Civil Case Risk Assessment Method, Apparatus, Device, Medium and Product

Abstract

The present application discloses a method, apparatus, device, medium and product for risk assessment of civil cases, which are applied to the technical field of case assessment. The method includes: obtaining a police situation text; processing the police situation text through NLP technology to obtain the similarity score between the police situation text and legal provisions, the keyword risk score and the emotional risk score; processing the police situation text through GCN technology to obtain the case node risk score corresponding to the police situation text; and performing weighted fusion on the similarity score between the police situation text and legal provisions, the keyword risk score, the emotional risk score and the case node risk score through an attention mechanism to obtain the risk assessment result of the civil case. The present application considers the correlation relationship between cases, and processes the police situation text through GCN technology to obtain the case node risk score corresponding to the police situation text, thereby improving the accuracy of the risk assessment result of civil cases.

Description

Technical Field

[0001] This application relates to the technical field of case assessment, and more particularly, to a civil case risk assessment method, device, equipment, medium and product. Background Art

[0002] In the current social environment, civil disputes occur frequently, and the number of cases is increasing continuously, posing a huge challenge to the work of public security organs. The traditional way of handling civil cases mainly relies on manual experience and intuition, lacking systematic and scientific analysis means, often resulting in low processing efficiency and frequent occurrence of misjudgments and missed judgments. Therefore, how to use advanced technical means to conduct scientific and effective risk assessment of civil cases has become an urgent problem to be solved.

[0003] The existing civil case risk assessment methods are mainly divided into the following four categories:

[0004] Rule-based method: Traditional civil case risk prediction mainly relies on manual experience, and classifies civil cases by setting a series of rules.

[0005] Statistics-based method: Such methods usually conduct statistical analysis on historical civil case data and establish a probability model to evaluate the risks of new civil cases.

[0006] Machine learning-based method: Existing research applies machine learning algorithms such as support vector machine (SVM), decision tree, and random forest to civil case risk assessment.

[0007] Deep learning-based method: Such methods use deep learning models (such as convolutional neural network, recurrent neural network, etc.) for automatic feature extraction and risk prediction of civil case texts.

[0008] However, the above methods ignore the correlation between civil cases, which may lead to low accuracy of civil case risk assessment results. Summary of the Invention

[0009] The main purpose of this application is to provide a civil case risk assessment method, device, equipment, medium and product to solve the problem of low accuracy of existing civil case risk assessment results.

[0010] To achieve the above purpose, in the first aspect, this application provides a civil case risk assessment method, including:

[0011] Obtain police situation text;

[0012] Process the police situation text through NLP technology to obtain the similarity score between the police situation text and legal provisions, keyword risk score, and sentiment risk score;

[0013] Process the police situation text using GCN technology to obtain the risk score of the case node corresponding to the police situation text;

[0014] Through the attention mechanism, the similarity score between the police situation text and legal provisions, the keyword risk score, the emotional risk score, and the case node risk score are weighted and fused to obtain the risk assessment result of civil cases.

[0015] In an embodiment, the police situation text is processed by NLP technology to obtain the similarity score between the police situation text and legal provisions, including:

[0016] Obtain the legal provisions corresponding to the police situation text;

[0017] Input the police situation text and legal provisions into the BERT similarity risk model, and output the similarity score between the police situation text and legal provisions.

[0018] In an embodiment, input the police situation text and legal provisions into the BERT similarity risk model, and output the similarity score between the police situation text and legal provisions, including:

[0019] Preprocess the police situation text and legal provisions to obtain the preprocessed police situation text and preprocessed legal provisions;

[0020] Obtain the first text feature corresponding to the preprocessed police situation text and the second text feature corresponding to the preprocessed legal provisions;

[0021] Use cosine similarity to calculate the similarity between the first text feature and the second text feature, and extract the maximum similarity as the similarity score between the police situation text and legal provisions.

[0022] In an embodiment, obtain the first text feature corresponding to the preprocessed police situation text and the second text feature corresponding to the preprocessed legal provisions, including:

[0023] Use the tokenizer of the BERT similarity risk model to tokenize and add markers to the preprocessed police situation text and preprocessed legal provisions to obtain the marked police situation text and marked legal provisions;

[0024] Use the BERT similarity risk model to vectorize the marked police situation text and marked legal provisions to obtain the first feature vector corresponding to the police situation text and the second feature vector corresponding to the legal provisions;

[0025] Extract the vectors carrying the target markers from the first feature vector and the second feature vector to obtain the first text feature and the second text feature.

[0026] In one embodiment, the police situation text is processed by NLP technology to obtain the risk score of the keywords corresponding to the police situation text, including:

[0027] Preprocess the police situation text to obtain the current police situation report;

[0028] Input the current police situation report into the TF-IDF keyword risk model, and output the risk score of the keywords corresponding to the police situation text.

[0029] In one embodiment, inputting the current police situation report into the TF-IDF keyword risk model and outputting the risk score of the keywords corresponding to the police situation text includes:

[0030] Perform word segmentation on the current police situation report and the historical police situation report to obtain the word-segmented current police situation report and the word-segmented historical police situation report;

[0031] Based on the word-segmented current police situation report and the word-segmented historical police situation report, calculate the TF value and IDF value of each keyword in the word-segmented current police situation report;

[0032] Based on the TF value and IDF value of each keyword, calculate the TF-IDF value of each keyword;

[0033] Based on the TF-IDF value of each keyword and the preset risk coefficient table, calculate the risk score of each keyword;

[0034] Add up the risk scores of each keyword to obtain the risk score of the keywords corresponding to the police situation text.

[0035] In one embodiment, based on the TF-IDF value of each keyword and the preset risk coefficient table, calculating the risk score of each keyword includes:

[0036] Obtain the preset risk coefficient table;

[0037] Select the risk coefficient matching each keyword from the preset risk coefficient table, and multiply the TF-IDF value of each keyword by the risk coefficient matching each keyword to obtain the risk score of each keyword.

[0038] In one embodiment, the police situation text is processed by NLP technology to obtain the emotional risk score of the police situation text, including:

[0039] Preprocess the police situation text to obtain the preprocessed police situation text;

[0040] Input the preprocessed police situation text into the ERNIE emotional risk model, and output the emotional risk score of the police situation text.

