Law enforcement basis judgment method and system in administrative law enforcement

By encoding and building legal graphs in administrative law enforcement, using graph neural network to predict the correctness probability of law enforcement basis, the problem of traditional models dealing with long text restrictions is solved, and the efficiency and accuracy of law enforcement basis judgment is improved.

CN118916481APending Publication Date: 2024-11-08SHANDONG UNIV
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410603717.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-05-15
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

During the administrative law enforcement process, long statutory provisions and case description text information exceed the maximum input length limit of the traditional model, and there is a lack of technical solutions to assist judicial personnel in determining the accuracy of the basis for law enforcement.

Method used

By searching relevant law enforcement bases from the legal corpus, coding the case description and law enforcement bases, building a legal diagram, using the graph neural network to fuse information, extracting features and predicting the correctness probability through a classifier, assisting judicial personnel to judge the correctness of law enforcement bases .

Benefits of technology

It effectively overcomes the limitations of long text information processing, improves the efficiency of judging the basis for law enforcement, and provides tools to assist judicial personnel to judge the rationality and correctness of the law enforcement process.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118916481A_ABST
    Figure CN118916481A_ABST
Patent Text Reader

Abstract

The invention provides a law enforcement basis judgment method and system in administrative law enforcement. The law enforcement basis judgment method comprises the following steps: extracting case description text information and law enforcement basis text information quoted by a document from a case document; the case description text information and the law enforcement basis text information are coded respectively and mapped to dense representation, and case description embedding and law enforcement basis embedding are obtained respectively; constructing a law diagram, fusing information in the law diagram to enrich law enforcement basis embedding, and obtaining a retrieved basis; extracting features of the basis embedded and retrieved according to the case description and law enforcement basis, and performing dimension transformation on the features to obtain feature representation; and based on the feature representation, determining a correctness probability prediction result by using a classifier, comparing the correctness probability prediction result with a preset probability threshold, and setting a judgment label according to a comparison result. The method can assist a judge in judging whether the law enforcement basis is used correctly or not.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of legal artificial intelligence, and specifically relates to a method and system for determining enforcement basis in administrative law enforcement. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] In the process of administrative law enforcement, it is necessary to consult relevant legal provisions and the enforcement of similar cases in the past to analyze the case. This part of the work often takes up a lot of time for administrative law enforcement personnel. The emergence of legal artificial intelligence has effectively improved the efficiency of this link.

[0004] Previous studies on the task of recommending legal provisions have been limited by the size of the data set. The research methods are mainly based on traditional machine learning. Currently, text classification models based on neural networks are often used to improve the prediction quality by extracting deep text features. The most commonly used models are CNN, TextCNN, LSTM, and GRU. Yang et al. and Xu et al. continued the classification idea and established the relationship between subtasks within the multi-task framework. While completing the task of recommending legal provisions, they can also predict the crime and sentence. The literature uses a network structure based on graph convolution.

[0005] However, according to the inventor's understanding, the current assistant system still has the problem that the statutory provisions and case descriptions are very long text sequences, which may be as long as thousands of words, and it is necessary to overcome the maximum input length limit, and there is no technical solution that can assist judicial personnel in determining the accuracy of the basis for administrative enforcement. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a method and system for determining the enforcement basis in administrative law enforcement. The present invention retrieves relevant enforcement basis from the legal corpus, and then classifies the case description, enforcement basis and retrieval results, and finally obtains the correctness probability result of the enforcement process to assist judicial personnel in judging whether the enforcement basis is used correctly.

[0007] According to some embodiments, the present invention adopts the following technical solutions:

[0008] A method for determining enforcement basis in administrative enforcement comprises the following steps:

[0009] Extracting case description text information and enforcement basis text information cited in the documents from case files;

[0010] Encode the case description text information and the law enforcement basis text information respectively, and map them into dense representations to obtain the case description embedding and the law enforcement basis embedding respectively;

[0011] Construct a legal graph, integrate the information in the legal graph to enrich the enforcement basis embedding, and obtain the retrieved basis;

[0012] Extract the features of the case description embedding, the enforcement basis embedding and the retrieved basis, perform dimension transformation on the features, and obtain feature representation;

[0013] Based on the feature representation, a classifier is used to determine a correctness probability prediction result, the correctness probability prediction result is compared with a preset probability threshold, and a determination label is set according to the comparison result.

