multi-chain perception, the present invention significantly improves the accuracy and reliability of semantic understanding and behavior determination. Through the multi-chain perception mechanism, the model can deeply mine the semantic and syntactic information in the case text, accurately capture the key information and deep semantics in the case, and avoid misjudgments caused by semantic deviations. At the same time, combined with legal knowledge and a rule engine, the model can comprehensively and accurately determine illegal behaviors, reducing the situations of misjudgments and missed judgments. Experimental results show that this method performs excellently in the handling of complex cases, can effectively improve judicial fairness and efficiency, and provides strong technical support for judicial practice. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the following will briefly introduce the accompanying drawings required for use in the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present disclosure. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0041] Figure 1 It is a schematic flowchart of a method for semantic understanding and behavior determination of illegal case handling events based on multi-chain perception provided by the embodiments of the present disclosure;

[0042] Figure 2 It is a system architecture diagram corresponding to a method for semantic understanding and behavior determination of illegal case handling events based on multi-chain perception provided by the embodiments of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] The following will describe the embodiments of the present disclosure in detail with reference to the accompanying drawings.

[0044] The following describes the embodiments of the present disclosure through specific examples. Those skilled in the art can easily understand the other advantages and effects of the present disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other. Based on the embodiments in the present disclosure, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present disclosure.

[0045] It should be noted that the following describes various aspects of the embodiments within the scope of the appended claims. It should be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is illustrative only. Based on the present disclosure, those skilled in the art should understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects described herein can be used to implement a device and / or practice a method. Additionally, this device and / or this method can be implemented using other structures and / or functions in addition to one or more of the aspects described herein.

[0046] It should also be noted that the diagrams provided in the following embodiments only illustrate the basic concept of the present disclosure in a schematic manner. The diagrams only show the components related to the present disclosure, rather than being drawn according to the number, shape, and size of the components in actual implementation. The type, quantity, and proportion of each component in its actual implementation can be arbitrarily changed, and the component layout type may also be more complex.

[0047] In addition, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0048] The embodiments of the present disclosure provide a method for semantic understanding and behavior determination of illegal case-handling events based on multi-chain perception. The method can be applied to the relationship extraction process in the knowledge graph construction scenario.

[0049] See Figure 1 , which is a schematic flowchart of a method for semantic understanding and behavior determination of illegal case-handling based on multi-chain perception provided by the embodiments of the present disclosure. As Figure 1 shown, the method mainly includes the following steps:

[0050] Step 1, obtain the original text data set to be processed;

[0051] Step 2: Construct a syntax graph and a semantic graph for the case handling event text;

[0052] Further, the specific steps of Step 2 include:

[0053] Step 2.1: Perform dependency syntactic analysis on the text to construct a syntax graph G syn =(V syn , E syn );

[0054] Step 2.2: Perform semantic dependency analysis on the text to construct a semantic graph G sem =(V sem , E sem );

[0055] When specifically implemented, constructing a syntax graph and a semantic graph for the case handling event text includes:

[0056] S2.1: Use the HanLP toolkit to perform dependency syntactic analysis on the text to construct a syntax graph G syn =(V syn , E syn ); where V syn ={w1, w2,..., w n} represents the words in the text, and E syn ={(w i , w j )|r ij} represents the syntactic relationship between words, and r ij is a specific syntactic relationship type. In the syntax graph, the general predicate serves as the root node, and all the child nodes connected to its root node are called the first chain, generally including the subject, object, and adverbial directly connected to the predicate, etc. All the grandchild nodes are the second chain.

[0057] S2.2: Construct a semantic graph G through semantic dependency analysis sem =(V sem , E sem ), V sem ={w1, w2,..., w n} also represents the words, and E sem ={(w i , w j )|s ij} represents the semantic relationship between words, and s ij is a specific semantic relationship type. Similarly, in the semantic graph, the general verb serves as the root node, and we also call the connection with other nodes a chain.

[0058] Step 3: Generate an adjacency matrix for the syntax graph and the semantic graph;

[0059] Further, step 3 specifically includes:

[0060] Step 3.1, generate an adjacency matrix A for the syntactic graph G syn ; syn ;

[0061] Step 3.2, generate an adjacency matrix A for the semantic graph G sem ; sem ;

[0062] Specifically in implementation, generating adjacency matrices for the syntactic graph and the semantic graph includes:

[0063] S3.1: Generate an adjacency matrix A for the syntactic graph G syn ; if there is a syntactic relationship between w syn and w j and w j , the corresponding element A syn [i][j] is set to a value representing the relationship strength, which can be determined according to the syntactic relationship type or a predefined weight mapping. Usually, we determine the starting weight based on the number of chains and components it belongs to. For example, the first chain is higher than the second chain, the subject and object are higher than the adverbial, and those without connection are set to 0.