[0041] In one embodiment, the preprocessed police situation text is input into the ERNIE sentiment risk model, and the sentiment risk score corresponding to the police situation text is output, including:

[0042] The preprocessed police situation text is vectorized using the ERNIE sentiment risk model to obtain a third feature vector corresponding to the preprocessed police situation text;

[0043] The feature vector carrying the target marker is extracted from the third feature vector as the text feature vector;

[0044] The sentiment category probability of each text feature vector in the text feature vector is obtained;

[0045] The sentiment category probability of each text feature vector is multiplied by the corresponding risk coefficient to obtain the risk score of each text feature vector;

[0046] The risk scores of each text feature vector are added together to obtain the sentiment risk score corresponding to the police situation text.

[0047] In one embodiment, the GCN technology is used to process the police situation text to obtain the case node risk score corresponding to the police situation text, including:

[0048] A network structure diagram of civil cases is constructed, where each node in the network structure diagram identifies a civil case, and each edge represents the association relationship between civil cases;

[0049] The case node feature matrix and the adjacency matrix are obtained;

[0050] The network structure diagram, the case node feature matrix, and the adjacency matrix are input into the GCN case network model, and the case node risk score corresponding to the police situation text is output.

[0051] In one embodiment, the network structure diagram, the case node feature matrix, and the adjacency matrix are input into the GCN case network model, and the case node risk score corresponding to the police situation text is output, including:

[0052] Based on the network structure diagram, the case node feature matrix, and the adjacency matrix, the high-level features of each node are obtained;

[0053] The risk score of the high-level feature of each node is calculated;

[0054] The activation function is used to transform the risk score of the high-level feature of each node to obtain the case node risk score corresponding to the police situation text.

[0055] In one embodiment, the similarity score, keyword risk score, sentiment risk score, and case node risk score between the police situation text and the legal provisions are weighted and fused through an attention mechanism to obtain the risk assessment result of a civil case, including:

[0056] Calculate the attention scores of the similarity score, keyword risk score, sentiment risk score, and case node risk score;

[0057] Based on the attention scores of the similarity score, keyword risk score, sentiment risk score, and case node risk score, calculate the attention weights of the similarity score, keyword risk score, sentiment risk score, and case node risk score;

[0058] Multiply the similarity score, keyword risk score, sentiment risk score, and case node risk score by their respective attention weights and sum them up to obtain the risk assessment result of the civil case.

[0059] In a second aspect, an embodiment of the present application provides a civil case risk assessment device, including:

[0060] A text acquisition module for acquiring the police situation text;

[0061] A first score calculation module for processing the police situation text through NLP technology to obtain the similarity score, keyword risk score, and sentiment risk score between the police situation text and the legal provisions;

[0062] A second score calculation module for processing the police situation text through GCN technology to obtain the case node risk score corresponding to the police situation text;

[0063] A risk assessment module for weighted and fusing the similarity score, keyword risk score, sentiment risk score, and case node risk score between the police situation text and the legal provisions through an attention mechanism to obtain the risk assessment result of the civil case.

[0064] In a third aspect, an embodiment of the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of any of the above methods are implemented.

[0065] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the steps of any of the above methods are implemented.

[0066] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of any of the above methods are implemented.

[0067] The embodiments of this application provide a civil case risk assessment method, device, equipment, medium and product, including: obtaining police situation texts, then processing the police situation texts through NLP technology to obtain the similarity scores between the police situation texts and legal provisions, keyword risk scores and sentiment risk scores, and then using GCN technology to process the police situation texts to obtain the case node risk scores corresponding to the police situation texts. Finally, through the attention mechanism, the similarity scores between the police situation texts and legal provisions, keyword risk scores, sentiment risk scores and case node risk scores are weighted and fused to obtain the risk assessment result of civil cases. This application considers the association relationships between cases, uses GCN technology to process the police situation texts, obtains the case node risk scores corresponding to the police situation texts, and improves the accuracy of the risk assessment results of civil cases. In addition, this application synthesizes multiple risk scores, effectively fuses the multiple scores, and finally obtains the risk assessment result of civil cases, improving the accuracy and reliability of risk assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The drawings constituting a part of this application are used to provide a further understanding of this application, making other features, purposes and advantages of this application more obvious. The schematic embodiments drawings of this application and their descriptions are used to explain this application and do not constitute an improper limitation of this application. In the drawings:

[0069] Figure 1 is a schematic structural diagram of a civil case risk assessment device provided by an embodiment of this application;

[0070] Figure 2 is a schematic flowchart of a civil case risk assessment method provided by an embodiment of this application;

[0071] Figure 3 is a schematic structural diagram of another civil case risk assessment device provided by an embodiment of this application;

[0072] Figure 4 is a schematic diagram of a computer device provided by an embodiment of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0073] To make the purposes, technical solutions and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of this application.

[0074] In the description and claims of this application and the above-mentioned drawings, the terms "first", "second", "third", "fourth", etc. (if any) are used to distinguish similar objects and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so that the embodiments of this application described herein can be implemented in an order other than those illustrated or described herein.

[0075] In the embodiments of this application, the term "module" or "unit" refers to a computer program with a predetermined function or a part of a computer program, which works together with other related parts to achieve a predetermined goal and can be fully or partially implemented by using software, hardware (such as a processing circuit or a memory), or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be a part of an overall module or unit that includes the function of that module or unit.

[0076] It should be understood that in various embodiments of this application, the magnitudes of the sequence numbers of the various processes do not mean the order of execution, and the order of execution of the various processes should be determined by their functions and internal logics, and should not constitute any limitation to the implementation process of the embodiments of this application.

[0077] It should be understood that in this application, "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0078] It should be understood that in this application, "a plurality of" means two or more. "And / or" is merely a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally indicates that the associated objects before and after are in an "or" relationship. "Including A, B, and C" and "including A, B, C" mean that all of A, B, and C are included, "including A, B, or C" means including one of A, B, and C, and "including A, B, and / or C" means including any one or any two or all three of A, B, and C.