[0014] As an optional implementation, the process of encoding the case description text information includes: inputting the case description text information into a pre-trained BERT-based model with a weight of θ, the model maps the input sequence of n tokens from the vocabulary W to a d-dimensional real-valued token embedding, and embedding the [CLS] token representation of the last layer as the case description.

[0015] As a further implementation method, in the pre-training stage of the BERT-based model, after the model is pre-trained and fine-tuned, the output of the [CLS] token can capture the semantics and information of the entire sentence or sequence.

[0016] As an optional implementation method, the specific process of encoding the law enforcement basis text information includes: each law enforcement basis is divided into a plurality of text paragraphs [p1, p2, ..., p m ], where paragraph p i is a token sequence of n≤512 The paragraphs are independently passed through a shared BERT-based model to extract a list of context-aware paragraph representations using their respective [CLS]token embeddings, the [CLS]token of each paragraph is added with a learnable paragraph position embedding, and the result is fed into a small Transformer encoder that computes the final enforcement embedding via a pooling operation on the context-aware paragraph representations.

[0017] As an optional implementation, the specific process of constructing the legal graph includes: formalizing the hierarchical organization of the law into a tree structure composed of two types of nodes, the node types include node nodes and legal article nodes, the node nodes are title structure units representing continuous divisions in the law, and the legal article nodes represent text content units of different statutory provisions;

[0018] Initialize node features, encode the semantic information of nodes, and use the resulting embedding as the initial node feature;

[0019] Graph neural networks are used to capture the structural properties of each node and update the legal graph.

[0020] As a further qualification, using graph neural networks to capture the structural properties of each node, the specific process of updating the legal graph includes:

[0021] The graph neural network consists of a stack of neural network layers, where each layer aggregates the local neighborhood information around each node and passes the aggregated information to the next layer. The graph neural network takes the feature matrix and the adjacency matrix of the graph as input and produces a node-level output that captures the structural properties of each node.

[0022] Each graph neural network is written as a nonlinear function, and when propagating layer by layer, a nonlinear activation function is used to update the hidden state of the node.

[0023] As an optional implementation, the specific process of using a classifier to determine the correctness probability prediction result includes: using a binary classifier to classify the feature representation after dimensionality transformation to obtain a label prediction result for displaying the confidence of the model in each category.

[0024] A system for determining enforcement basis in administrative enforcement, comprising:

[0025] An information extraction module is configured to extract case description text information and enforcement basis text information cited in the document from the case file document;

[0026] The encoding module is configured to encode the case description text information and the law enforcement basis text information respectively, and map them into dense representations to obtain the case description embedding and the law enforcement basis embedding respectively;

[0027] A legal graph encoding module is configured to construct a legal graph, fuse the information in the legal graph to enrich the enforcement basis embedding, and obtain the retrieved basis;

[0028] A feature representation module is configured to extract features of the case description embedding and the enforcement basis embedding and retrieved basis, perform dimension transformation on the features, and obtain feature representation;

[0029] The classification module is configured to determine the correctness probability prediction result based on the feature representation using a classifier, compare the correctness probability prediction result with a preset probability threshold, and set a judgment label according to the comparison result.

[0030] A computer-readable storage medium is used to store computer instructions, and when the computer instructions are executed by a processor, the steps in the above method are completed.

[0031] An electronic device comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps in the above method are completed.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] The present invention first retrieves relevant law enforcement basis from the legal corpus, then classifies the case description, law enforcement basis and search results, and finally outputs a correctness judgment result label of the law enforcement process, which can effectively assist judges and other judicial workers in judging the rationality and correctness of the law enforcement basis during the law enforcement process.