[0064] S3.2: Generate an adjacency matrix A for the semantic graph G sem ; if there is a semantic relationship between w sem and w i and w j , the corresponding element A sem [i][j] is set to a value representing the relationship strength, which can be determined according to the semantic relationship

[0065] type or a predefined weight mapping. Usually, we determine the starting weight based on the number of chains and components it belongs to.

[0066] Step 4, use the multi-chain mechanism to perform multi-step propagation to capture long-distance dependency relationships;

[0067] Further, step 4 specifically includes:

[0068] Step 4.1, initialize the text embedding vectors through a pre-trained Bert model;

[0069] Step 4.2, use the multi-chain mechanism to perform multi-step propagation in the chain structure of the adjacency matrix to capture the long-distance dependency relationships between words;

[0070] Specifically in implementation, using the multi-chain mechanism to perform multi-step propagation to capture long-distance dependency relationships includes:

[0071] S4.1: Initialize the text embedding vectors through the pre-trained Bert model. The Bert model will tokenize and tokenize the input text, and convert each word into the corresponding word vector.

[0072] S4.2: Use the multi-chain mechanism to perform multi-step propagation in the chain structure of the adjacency matrix to capture the long-distance dependencies between words. Let h t denote the hidden state of the node at the t-th chain in the propagation. The initial hidden state h0 is initialized according to the word embeddings of the nodes in the syntax graph and the semantic graph. Use the information propagation formula: h t+1 = to perform information propagation and capture the long-distance dependencies between words.

[0073] Step 5, the subgraph attention mechanism dynamically allocates attention;

[0074] Further, the specific implementation of step 5 includes:

[0075] Step 5.1, introduce the subgraph attention mechanism to calculate the attention scores for the internal nodes of each subgraph with different numbers of chains;

[0076] Step 5.2, normalize to obtain the attention coefficients, and update the node features by weighted summation;

[0077] Specifically, when implementing, the subgraph attention mechanism dynamically allocates attention, including:

[0078] S5.1: Introduce the subgraph attention mechanism, which allows the model to dynamically allocate attention within each subgraph, highlighting the nodes and edges that are more critical to the current task. For node i in each subgraph, for each subgraph, we calculate the attention scores between nodes. Let the nodes in the subgraph be represented as v1, v2,..., v m , and their corresponding feature vectors be h1, h2,..., h m . First, map the node features to a new space through a shared linear transformation: e ij = LeakyReLU(W1 · [h i ; h j ).

[0079] S5.2: Normalize through the softmax function to obtain the attention coefficients Update the node features by weighted summation

[0080] Step 6, introduce a gating mechanism to control the information flow;

[0081] Further, the specific implementation of step 6 includes:

[0082] Introduce a gating mechanism, calculate the gating signal for each node, and adjust the node representation through the gating signal to obtain the final node representation.

[0083] Specifically, when implementing, introduce a gating mechanism to control the information flow, including:

[0084] S6: Control the information flow and prevent overfitting, calculate the gating signal g for each node i =σ(W g [h i ||h t+1 [i]), and adjust the node representation through the gating signal to obtain the final node representation H final [i]=g i ⊙h t+1 [i]+(1 - g i )⊙h i .

[0085] Step 7, the multi-layer perceptron maps the high-dimensional features to the probability distribution space;

[0086] Furthermore, the specific steps of step 7 include:

[0087] Step 7.1, use the multi-layer perceptron MLP to map the final node representation, and map the high-dimensional features to the probability distribution space of the violation event categories;

[0088] Step 7.2, obtain the category prediction score vector through the fully connected layer, and use the softmax function to normalize the prediction score vector to obtain the prediction probability distribution of each crime name;

[0089] Specifically, when implementing, the multi-layer perceptron maps the high-dimensional features to the probability distribution space, including:

[0090] S7.1: Use the multi-layer perceptron MLP to map the final node representation, and map the high-dimensional features to the probability distribution space of the violation event categories;

[0091] S7.2: Obtain the category prediction score vector z = W fc H final +b fc , and use the softmax function to normalize the prediction score vector to obtain the prediction probability distribution of each crime name

[0092] Step 8, the rule engine adjusts the prediction scores to conform to legal practice;

[0093] Furthermore, the specific steps of step 8 include:

[0094] Introduce a rule engine, set the threshold θ for specific laws and regulations, and adjust the prediction scores to conform to the actual legal judgment logic.

[0095] In specific implementation, the rule engine adjusts the prediction score to conform to legal practice, including:

[0096] S8: Introduce a rule engine to deeply analyze specific laws and regulations, sort out the key judgment elements and criteria, set a threshold θ. If the behavioral characteristics corresponding to the prediction score trigger the preset threshold, the rule engine adjusts the prediction score to conform to the actual legal judgment logic.

[0097] Step 9, select the behavioral judgment result with the highest probability.