[0079] It should be understood that in this application, "B corresponding to A", "B corresponding to A relatively", "A corresponding to B relatively", or "B corresponding to A relatively" means that B is associated with A, and B can be determined according to A. Determining B according to A does not mean determining B only according to A, and B can also be determined according to A and / or other information. The matching of A and B means that the similarity between A and B is greater than or equal to a preset threshold.

[0080] Depending on the context, as used herein, "if" can be interpreted as "when", or "while", or "in response to determining", or "in response to detecting".

[0081] The data involved in this application can be data authorized by testers or fully authorized by all parties. The collection, dissemination, use, etc. of the data all comply with the requirements of relevant laws, regulations and standards in relevant countries and regions. The implementation manners / embodiments of this application can be combined with each other.

[0082] The technical solutions of this application will be described in detail below with specific embodiments. These several specific embodiments below can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0083] To facilitate understanding of the solution of this application, the following terms will be explained first:

[0084] NLP: Natural Language Processing is an interdisciplinary field of computer science, artificial intelligence and linguistics, dedicated to researching and developing computer systems that can recognize and interpret human language. NLP includes multiple application fields such as text processing, speech recognition, sentiment analysis, machine translation, etc. In this application, NLP technology is used to process police situation texts and extract case-related information, such as similarity scores, keyword scores and sentiment analysis scores.

[0085] BERT: BERT (Bidirectional Encoder Representations from Transformers) is a deep learning model in the field of Natural Language Processing (NLP). Its core lies in using the Transformer architecture to understand the bidirectional context of text data. BERT performs excellently in various natural language processing tasks, such as question answering systems, text classification and named entity recognition. In this application, the BERT model is used to extract text features from police situation texts.

[0086] TF-IDF: TF-IDF (Term Frequency-Inverse Document Frequency) is a statistical method widely used in information retrieval and text mining to evaluate the importance of a word for a document set or a single document in a corpus. It is a commonly used feature vectorization method. The core idea is that the higher the frequency of a word in a certain document and the lower its frequency in other documents, the more important this word is for distinguishing this document. In this application, TF-IDF is used for keyword analysis, extracting important keywords from police situation texts, and calculating keyword scores.

[0087] ERNIE (Enhanced Representation through kNowledge Integration) is a pre-trained language model proposed by Baidu, aiming to improve the quality of language representation through knowledge enhancement. The core idea of the ERNIE model is to integrate lexical, syntactic, and knowledge information during the pre-training process, enabling the model to better understand language. In this application, the ERNIE model is used for sentiment analysis of police situation texts, extracting sentiment features, and calculating sentiment analysis scores.

[0088] GCN: Graph Convolutional Network is a neural network for processing graph-structured data. It extracts information from the graph structure and node features through convolutional operations to achieve aggregation and update of node features. GCN is commonly used in fields such as social network analysis, recommendation systems, and chemical molecular structure analysis. In this application, GCN is used to construct a relationship graph between cases and calculate the risk score of each case through the graph convolutional network.

[0089] Next, the solution of this application will be described through specific embodiments in conjunction with the accompanying drawings.

[0090] Please refer to Figure 1 , Figure 1 which is a schematic diagram of a civil case risk assessment device provided by an embodiment of this application, including:

[0091] NLP module, GCN module, and attention mechanism module.

[0092] Among them, the NLP module is responsible for preprocessing and feature extraction of police situation texts (police situation information), including calculating the similarity score between cases and criminal law through the BERT similarity risk model, obtaining risk keyword scores through the TF-IDF keyword risk model, and performing sentiment analysis through the ERNIE sentiment risk model to obtain risk sentiment scores. These features provide rich text features for subsequent risk prediction.

[0093] The GCN module treats cases as network nodes, where each node represents an independent case, and the edges between nodes represent various correlations between cases, such as common involved persons, similar case types, proximate geographical locations, or temporal correlations, etc. This network model not only captures the inherent characteristics of cases but also deeply analyzes the complex interrelationships and network structures between cases, thereby revealing the potential patterns of case risk propagation and influence. Through in-depth feature learning and information aggregation of the case network by GCN, the model can identify and quantify the mutual influence between cases. For example, the high risk of one case may be transmitted to other cases through the correlations in the network, thus affecting the overall risk assessment. This method enables risk prediction not to be limited to analyzing a single case in isolation but to be able to consider the dynamic changes and interdependencies of the set of cases from a global perspective. Further, the multi-level convolutional operation of the GCN model can capture case relationships at different levels, from direct connections to indirect influences, providing a more abundant and detailed perspective for risk prediction. By capturing the propagation of risk in the network, a risk score is calculated for each case node, fully considering the mutual influence between cases, thereby providing a global risk assessment perspective.

[0094] The attention mechanism module fuses the similarity score, keyword risk score, and sentiment risk score extracted by the NLP module and the case node risk score obtained by the GCN module. By learning the importance of different features for case risk prediction, this module assigns corresponding weights to each feature, enabling the model to pay more attention to the features that contribute more to risk prediction and making the final risk prediction score more comprehensive and accurate.

[0095] The above device ensures that when dealing with civil case risk prediction, it can not only comprehensively utilize text information but also fully consider the complex relationships between cases, and finally provides an efficient and accurate risk assessment method.

[0096] Please refer to Figure 2 , Figure 2 which is a schematic flowchart of a civil case risk assessment method provided by an embodiment of this application. It includes the following steps:

[0097] Step S201: Obtain police situation text;

[0098] Step S202: Process the police situation text through NLP technology to obtain the similarity score between the police situation text and legal provisions, keyword risk score, and sentiment risk score.

[0099] For processing police situation texts through NLP technology to obtain the similarity scores between police situation texts and legal provisions, it is necessary to first obtain the legal provisions corresponding to the police situation texts, and then input the police situation texts and legal provisions into the BERT similarity risk model to output the similarity scores between the police situation texts and legal provisions.