[0034] The present invention learns high-quality low-dimensional embedding spaces of case descriptions and enforcement basis so that related case description-enforcement basis pairs appear closer than unrelated ones in these spaces, represents them as embeddings, and extracts features. This can effectively solve the problem of long text information of statutory provisions and case descriptions, overcome the maximum input length limitation of 512 tokens imposed by the BERT-based model, and improve processing efficiency.

[0035] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0037] Figure 1 It is a schematic diagram of the flow chart of this embodiment. DETAILED DESCRIPTION

[0038] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0039] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0040] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0041] In the absence of conflict, the embodiments in this application and the features in the embodiments may be combined with each other.

[0042] Embodiment 1

[0043] A method for determining the basis for enforcement in administrative law enforcement, such as Figure 1 As shown, the following steps are included:

[0044] Extracting case description text information and enforcement basis text information cited in the documents from case files;

[0045] Encode the case description text information and the law enforcement basis text information respectively, and map them into dense representations to obtain the case description embedding and the law enforcement basis embedding respectively;

[0046] Construct a legal graph, integrate the information in the legal graph to enrich the enforcement basis embedding, and obtain the retrieved basis;

[0047] Extract the features of the case description embedding, the enforcement basis embedding and the retrieved basis, perform dimension transformation on the features, and obtain feature representation;

[0048] Based on the feature representation, a classifier is used to determine a correctness probability prediction result, the correctness probability prediction result is compared with a preset probability threshold, and a determination label is set according to the comparison result.

[0049] In this embodiment, the process of encoding the case description text information includes: inputting the case description text information into a pre-trained BERT-based model with a weight of θ, the model maps the input sequence of n tokens from the vocabulary W to a d-dimensional real-valued token embedding, and embedding the [CLS] token representation of the last layer as the case description.

[0050] In this embodiment, in the pre-training stage of the BERT-based model, after the model is pre-trained and fine-tuned, the output of the [CLS] token can capture the semantics and information of the entire sentence or sequence.

[0051] In this embodiment, the specific process of encoding the law enforcement basis text information includes: each law enforcement basis is divided into a plurality of text paragraphs [p1, p2, ..., p m ], where paragraph p i is a token sequence of n≤512 The paragraphs are independently passed through a shared BERT-based model to extract a list of context-aware paragraph representations using their respective [CLS]token embeddings, the [CLS]token of each paragraph is added with a learnable paragraph position embedding, and the result is fed into a small Transformer encoder that computes the final enforcement embedding via a pooling operation on the context-aware paragraph representations.

[0052] In this embodiment, the specific process of constructing the legal graph includes: formalizing the hierarchical organization of the law into a tree structure composed of two types of nodes, the node types include node nodes and legal article nodes, the node nodes are title structure units representing continuous divisions in the law, and the legal article nodes represent text content units of different statutory provisions;

[0053] Initialize node features, encode the semantic information of nodes, and use the resulting embedding as the initial node feature;

[0054] Graph neural networks are used to capture the structural properties of each node and update the legal graph.

[0055] In this embodiment, the structural attributes of each node are captured by using a graph neural network, and the specific process of updating the legal graph includes:

[0056] The graph neural network consists of a stack of neural network layers, where each layer aggregates the local neighborhood information around each node and passes the aggregated information to the next layer. The graph neural network takes the feature matrix and the adjacency matrix of the graph as input and produces a node-level output that captures the structural properties of each node.

[0057] Each graph neural network is written as a nonlinear function, and when propagating layer by layer, a nonlinear activation function is used to update the hidden state of the node.

[0058] In this embodiment, the specific process of using a classifier to determine the correctness probability prediction result includes: using a binary classifier to classify the feature representation after dimensionality transformation to obtain a label prediction result for displaying the confidence of the model in each category.

[0059] Embodiment 2

[0060] A law enforcement basis determination system in administrative law enforcement includes a data acquisition module, a law enforcement basis retrieval module and a classification module, wherein:

[0061] The data acquisition module is used to obtain law enforcement files or judgment documents, extract the text information of the case description and the text information of the law enforcement basis cited in the document.