[0098] Furthermore, the specific content of step 9 includes:

[0099] According to the prediction score adjusted by the rule engine, select the behavioral judgment with the highest probability as the final prediction result.

[0100] In specific implementation, selecting the behavioral judgment result with the highest probability includes:

[0101] S9: According to the prediction score adjusted by the rule engine, select the behavioral judgment with the highest probability as the final prediction.

[0102] The method for semantic understanding and behavioral judgment of illegal case handling events based on multi-chain perception provided in this embodiment constructs a syntax graph and a semantic graph through dependency syntactic analysis and semantic dependency analysis, uses the multi-chain perception mechanism for long-distance propagation and multi-step reasoning of information, dynamically adjusts the information propagation weight, and deeply mines the semantic and syntactic information in the case text. At the same time, introduce a subgraph attention mechanism and a gating mechanism to focus on important information and control the information flow to prevent overfitting. In addition, map high-dimensional features to the probability distribution space of illegal event categories through a multi-layer perceptron. And in combination with the rule engine module, adjust the probability distribution predicted by the model based on legal rules to ensure that the prediction result conforms to judicial practice. Finally, select the behavioral judgment with the highest probability as the prediction result. It effectively solves the problems of inaccurate semantic understanding and unreliable behavioral judgment in the existing semantic understanding and behavioral judgment technologies for illegal case handling events, significantly improves the accuracy and reliability of semantic understanding and behavioral judgment, reduces misjudgment and missed judgment, and further improves judicial fairness and efficiency. It has broad application prospects in emerging fields such as the judicial system and legal big data, can create considerable economic benefits, and strongly promotes the process of judicial modernization.

[0103] It should be understood that each part of the present disclosure can be implemented by hardware, software, firmware, or a combination thereof.

[0104] As described above, it is only the specific implementation manner of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present disclosure should be covered within the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure shall be subject to the protection scope of the claims.

Claims

1. A semantic understanding and behavior determination method for illegal case-handling events based on multi-chain perception, characterized in that, Including: Step 1: Obtain the original text dataset to be processed; Step 2: Construct a syntax graph and a semantic graph for the case handling event text; Step 3: Generate an adjacency matrix for the syntax graph and the semantic graph; Step 4: Use a multi-chain mechanism to perform multi-step propagation to capture long-distance dependency relationships; Step 5: Use a subgraph attention mechanism to dynamically allocate attention; Step 6: Introduce a gating mechanism to control the information flow; Step 7: Use a multi-layer perceptron to map high-dimensional features to a probability distribution space; Step 8: Use a rule engine to adjust the prediction scores to conform to legal practice; Step 9: Select the behavior determination result with the highest probability.

2. The method according to claim 1, characterized in that, The specific content of Step 2 includes: Step 2.1, perform dependency syntactic analysis on the text to construct a syntax graph G syn =(V syn , E syn ); Step 2.2, perform semantic dependency analysis on the text to construct a semantic graph G sem =(V sem , E sem ).

3. The method according to claim 2, wherein The specific content of Step 3 includes: Step 3.1, generate an adjacency matrix A for the syntax graph G syn according to the relationship strength between nodes syn ; Step 3.2, generate an adjacency matrix A for the semantic graph G sem according to the relationship strength between nodes sem .

4. The method according to claim 3, characterized in that, The specific content of Step 4 includes: Step 4.1: Initialize the text embedding vector through a pre-trained Bert model; Step 4.2: Use a multi-chain mechanism to perform multi-step propagation in the chain structure of the adjacency matrix to capture the long-distance dependency relationships between words.

5. The method according to claim 4, wherein The specific content of Step 5 includes: Step 5.1: Introduce a subgraph attention mechanism to calculate the attention scores for the internal nodes of each subgraph with different numbers of chains; Step 5.2: Normalize to obtain the attention coefficients, and update the node features by weighted summation.

6. The method according to claim 5, wherein The specific content of Step 6 includes: Introduce a gating mechanism, calculate the gating signal for each node, and adjust the node representation through the gating signal to obtain the final node representation.

7. The method according to claim 6, wherein The specific content of Step 7 includes: Step 7.1: Use a multi-layer perceptron (MLP) to map the final node representation, and map the high-dimensional features to the probability distribution space of the violation event categories; Step 7.2: Obtain the category prediction score vector through a fully connected layer, and use the softmax function to normalize the prediction score vector to obtain the prediction probability distribution for each crime name.

8. The method according to claim 7, wherein The specific content of Step 8 includes: Introduce a rule engine, set a threshold θ for specific laws and regulations, and adjust the prediction scores to conform to the actual legal determination logic.

9. The method according to claim 8, wherein The specific content of Step 9 includes: According to the prediction scores adjusted by the rule engine, select the behavior determination with the highest probability as the final prediction result.

Images