[0100] Among them, inputting the police situation text and legal provisions into the BERT similarity risk model to output the similarity scores between the police situation text and legal provisions includes: preprocessing the police situation text and legal provisions to obtain the preprocessed police situation text and preprocessed legal provisions; obtaining the first text feature corresponding to the preprocessed police situation text and the second text feature corresponding to the preprocessed legal provisions; calculating the similarity between the first text feature and the second text feature using cosine similarity, and extracting the maximum similarity as the similarity score between the police situation text and legal provisions.

[0101] Among them, obtaining the first text feature corresponding to the preprocessed police situation text and the second text feature corresponding to the preprocessed legal provisions includes: using the tokenizer of the BERT similarity risk model to tokenize and add markers to the preprocessed police situation text and preprocessed legal provisions to obtain the marked police situation text and marked legal provisions; using the BERT similarity risk model to vectorize the marked police situation text and marked legal provisions to obtain the first feature vector corresponding to the police situation text and the second feature vector corresponding to the legal provisions; extracting the vectors carrying the target markers from the first feature vector and the second feature vector to obtain the first text feature and the second text feature.

[0102] Exemplarily, let the input be the police situation text , and the legal provision be , and the output be the similarity score .

[0103] First, preprocess the text, including removing stop words, punctuation marks, etc. The preprocessed police situation text and legal provisions are as follows:

[0104]

[0105]

[0106] Then, use the BERT similarity risk model to represent the above text as feature vectors

[0107] Tokenization: Use the tokenizer of the BERT similarity risk model to tokenize the above text and add special markers, namely [CLS] and [SEP]:

[0108]

[0109]

[0110] Embedding: Obtain the feature vector of the text:

[0111]

[0112]

[0113] Pooling: Extract the vector corresponding to the [CLS] token as the text feature:

[0114]

[0115]

[0116] Secondly, use cosine similarity to calculate the similarity between the police situation text and the criminal law article text:

[0117]

[0118] Among them, represents the dot product, represents the norm of the vector.

[0119] Finally, calculate the similarity between each police situation text and all criminal law article texts, and take the highest value as the similarity score:

[0120]

[0121] Among them, represents the th feature vector of the police situation text, represents the th feature vector of the criminal law article text.

[0122] This application uses the BERT similarity risk model for text feature extraction, which can effectively capture the deep semantics of the police situation text, thereby improving the accuracy of similarity calculation. This similarity score is a key factor in the comprehensive risk assessment and plays a crucial role in the final risk prediction. In addition, in addition to the BERT similarity risk model, there are other pre-trained language models such as RoBERTa, XLNet, ALBERT, etc., which can also be used for this purpose.

[0123] For processing the police situation text through NLP technology to obtain the risk score of the keywords corresponding to the police situation text, it is necessary to preprocess the police situation text first to obtain the current police situation report, and then input the current police situation report into the TF-IDF keyword risk model to output the risk score of the keywords corresponding to the police situation text.

[0124] Among them, the current police situation report is input into the TF-IDF keyword risk model, and the keyword risk score corresponding to the police situation text is output, including: performing word segmentation on the current police situation report and the historical police situation report to obtain the word-segmented current police situation report and the word-segmented historical police situation report; calculating the TF value and IDF value of each keyword in the word-segmented current police situation report based on the word-segmented current police situation report and the word-segmented historical police situation report; calculating the TF-IDF value of each keyword based on the TF value and IDF value of each keyword; calculating the risk score of each keyword based on the TF-IDF value of each keyword and a preset risk coefficient table; adding up the risk scores of each keyword to obtain the keyword risk score corresponding to the police situation text.

[0125] Among them, calculating the risk score of each keyword based on the TF-IDF value of each keyword and a preset risk coefficient table includes: obtaining the preset risk coefficient table; selecting the risk coefficient matching each keyword from the preset risk coefficient table, and multiplying the TF-IDF value of each keyword by the risk coefficient matching each keyword to obtain the risk score of each keyword.

[0126] Exemplarily, let the input be the current police situation report , and the output be the keyword risk score .

[0127] First, perform data preprocessing on the current police situation report and the historical police situation report, and use a word segmentation tool for word segmentation. Common word segmentation tools include jieba, NLTK, and spacy. In this method, the jieba word segmentation tool is used.

[0128] Then, calculate the frequency of occurrence (TF, term frequency value) of each word in the document of the current police situation report :

[0129]

[0130] Among them, is the term (keyword) in the document (historical police situation report), is the total number of all terms in the document .

[0131] Secondly, calculate the frequency of occurrence (IDF, inverse document frequency) of each word in the entire police situation data set (including the current police situation report and the historical police situation report):

[0132]

[0133] Among them, N is the total number of documents, is the number of documents containing the term .

[0134] Furthermore, multiply the term frequency matrix by the IDF value to obtain the TF-IDF value of each term in the police situation report:

[0135]

[0136] Finally, match the keywords according to the predefined keyword and its risk coefficient table (preset risk coefficient table) and calculate the risk score. These risk coefficients can be set manually according to expert knowledge or obtained through statistical analysis of historical data. For example:

[0137] "Fire": 0.8

[0138] "Production accident": 0.9

[0139] "Traffic accident": 0.5

[0140]

[0141] Among them, represents the risk coefficient of the term .

[0142] Through the above process, the keyword risk score of the police situation text can be obtained. This method can more dynamically reflect the specific content and characteristics of each police situation text without relying on a pre-established keyword library.

[0143] For processing the police situation text through NLP technology to obtain the corresponding emotional risk score of the police situation text, including: first preprocessing the police situation text to obtain the preprocessed police situation text, and then inputting the preprocessed police situation text into the ERNIE emotional (emotional) risk model to output the corresponding emotional risk score of the police situation text.

[0144] Among them, inputting the preprocessed police situation text into the ERNIE emotional risk model to output the corresponding emotional risk score of the police situation text, including: vectorizing the preprocessed police situation text using the ERNIE emotional risk model to obtain the third feature vector corresponding to the preprocessed police situation text; extracting the feature vector carrying the target marker from the third feature vector as the text feature vector; obtaining the emotional category probability of each text feature vector in the text feature vector; multiplying the emotional category probability of each text feature vector by the corresponding risk coefficient to obtain the risk score of each text feature vector; adding up the risk scores of each text feature vector to obtain the corresponding emotional risk score of the police situation text.

