Case file auditing method and device, computer device, storage medium and program product

Abstract

This application relates to a case file review method, apparatus, computer equipment, computer-readable storage medium, and computer program product. The method is executed by a review model and includes: acquiring a case file to be reviewed; extracting case file keywords from the case file; determining the review type of the case file based on the case file keywords; searching for review prompt words corresponding to the case file in a prompt word rule base according to the review type and case file keywords, and combining the review prompt words corresponding to the case file to be reviewed to obtain a review prompt word chain; and reviewing the content in the case file to be reviewed sequentially according to the review prompt word chain to obtain a review result. This method enables efficient and accurate case file review.

Description

Technical Field

[0001] This application relates to the field of case file review technology, and in particular to a case file review method, apparatus, computer equipment, computer-readable storage medium, and computer program product. Background Technology

[0002] After investigating, collecting evidence, and approving procedures for illegal or irregular activities, all case files and documents involved in the procedures need to be integrated to create a complete case file. Then, a review body will assess the legality, compliance, completeness, and logical coherence of the case file.

[0003] Traditional techniques primarily rely on manual review for analysis. However, since case files are a compilation of documents, they involve cross-disciplinary knowledge and are prone to subjective biases, resulting in inefficient and inaccurate evaluation standards. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can efficiently and accurately review case files in response to the above-mentioned technical problems.

[0005] Firstly, this application provides a case file review method, executed by a review model, the method comprising:

[0006] Obtain case files to be reviewed, and extract case file keywords from the case files to be reviewed;

[0007] The review type of the case file to be reviewed is determined based on the keywords in the case file.

[0008] In the prompt word rule base, the prompt words corresponding to the case file to be reviewed are found according to the review type and case file keywords, and the prompt words corresponding to the case file to be reviewed are combined to obtain the review prompt word chain;

[0009] Based on the aforementioned audit prompt keyword chain, the contents of the case file to be audited are audited sequentially to obtain the audit results.

[0010] In one embodiment, determining the review type of the case file to be reviewed based on the case file keywords includes:

[0011] Based on the aforementioned case file keywords, determine the case file type corresponding to the applicable procedure for the case file to be reviewed;

[0012] The content type is determined based on the content represented by the keywords in the case file;

[0013] In the matching rule base, based on the case file type and the content type, search for matching rules for the prompt words;

[0014] The review prompt words are searched according to the prompt word matching rules, and the review prompt words corresponding to the case files to be reviewed are combined to obtain the review prompt word chain.

[0015] In one embodiment, determining the content type based on the content represented by the case file keywords includes:

[0016] Identify each document in the case file to be reviewed, and classify the content elements based on the case file keywords of each document to obtain the content elements of each document;

[0017] The document type of each document is determined based on the content elements described;

[0018] The step of searching for matching rules for prompt words in the matching rule base based on the case file type and the content type includes:

[0019] In the matching rule base, the prompt word matching rules for each of the documents are determined based on the case file type and the document type.

[0020] In one embodiment, extracting case file keywords from the case file to be reviewed includes:

[0021] The case files to be reviewed are identified to obtain the initial case file content;

[0022] The initial case file content is divided into a catalog to obtain a case file catalog;

[0023] Select case file keywords from the structured fields under the case file directory.

[0024] In one embodiment, the step of sequentially reviewing the content of the case file to be reviewed based on the review prompt word chain to obtain the review result includes:

[0025] In the case file to be reviewed, the content to be reviewed corresponding to each of the review prompt words in the review prompt word chain and the review rules corresponding to each of the review prompt words are determined in sequence;

[0026] Based on the review rules that each of the items to be reviewed conforms to, a first review result is generated;

[0027] A second review result is generated based on the review rules that each of the items to be reviewed does not comply with; the first review result is different from the second review result.

[0028] In one embodiment, the method further includes:

[0029] According to the evaluation dimensions, the first review results and the second review results are statistically analyzed to determine the score for each branch;

[0030] The case file score is obtained by combining the scores from the aforementioned branches.