[0062] Law enforcement basis search module

[0063] In the law enforcement basis retrieval module, a graph-enhanced legal retrieval method is proposed, which uses the text semantic information in the law enforcement basis and the structural graph information in the legislation to learn and retrieve relevant laws. It is implemented using a graph-enhanced legal retriever, which mainly includes a legal retriever and a legal graph encoder.

[0064] Legal Search Engine

[0065] The basis of the legal searcher is the Bi-encoder architecture, which maps case descriptions and enforcement bases into dense representations and calculates relevance scores. The legal searcher aims to learn high-quality low-dimensional embedding spaces for case descriptions and enforcement bases, so that related case description-enforcement base pairs appear closer in these spaces than unrelated ones. The relevance score is calculated as:

[0066]

[0067] in and They represent case description embedding and law enforcement basis embedding respectively, and sim is a similarity function, such as cosine similarity or dot product similarity.

[0068] Bi-encoder is further divided into query encoder and hierarchical article encoder.

[0069] To encode the case description (query), it is input into a BERT-based model with weights θ, which maps the input sequence of n tokens from the vocabulary W to d-dimensional real-valued token embeddings, and then uses the [CLS] token representation of the last layer as the case description embedding, that is:

[0070]

[0071] After training, the output of this special token has been designed to be used as a representation for the entire sentence or sequence.

[0072] During the pre-training stage of the BERT model, it is used for classification tasks in downstream tasks, so after pre-training and fine-tuning, the output of the [CLS] token is expected to capture the semantics and information of the entire sentence or sequence.

[0073] This output already contains important information about the entire case description sequence.

[0074] To encode the law enforcement article, since the legal article may be longer than the maximum input length of the standard BERT-based encoder, G-DSR uses a hierarchical variant that can handle longer text sequences. Each law enforcement article a is first divided into smaller text paragraphs [p1, p2, …, p m ], where paragraph p i is a token sequence of n≤512 The paragraphs are then passed independently through a shared BERT-based model to extract a list of context-agnostic paragraph representations using their respective [CLS]token embeddings. Next, the chunking model adds each paragraph’s [CLS]token with a learnable paragraph position embedding and feeds the result into a small Transformer encoder to make it aware of the surrounding paragraphs. The final enforcement embedding is computed via a pooling operation on the context-aware paragraph representations, i.e.:

[0075]

[0076] 2. Legal Graph Encoder

[0077] The legal graph encoder aims to enrich the enforcement evidence embeddings provided by the trained retriever’s enforcement evidence encoder by fusing information from the legal graph. The construction of the legal graph and the graph training process are described in detail below.

[0078] Legal graph construction. In order to take advantage of the hierarchical organization of law, this method formalizes it into a tree structure consisting of two types of nodes: node nodes and legal article nodes. Node nodes are title structure units representing continuous divisions in law (i.e., legal name, title, chapter name, etc.), and legal article nodes represent text content units of different statutory provisions. Such a tree can be represented as a directed acyclic graph G = (V, E).

[0079] Node feature initialization. Nodes in V are usually associated with d-dimensional features. This method can apply the previously trained dual encoder block-based encoder to encode the semantic information of the node offline, and use the resulting embedding as the initial node feature.

[0080] Legal graph update. In order to use the graph structure to fuse the feature information of nodes, this method uses a graph neural network (GNN). Such a model consists of a stack of neural network layers, where each layer aggregates the local neighborhood information around each node (i.e., the features of the neighbors) and then passes this aggregated information to the next layer. In general, GNN takes the feature matrix X and the adjacency matrix of the graph as is the input, A i,j is the edge weight between nodes i and j, and generates node-level output This output captures the structural properties of each node.

[0081] Each GNN can be written as a nonlinear function:

[0082] H (l+1) =f(H (l) ,A) (4)

[0083] Among them, H 0 =X and H L=Z, L is the number of layers of the graph neural network. In its simplest form, the rule of layer-by-layer propagation is as follows:

[0084] f(H (l) ,A)=σ(AH (l) W (l) ) (5)

[0085] Where W (l) is the weight matrix of the linear transformation of the input of the l-th layer neural network, and σ(·) is the nonlinear activation function. Here we can use the GATv2 network, which is a variant of GAT that can learn the connection strength of adjacent nodes through a dynamic attention mechanism. Formally, GATv2 updates the hidden state of a node as follows:

[0086]

[0087] Where N(i) is the first-order neighbor set of node i, is the normalized attention coefficient, indicating the importance of the feature of node j to node i in the lth layer. The latter is calculated as:

[0088]

[0089] Where att is the attention function.