[0145] Exemplarily, let the input be the police situation text , and the output be the emotional risk score .

[0146] First, preprocess the police situation text, such as removing stop words, punctuation marks, etc.:

[0147]

[0148] Then, use the ERNIE sentiment risk model for text feature representation:

[0149] Tokenization: Segment the text and add special tokens.

[0150]

[0151] Embedding: Obtain the feature vector of the text

[0152]

[0153] Secondly, conduct sentiment classification:

[0154] Feature vector extraction: Extract the vector corresponding to the [CLS] token as the text feature.

[0155]

[0156] Fully connected layer: Convert the feature vector into a sentiment score.

[0157]

[0158] Among them, is a constant.

[0159] Activation function: Convert the logits into sentiment class probabilities.

[0160]

[0161] Furthermore, set risk coefficients for different sentiment classes: The setting of risk coefficients is based on in-depth analysis of historical case data and expert knowledge. By studying the correlation between different sentiment tendencies and case risks, a coefficient reflecting its risk level can be assigned to each sentiment class. For example, positive sentiment may be associated with lower risk, so a lower risk coefficient (e.g., 0.2) is assigned, while negative sentiment or anxiety / panic may be associated with higher risk, so a higher risk coefficient (e.g., 0.8 or 1.0) is assigned. These coefficients are not fixed and can be adjusted according to new data and expert feedback. We recommend regularly reviewing and updating these coefficients to ensure the prediction accuracy and adaptability of the model.

[0162] Positive: 0.2

[0163] Neutral: 0.5

[0164] Negative: 0.8

[0165] Anxiety / panic: 1.0

[0166] Finally, allocate risk coefficients according to the sentiment analysis results and calculate the sentiment risk scores.

[0167]

[0168] Among them, represents the sentiment category probability of the th text feature vector, represents the risk coefficient matched by the th text feature vector.

[0169] Step S203: Process the police situation text using the GCN technology to obtain the risk score of the case node corresponding to the police situation text.

[0170] For processing the police situation text using the GCN technology to obtain the risk score of the case node corresponding to the police situation text, first construct a network structure diagram of civil cases. Among them, each node in the network structure diagram identifies a civil case, and each edge represents the association relationship between civil cases. Then, obtain the case node feature matrix and the adjacency matrix. Next, input the network structure diagram, the case node feature matrix, and the adjacency matrix into the GCN case network model to output the risk score of the case node corresponding to the police situation text.

[0171] Among them, inputting the network structure diagram, the case node feature matrix, and the adjacency matrix into the GCN case network model to output the risk score of the case node corresponding to the police situation text includes: obtaining the high-level features of each node based on the network structure diagram, the case node feature matrix, and the adjacency matrix; calculating the risk scores of the high-level features of each node; using an activation function to transform the risk scores of the high-level features of each node to obtain the risk score of the case node corresponding to the police situation text.

[0172] Exemplarily, assume the input is the network structure diagram of civil cases, the node feature matrix X, and the adjacency matrix A, and the output is the risk score of the case node .

[0173] First, design the case node features. The features of each case node include: police situation information such as case content, case type, alarm time, and alarm location. Taken together, the node feature matrix X is expressed as:

[0174]

[0175] Among them, each feature vector of the case node includes police situation information such as case content, case type, alarm time, and alarm location.

[0176] Then, for the type design of the edges between cases, the association relationships between cases can be constructed in various ways:

[0177] Alarm reporter association: If two cases involve the same alarm reporter, an edge is established between the two cases.

[0178] Involved person association: If two cases involve the same involved person, an edge is established between the two cases.

[0179] Geographical location association: If two cases occur in the same or adjacent geographical locations, an edge is established between the two cases.

[0180] Case type association: If two cases belong to the same type, an edge is established between the two cases.

[0181] Time association: If two cases occur within a close time range, an edge is established between the two cases.

[0182] Secondly, use the graph convolutional layer to extract features from the graph structure, namely the adjacency matrix:

[0183]

[0184] Among them, represents that there is an edge between case and case , otherwise .

[0185] Furthermore, use the graph convolutional layer to extract features from the graph structure:

[0186]

[0187] Where:

[0188] represents the node feature matrix of the -th layer, ;

[0189] is the normalized version of the adjacency matrix , defined by the following formula, where is the degree matrix:

[0190]

[0191] is the weight matrix of the -th layer,

[0192] Finally, risk score calculation:

[0193] Calculate the risk score for the high-level feature representation of each node:

[0194]

[0195] where, represents the node feature matrix of the th layer, represents the weight coefficient, represents a constant.

[0196] Use an activation function to convert the logits into the risk score of the case node:

[0197]

[0198] Step S204: Through the attention mechanism, weightedly fuse the similarity score between the police situation text and the legal provisions, the keyword risk score, the sentiment risk score, and the case node risk score to obtain the risk assessment result of the civil case.

[0199] Regarding the weighted fusion of the similarity score between the police situation text and the legal provisions, the keyword risk score, the sentiment risk score, and the case node risk score through the attention mechanism to obtain the risk assessment result of the civil case, it is necessary to first calculate the attention scores of the similarity score, the keyword risk score, the sentiment risk score, and the case node risk score, and then calculate the attention weights of the similarity score, the keyword risk score, the sentiment risk score, and the case node risk score based on the attention scores of the similarity score, the keyword risk score, the sentiment risk score, and the case node risk score. Then multiply the similarity score, the keyword risk score, the sentiment risk score, and the case node risk score by their respective attention weights and sum them up to obtain the risk assessment result of the civil case.

[0200] Exemplarily, first calculate the attention scores:

[0201] Apply a linear layer to each score to transform it into a common space, and then calculate their attention scores:

[0202]

[0203]

[0204]

[0205]

[0206] where, , , and is a constant and can be set according to specific circumstances. W1, W2, W3, and W4 are the weight matrices of each linear layer, and tanh is a constant that can be set according to specific circumstances.