[0031] Secondly, this application also provides a case file review device for use in a review model, the device comprising:

[0032] The extraction module is used to obtain case files to be reviewed and extract case file keywords from the case files to be reviewed;

[0033] The determination module is used to determine the review type of the case file to be reviewed based on the case file keywords;

[0034] The query module is used to search for the audit prompt words corresponding to the case file to be audited in the prompt word rule base according to the audit type and case file keywords, and combine the audit prompt words corresponding to the case file to be audited to obtain the audit prompt word chain;

[0035] The review module is used to review the contents of the case file to be reviewed in sequence according to the review prompt word chain, and obtain the review result.

[0036] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the case file review method in any of the above embodiments.

[0037] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the case file review method in any of the above embodiments.

[0038] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps of the case file review method in any of the above embodiments.

[0039] The aforementioned case file review methods, devices, computer equipment, computer-readable storage media, and computer program products are executed by the review model. First, the keywords within the documents are identified through case file keywords, clarifying the elements in the case file to be reviewed. These elements determine the review type corresponding to the legal procedure, avoiding data anomalies caused by factors such as field errors and ensuring accuracy. Then, a two-dimensional strategy of obtaining review prompts using both review type and case file keywords is employed, enhancing the filtering dimensions of review prompts to obtain more accurate prompts and combine them into a chain, establishing a sequential relationship for review. The appropriate usage order further ensures more accurate acquisition of review prompts. Content is then reviewed sequentially using the chain of review prompts, ensuring stable and accurate review standards through the sequential task of word chains. Furthermore, this method can assist the manual review process, guaranteeing high review efficiency. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a diagram illustrating the application environment of a case file review method in one embodiment.

[0042] Figure 2 This is a flowchart illustrating a case file review method in one embodiment;

[0043] Figure 3 This is a schematic diagram of the process for determining the audit prompt word chain in one embodiment;

[0044] Figure 4 This is a structural block diagram of the volume review device in another embodiment;

[0045] Figure 5 This is a structural block diagram of a case file review device in one embodiment;

[0046] Figure 6 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0048] It should be noted that the terms "first," "second," etc., used in this application can be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish between the first type of data and the second type of data. The term "multiple" used in this application refers to two or more. The term "and / or" used in this application refers to one of the solutions, or any combination of multiple solutions.

[0049] The case file review method provided in this application embodiment can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or located on the cloud or other network servers. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, projection devices, etc. Portable wearable devices can include smartwatches, smart bracelets, head-mounted devices, etc. Head-mounted devices can be virtual reality (VR) devices, augmented reality (AR) devices, smart glasses, etc. Server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0050] In one exemplary embodiment, such as Figure 2 As shown, a case file review method is provided, which is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 202 to 208. Wherein:

[0051] Step 202: Obtain the case files to be reviewed and extract case file keywords from them.

[0052] The case file pending review is a record of the procedure for handling violations; it is the original case file that needs to be reviewed. The procedure for handling violations includes investigation, evidence collection, and approval processes for illegal or irregular conduct. The case file pending review can be a collection of case documents involved in the procedure. A complete case file is created to review its legality, compliance, completeness, and logical consistency.

[0053] Case file keywords are the content and procedural elements in the processing procedure. Content elements include the name of the party involved, the facts of the violation, and the time and place of the violation; procedural elements include, but are not limited to, the list of evidence, the legal provisions on which the penalty is based, the amount of the fine, and the date of the decision. For example, case file keywords can be extracted based on the processing procedure of environmental law enforcement.

[0054] In some embodiments, the server has an enforcement case file database for storing electronic files of cases to be reviewed, such as penalty decisions, on-site inspection records, evidence materials, etc. in PDF, Word, and image formats; and also historical review records and results.

[0055] In some embodiments, when the case file to be reviewed is in PDF or image format, optical character recognition (OCR) and information extraction are required to obtain the keyword categories and corresponding values ​​for the case file in a large model, so that the case file keywords can be processed through the large model. For example, OCR technology can be used to convert non-text format case files such as scanned documents and photos to convert the text in the document images into editable and processable text data; at the same time, predefined information fragments with specific meanings can be identified and extracted from the OCR results or the original case file text as case file keywords, for example, to find and mark the amount of the fine and the decision date in the case file.