[0090] Classification Module

[0091] This module receives the results of the enforcement basis search module and the input case description and enforcement basis, and outputs the legality judgment of the enforcement basis. The specific process is as follows:

[0092] Feature representation: The preprocessed case description, enforcement basis, and retrieved basis are simultaneously input into the BERT model. Feature extraction is performed through the Transformer layer of the BERT model to generate the corresponding feature representation. This step is to convert natural language text into a form that can be understood by the machine.

[0093] Dimension transformation: Perform downstream task processing on feature representations, such as dimension transformation through a fully connected layer. The main function of a fully connected layer is to combine and transform the extracted features to generate feature representations suitable for downstream tasks.

[0094] Classification prediction: The processed feature representation is input into a binary classifier, such as logistic regression, SVM, etc., to output the probability of whether the law enforcement process is correct. Specifically, the classifier can output the probability distribution of two labels, which can show the confidence of the model in each category rather than just the final prediction result. This step is to perform the actual prediction task based on the extracted features.

[0095] Result determination: According to the set threshold, the predicted probability is converted into a binary classification result. If the probability is greater than the threshold, the law enforcement process is judged to be correct; otherwise, the law enforcement process is judged to be wrong. This step is to convert the machine's prediction results into a form that can be understood by humans. It is not a final judgment, but can be used as a reference for judicial personnel.

[0096] In this way, the structure of the entire model is an end-to-end process, from inputting the case description, the adopted law enforcement basis, to outputting the legality judgment result of the law enforcement process.

[0097] Embodiment 3

[0098] A system for determining enforcement basis in administrative enforcement, comprising:

[0099] An information extraction module is configured to extract case description text information and enforcement basis text information cited in the document from the case file document;

[0100] The encoding module is configured to encode the case description text information and the law enforcement basis text information respectively, and map them into dense representations to obtain the case description embedding and the law enforcement basis embedding respectively;

[0101] A legal graph encoding module is configured to construct a legal graph, fuse the information in the legal graph to enrich the enforcement basis embedding, and obtain the retrieved basis;

[0102] A feature representation module is configured to extract features of the case description embedding and the enforcement basis embedding and retrieved basis, perform dimension transformation on the features, and obtain feature representation;

[0103] The classification module is configured to determine the correctness probability prediction result based on the feature representation using a classifier, compare the correctness probability prediction result with a preset probability threshold, and set a judgment label according to the comparison result.

[0104] Embodiment 4

[0105] A computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the steps in the method provided in the above embodiment are completed.

[0106] Embodiment 5

[0107] An electronic device includes a memory and a processor, and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps in the method provided in the above embodiment are completed.

[0108] The above process of the above embodiment can be executed on a processor. It should be understood that the processor can be a different type of graphics processing unit GPU including RTX 3090 and other processors.

[0109] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type. And a computer-readable storage medium for storing computer instructions, which, when executed by the processor, completes the method described in Embodiment 1.

[0110] The method in the first embodiment can be directly implemented by combining hardware and software modules in the processor. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.

[0111] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0112] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0113] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0114] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0115] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principle of the present invention without creative labor shall be included in the protection scope of the present invention.

Claims

1. A method for determining the basis for administrative law enforcement. Its characteristics are: The following steps are involved: Extracting case description text information and enforcement basis text information cited in the documents from case files; Encode the case description text information and the law enforcement basis text information respectively, and map them into dense representations to obtain the case description embedding and the law enforcement basis embedding respectively; Construct a legal graph, integrate the information in the legal graph to enrich the enforcement basis embedding, and obtain the retrieved basis; Extract the features of the case description embedding, the enforcement basis embedding and the retrieved basis, perform dimension transformation on the features, and obtain feature representation; Based on the feature representation, a classifier is used to determine a correctness probability prediction result, the correctness probability prediction result is compared with a preset probability threshold, and a determination label is set according to the comparison result.