[0207] Then, calculate the attention weights

[0208] Pass these scores through a softmax layer to obtain the weights of each score:

[0209]

[0210] The specific calculation method is:

[0211]

[0212] Finally, fuse the features, that is, multiply the original scores by their corresponding attention weights and then sum to obtain the final scores:

[0213]

[0214] This application uses the attention mechanism for score fusion, which can dynamically adjust the weights of features according to the context of a specific case, enabling the model to pay more attention to the features that contribute more to risk prediction, thereby improving the accuracy and comprehensiveness of the final risk assessment.

[0215] The embodiments of this application provide a civil case risk assessment method, including: obtaining the police situation text, then processing the police situation text through NLP technology to obtain the similarity score between the police situation text and legal provisions, the keyword risk score, and the emotional risk score, then using GCN technology to process the police situation text to obtain the case node risk score corresponding to the police situation text, and finally using the attention mechanism to perform weighted fusion on the similarity score between the police situation text and legal provisions, the keyword risk score, the emotional risk score, and the case node risk score to obtain the risk assessment result of the civil case. This application considers the correlation between cases, uses GCN technology to process the police situation text to obtain the case node risk score corresponding to the police situation text, and improves the accuracy of the civil case risk assessment result. In addition, this application synthesizes multiple risk scores, effectively fuses multiple scores, and finally obtains the civil case risk assessment result, improving the accuracy and reliability of the risk assessment.

[0216] It should be understood that the magnitudes of the sequence numbers of the steps in the above embodiments do not mean the order of execution. The order of execution of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of this application.

[0217] The following are the device embodiments of this application. For the details not described in detail, reference can be made to the corresponding method embodiments above.

[0218] Figure 3 The structure diagram of a civil case risk assessment device provided by an embodiment of the present application is shown. For ease of explanation, only the parts related to the embodiment of the present application are shown. A civil case risk assessment device includes a text acquisition module 301, a first score calculation module 302, a second score calculation module 303, and a risk assessment module 304, specifically as follows:

[0219] The text acquisition module 301 is used to acquire police situation texts;

[0220] The first score calculation module 302 is used to process the police situation texts through NLP technology to obtain the similarity score between the police situation texts and legal provisions, the keyword risk score, and the sentiment risk score;

[0221] The second score calculation module 303 is used to process the police situation texts by using GCN technology to obtain the case node risk score corresponding to the police situation texts;

[0222] The risk assessment module 304 is used to perform weighted fusion on the similarity score between the police situation texts and legal provisions, the keyword risk score, the sentiment risk score, and the case node risk score through an attention mechanism to obtain the risk assessment result of the civil case.

[0223] In an embodiment, the first score calculation module 302 is further used to acquire the legal provisions corresponding to the police situation texts;

[0224] Input the police situation texts and legal provisions into the BERT similarity risk model, and output the similarity score between the police situation texts and legal provisions.

[0225] In an embodiment, the first score calculation module 302 is further used to preprocess the police situation texts and legal provisions to obtain the preprocessed police situation texts and preprocessed legal provisions;

[0226] Obtain the first text features corresponding to the preprocessed police situation texts and the second text features corresponding to the preprocessed legal provisions;

[0227] Use cosine similarity to calculate the similarity between the first text features and the second text features, and extract the maximum similarity as the similarity score between the police situation texts and legal provisions.

[0228] In an embodiment, the first score calculation module 302 is further used to tokenize and add markers to the preprocessed police situation texts and preprocessed legal provisions by using the tokenizer of the BERT similarity risk model to obtain the tokenized police situation texts and tokenized legal provisions;

[0229] Using the BERT similarity risk model, the labeled police situation text and the labeled legal provisions are vectorized to obtain the first feature vector corresponding to the police situation text and the second feature vector corresponding to the legal provisions;

[0230] Extract the vectors carrying the target labels in the first feature vector and the second feature vector to obtain the first text feature and the second text feature.

[0231] In one embodiment, the first score calculation module 302 is further configured to preprocess the police situation text to obtain the current police situation report;

[0232] Input the current police situation report into the TF-IDF keyword risk model, and output the keyword risk score corresponding to the police situation text.

[0233] In one embodiment, the first score calculation module 302 is further configured to perform word segmentation on the current police situation report and the historical police situation reports to obtain the word-segmented current police situation report and the word-segmented historical police situation reports;

[0234] Based on the word-segmented current police situation report and the word-segmented historical police situation reports, calculate the TF value and IDF value of each keyword in the word-segmented current police situation report;

[0235] Based on the TF value and IDF value of each keyword, calculate the TF-IDF value of each keyword;

[0236] Based on the TF-IDF value of each keyword and the preset risk coefficient table, calculate the risk score of each keyword;

[0237] Add up the risk scores of each keyword to obtain the keyword risk score corresponding to the police situation text.

[0238] In one embodiment, the first score calculation module 302 is further configured to obtain the preset risk coefficient table;

[0239] Select the risk coefficient matching each keyword from the preset risk coefficient table, and multiply the TF-IDF value of each keyword by the risk coefficient matching each keyword to obtain the risk score of each keyword.

[0240] In one embodiment, the first score calculation module 302 is further configured to preprocess the police situation text to obtain the preprocessed police situation text;

[0241] Input the preprocessed police situation text into the ERNIE sentiment risk model, and output the sentiment risk score corresponding to the police situation text.

[0242] In one embodiment, the first score calculation module 302 is further configured to vectorize the preprocessed police situation text by using the ERNIE sentiment risk model to obtain a third feature vector corresponding to the preprocessed police situation text;

[0243] Extract the feature vector carrying the target label from the third feature vector as the text feature vector;

[0244] Obtain the sentiment category probability of each text feature vector in the text feature vector;

[0245] Multiply the sentiment category probability of each text feature vector by the corresponding risk coefficient to obtain the risk score of each text feature vector;

[0246] Add up the risk scores of each text feature vector to obtain the sentiment risk score corresponding to the police situation text.

[0247] In one embodiment, the second score calculation module 303 is further configured to construct a network structure diagram of civil cases, where each node in the network structure diagram identifies a civil case, and each edge represents the association relationship between civil cases;

[0248] Obtain the case node feature matrix and the adjacency matrix;

[0249] Input the network structure diagram, the case node feature matrix, and the adjacency matrix into the GCN case network model, and output the case node risk score corresponding to the police situation text.

[0250] In one embodiment, the second score calculation module 303 is further configured to obtain the high-level features of each node based on the network structure diagram, the case node feature matrix, and the adjacency matrix;

[0251] Calculate the risk score of the high-level features of each node;

[0252] Use the activation function to transform the risk score of the high-level features of each node to obtain the case node risk score corresponding to the police situation text.

[0253] In one embodiment, the risk assessment module 304 is further configured to calculate the attention scores of the similarity score, the keyword risk score, the sentiment risk score, and the case node risk score;

[0254] Based on the attention scores of the similarity score, the keyword risk score, the sentiment risk score, and the case node risk score, calculate the attention weights of the similarity score, the keyword risk score, the sentiment risk score, and the case node risk score;

[0255] Multiply the similarity score, keyword risk score, sentiment risk score, and case node risk score by their respective attention weights and sum them up to obtain the risk assessment result of the civil case.

[0256] This application Figure 4 provides a schematic diagram of a computer device. As Figure 4 shown, the computer device 4 of this embodiment includes: a processor 401, a memory 402, and a computer program 403 stored in the memory 402 and executable on the processor 401. When the processor 401 executes the computer program 403, it implements the steps in the above-mentioned embodiments of the civil case risk assessment method, such as Figure 2 the steps 201 to 204 shown. Alternatively, when the processor 401 executes the computer program 403, it implements the functions of each module / unit in the above-mentioned embodiments of the civil case risk assessment device, such as Figure 3 the functions of the modules / units 301 to 304 shown.

[0257] This application also provides a readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, it is used to implement the civil case risk assessment method provided by the above various implementation manners.

[0258] Among them, the readable storage medium can be a computer storage medium or a communication medium. The communication medium includes any medium that facilitates the transmission of a computer program from one place to another. The computer storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer. For example, the readable storage medium is coupled to the processor, so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application-specific integrated circuit (ASIC). In addition, the ASIC can be located in the user equipment. Of course, the processor and the readable storage medium can also exist as discrete components in the communication device. The readable storage medium can be a read-only memory (ROM), a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0259] This application also provides a computer program product, which includes execution instructions stored in a readable storage medium. At least one processor of the device can read the execution instructions from the readable storage medium, and at least one processor executes the execution instructions to enable the device to implement the civil case risk assessment method provided by the above various implementation manners.

[0260] In an embodiment of the above device, it should be understood that the processor may be a central processing unit (CPU), or may also be other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in conjunction with the present application may be directly embodied as being executed and completed by a hardware processor, or may be executed and completed by a combination of hardware and software modules in the processor.

[0261] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A civil case risk assessment method, characterized in that: include: Get the police information text; The police text is processed by NLP technology to obtain a similarity score between the police text and the legal text, a keyword risk score corresponding to the police text, and an emotional risk score corresponding to the police text; The police text is processed using GCN technology to obtain the case node risk score corresponding to the police text; The similarity score between the police text and the legal text, the keyword risk score, the emotional risk score and the case node risk score are weighted and integrated through the attention mechanism to obtain the risk assessment result of the civil case; The method of processing the warning text by using NLP technology to obtain a keyword risk score corresponding to the warning text includes: Preprocessing the alarm text to obtain a current alarm report; Input the current police report into the TF-IDF keyword risk model, and output the keyword risk score corresponding to the police text; The step of inputting the current police report into the TF-IDF keyword risk model and outputting the keyword risk score corresponding to the police text includes: Performing word segmentation processing on the current police situation report and the historical police situation report to obtain a word-segmented current police situation report and a word-segmented historical police situation report; Based on the current police situation report after word segmentation and the historical police situation reports after word segmentation, calculating the TF value and IDF value of each keyword in the current police situation report after word segmentation; Calculate the TF-IDF value of each keyword based on the TF value and IDF value of each keyword; Calculate the risk score of each keyword based on the TF-IDF value of each keyword and a preset risk coefficient table; Add the risk scores of each keyword to obtain the risk score of the keyword corresponding to the warning text; The step of calculating the risk score of each keyword based on the TF-IDF value of each keyword and a preset risk coefficient table includes: Obtain the preset risk factor table; Selecting a risk coefficient matching each keyword from the preset risk coefficient table, and multiplying the TF-IDF value of each keyword by the risk coefficient matching each keyword to obtain a risk score for each keyword; The method of processing the warning text by using NLP technology to obtain the emotional risk score corresponding to the warning text includes: Preprocessing the warning text to obtain a preprocessed warning text; Input the preprocessed warning text into the ERNIE emotional risk model, and output the emotional risk score corresponding to the warning text; The step of inputting the pre-processed warning text into the ERNIE emotional risk model and outputting the emotional risk score corresponding to the warning text includes: Vectorizing the preprocessed warning text using the ERNIE emotional risk model to obtain a third feature vector corresponding to the preprocessed warning text; Extracting a feature vector carrying a target tag from the third feature vector as a text feature vector; Obtaining the sentiment category probability of each text feature vector in the text feature vector; Multiplying the sentiment category probability of each text feature vector by the corresponding risk coefficient to obtain a risk score for each text feature vector; Adding the risk scores of each text feature vector to obtain the emotional risk score corresponding to the warning text; The GCN technology is used to process the police text to obtain the case node risk score corresponding to the police text, including: Constructing a network structure diagram of civil cases, wherein each node in the network structure diagram identifies a civil case, and each edge represents an association relationship between civil cases; Obtain case node feature matrix and adjacency matrix; Input the network structure diagram, the case node feature matrix and the adjacency matrix into the GCN case network model, and output the case node risk score corresponding to the police situation text; The step of inputting the network structure diagram, the case node feature matrix and the adjacency matrix into the GCN case network model and outputting the case node risk score corresponding to the police situation text includes: Based on the network structure diagram, the case node feature matrix and the adjacency matrix, obtaining high-level features of each node; Calculating a risk score for the high-level features of each node; An activation function is used to transform the risk score of the high-level features of each node to obtain the case node risk score corresponding to the police text.