2. A method for determining enforcement basis in administrative enforcement as claimed in claim 1, Its characteristics are: The process of encoding the case description text information includes: inputting the case description text information into the weight In the pre-trained BERT-based model, the model takes the words from the vocabulary W n The input sequence of tokens is mapped to d The real-valued token is embedded and the [CLS] token representation of the last layer is embedded as the case description.

3. A method for determining enforcement basis in administrative enforcement as claimed in claim 2, Its characteristics are: In the pre-training phase of the BERT-based model, after pre-training and fine-tuning, the output of the [CLS] token can capture the semantics and information of the entire sentence or sequence.

4. A method for determining enforcement basis in administrative enforcement as claimed in claim 1, Its characteristics are: The specific process of encoding the text information of law enforcement basis includes: each law enforcement basis is divided into multiple text paragraphs , where paragraph yes n Token sequence ≤ 512 , the paragraphs are independently passed through a shared BERT-based model to extract a list of context-independent paragraph representations using their respective [CLS]token embeddings, the [CLS]token of each paragraph is added with the learnable paragraph position embedding, and the result is fed into a small Transformer encoder that computes the final enforcement embedding via a pooling operation on the context-aware paragraph representations.

5. A method for determining enforcement basis in administrative enforcement as claimed in claim 1, Its characteristics are: The specific process of constructing the legal graph includes: formalizing the hierarchical organization of the law into a tree structure composed of two types of nodes, the node types include node nodes and legal article nodes. The node nodes are the title structure units representing the continuous divisions in the law, and the legal article nodes represent the text content units of different statutory provisions; Initialize node features, encode the semantic information of nodes, and use the resulting embedding as the initial node feature; Graph neural networks are used to capture the structural properties of each node and update the legal graph.

6. A method for determining enforcement basis in administrative enforcement as claimed in claim 5, Its characteristics are: Using graph neural networks to capture the structural properties of each node, the specific process of updating the legal graph includes: The graph neural network consists of a stack of neural network layers, where each layer aggregates the local neighborhood information around each node and passes the aggregated information to the next layer. The graph neural network takes the feature matrix and the adjacency matrix of the graph as input and produces a node-level output that captures the structural properties of each node. Each graph neural network is written as a nonlinear function, and when propagating layer by layer, a nonlinear activation function is used to update the hidden state of the node.

7. A method for determining enforcement basis in administrative enforcement as claimed in claim 1, Its characteristics are: The specific process of using a classifier to determine the correctness probability prediction result includes: using a binary classifier to classify the feature representation after dimension transformation to obtain a label prediction result, which is used to show the confidence of the model in each category.

8. A system for determining the basis for administrative law enforcement. Its characteristics are: include: An information extraction module is configured to extract case description text information and enforcement basis text information cited in the document from the case file document; The encoding module is configured to encode the case description text information and the law enforcement basis text information respectively, and map them into dense representations to obtain the case description embedding and the law enforcement basis embedding respectively; A legal graph encoding module is configured to construct a legal graph, fuse the information in the legal graph to enrich the enforcement basis embedding, and obtain the retrieved basis; A feature representation module is configured to extract features of the case description embedding and the enforcement basis embedding and retrieved basis, perform dimension transformation on the features, and obtain feature representation; The classification module is configured to determine the correctness probability prediction result based on the feature representation using a classifier, compare the correctness probability prediction result with a preset probability threshold, and set a judgment label according to the comparison result.

9. A computer-readable storage medium, Its characteristics are: Used to store computer instructions, which, when executed by a processor, complete the steps of the method according to any one of claims 1 to 7.

10. An electronic device, Its characteristics are: The method comprises a memory and a processor, and computer instructions stored in the memory and executed on the processor, wherein when the computer instructions are executed by the processor, the steps in the method according to any one of claims 1 to 7 are completed.

Citation Information

Cited By

  • Food and medicine investigation and law enforcement comprehensive information system supporting multi-level user approval

    CN120563058A