2. The civil case risk assessment method according to claim 1, characterized in that: The process of processing the police text by using NLP technology to obtain a similarity score between the police text and the legal provisions includes: Obtain the legal provisions corresponding to the police text; The police text and the legal text are input into the BERT similarity risk model, and the similarity score between the police text and the legal text is output.

3. The civil case risk assessment method according to claim 2, characterized in that: The step of inputting the police text and the legal text into the BERT similarity risk model and outputting a similarity score between the police text and the legal text includes: Preprocessing the police situation text and the legal provisions to obtain a preprocessed police situation text and a preprocessed legal provision; Acquire a first text feature corresponding to the preprocessed police text and a second text feature corresponding to the preprocessed legal text; The similarity between the first text feature and the second text feature is calculated using cosine similarity, and the maximum similarity is extracted as the similarity score between the police text and the legal provision.

4. The civil case risk assessment method according to claim 3, characterized in that: The obtaining of the first text feature corresponding to the preprocessed police text and the second text feature corresponding to the preprocessed legal text includes: Using a word segmenter of a BERT similarity risk model, the preprocessed police text and the preprocessed legal text are word segmented and marked to obtain a marked police text and a marked legal text; The marked police text and the marked legal provisions are vectorized by using the BERT similarity risk model to obtain a first feature vector corresponding to the police text and a second feature vector corresponding to the legal provisions; Vectors carrying target tags are extracted from the first feature vector and the second feature vector to obtain the first text feature and the second text feature.

5. The civil case risk assessment method according to claim 1, characterized in that: The attention mechanism is used to weight the similarity score between the police text and the legal text, the keyword risk score, the emotional risk score and the case node risk score to obtain the risk assessment result of the civil case, including: Calculating an attention score of the similarity score, the keyword risk score, the sentiment risk score, and the case node risk score; Calculate the attention weights of the similarity score, the keyword risk score, the emotion risk score, and the case node risk score based on the attention scores of the similarity score, the keyword risk score, the emotion risk score, and the case node risk score; The similarity score, the keyword risk score, the emotional risk score and the case node risk score are respectively multiplied by their respective attention weights and summed to obtain a risk assessment result of the civil case.

6. A civil case risk assessment device, characterized in that: include: A text acquisition module is used to obtain the warning text; A first score calculation module is used to process the police text by using NLP technology to obtain a similarity score between the police text and the legal text, a keyword risk score corresponding to the police text, and an emotional risk score corresponding to the police text; The second score calculation module is used to process the police text using the GCN technology to obtain the case node risk score corresponding to the police text; A risk assessment module is used to perform weighted fusion of the similarity scores between the police text and the legal text, the keyword risk score, the sentiment risk score and the case node risk score through an attention mechanism to obtain a risk assessment result of the civil case; The method of processing the warning text by using NLP technology to obtain a keyword risk score corresponding to the warning text includes: Preprocessing the alarm text to obtain a current alarm report; Input the current police report into the TF-IDF keyword risk model, and output the keyword risk score corresponding to the police text; The step of inputting the current police report into the TF-IDF keyword risk model and outputting the keyword risk score corresponding to the police text includes: Performing word segmentation processing on the current police situation report and the historical police situation report to obtain a word-segmented current police situation report and a word-segmented historical police situation report; Based on the current police situation report after word segmentation and the historical police situation reports after word segmentation, calculating the TF value and IDF value of each keyword in the current police situation report after word segmentation; Calculate the TF-IDF value of each keyword based on the TF value and IDF value of each keyword; Calculate the risk score of each keyword based on the TF-IDF value of each keyword and a preset risk coefficient table; Add the risk scores of each keyword to obtain the risk score of the keyword corresponding to the warning text; The step of calculating the risk score of each keyword based on the TF-IDF value of each keyword and a preset risk coefficient table includes: Obtain the preset risk factor table; Selecting a risk coefficient matching each keyword from the preset risk coefficient table, and multiplying the TF-IDF value of each keyword by the risk coefficient matching each keyword to obtain a risk score for each keyword; The method of processing the warning text by using NLP technology to obtain the emotional risk score corresponding to the warning text includes: Preprocessing the warning text to obtain a preprocessed warning text; Input the preprocessed warning text into the ERNIE emotional risk model, and output the emotional risk score corresponding to the warning text; The step of inputting the pre-processed warning text into the ERNIE emotional risk model and outputting the emotional risk score corresponding to the warning text includes: Vectorizing the preprocessed warning text using the ERNIE emotional risk model to obtain a third feature vector corresponding to the preprocessed warning text; Extracting a feature vector carrying a target tag from the third feature vector as a text feature vector; Obtaining the sentiment category probability of each text feature vector in the text feature vector; Multiplying the sentiment category probability of each text feature vector by the corresponding risk coefficient to obtain a risk score for each text feature vector; Adding the risk scores of each text feature vector to obtain the emotional risk score corresponding to the warning text; The GCN technology is used to process the police text to obtain the case node risk score corresponding to the police text, including: Constructing a network structure diagram of civil cases, wherein each node in the network structure diagram identifies a civil case, and each edge represents an association relationship between civil cases; Obtain case node feature matrix and adjacency matrix; Input the network structure diagram, the case node feature matrix and the adjacency matrix into the GCN case network model, and output the case node risk score corresponding to the police situation text; The step of inputting the network structure diagram, the case node feature matrix and the adjacency matrix into the GCN case network model and outputting the case node risk score corresponding to the police situation text includes: Based on the network structure diagram, the case node feature matrix and the adjacency matrix, obtaining high-level features of each node; Calculating a risk score for the high-level features of each node; An activation function is used to transform the risk score of the high-level features of each node to obtain the case node risk score corresponding to the police text.

7. A computer device, characterized in that: comprising a memory, and one or more processors communicatively connected to the memory; The memory stores instructions that can be executed by the one or more processors, and the instructions are executed by the one or more processors to enable the one or more processors to implement the civil case risk assessment method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: It includes a program or instruction, which, when executed on a computer, implements the civil case risk assessment method as described in any one of claims 1 to 5.

9. A computer program product, characterized in that It includes a computer program, which, when executed by a processor, implements the civil case risk assessment method as described in any one of claims 1 to 5.

Images