[0056] Step 204: Determine the review type of the case file to be reviewed based on the case file keywords.

[0057] The review type indicates the type of legal procedure to which the case file applies. By determining the review type through case file keywords, analysis of the review type can be performed from the perspective of the case file's content. Even if the original review type field in the case file contains errors or anomalies, it will not be misleading, thus allowing for a more accurate determination of the review type. Optionally, the review type includes, but is not limited to, summary procedure types, and general procedure types can be general procedure penalty decisions.

[0058] In some embodiments, case file keywords are arranged according to a preset order corresponding to their respective categories, resulting in arranged case file keywords. A mapping rule is determined based on the position of the extracted case file keywords within the preset order. This mapping rule is then used to map the arranged case file keywords, yielding a keyword mapping result. Finally, the review type of the case file to be reviewed is determined based on this keyword mapping result. Therefore, redefining case file keywords by using a preset order avoids information misalignment due to errors in case file completion, helping to ensure the correct keyword extraction order. Since some prompts in the preset order are omitted or incorrect, a mapping rule can be selected based on the position of the case file keywords within the preset order, and then mapped to determine the mapping rule for this situation. Finally, the review type is determined based on the keyword mapping result, ensuring high accuracy.

[0059] Step 206: In the prompt word rule base, search for the corresponding audit prompt words for the case file to be audited based on the audit type and case file keywords, and combine the audit prompt words corresponding to the case file to be audited to obtain the audit prompt word chain.

[0060] The prompt word rule base is a collection of prompt words corresponding to the case file review rules. It provides review rules for the evaluation process and can be deployed on a server or in the cloud. The prompt word rule base is derived from the extraction and combination of review rules from the knowledge base; it is equivalent to structured review rules and standards. Review prompt words are text instructions input into the large model to guide its review process. Review prompt words are dynamically combined, and the review prompt word chain is a standardized query text containing multi-step review instructions. It is a digital review checklist that precisely controls the large model's review logic and output format during the review process.

[0061] The review rules corresponding to the prompt word rule base include, but are not limited to, the clauses and rules corresponding to the legal and regulatory clauses base, procedural normative rules, document element normative base, and typical problem case base. The legal and regulatory clauses base includes the core clauses of relevant laws on environmental protection and administrative penalties, as well as various environmental protection single-item laws. The procedural normative rules base includes the statutory steps and time limits for each enforcement stage, such as case filing, investigation and evidence collection, notification, hearing, decision, and service. The document element normative base includes the statutory elements and their format norms for various enforcement documents (such as penalty decisions). The typical problem case base includes common errors and compliance examples summarized from historical reviews. Statutory elements include, but are not limited to, party information, facts of the violation, evidence listing, legal basis, penalty content, and remedies. The review prompt word chain is a sequence of review prompt words arranged according to the review order. Following the order of the review prompt words can form a review strategy to ensure a high degree of accuracy in the review.

[0062] In some embodiments, the order and combination of case file keywords are determined in the prompt word rule base based on the review type; according to the combination method, case file keywords are selected and combined to obtain combined keywords; according to the order of use, the combined keywords are used sequentially to find the review prompt words corresponding to the case file to be reviewed. Thus, by determining the dynamically changing keyword combination relationship through the review type, and by determining and combining case file keywords through this keyword combination relationship, combined keywords are obtained. This dynamically changing combination relationship and order of use forms at least two dimensions to more accurately obtain review prompt words.

[0063] In some embodiments, the audit prompts corresponding to the case files to be audited can be combined to obtain an audit prompt chain based on the result of the joint mapping between audit types and audit prompts. For example, mapping can be performed based on audit types of multiple granularities to obtain audit type features of the target dimension; mapping can be performed based on the type to which the audit prompts belong to so that multi-dimensional audit prompts form prompt features of the target dimension; the audit type features and prompt features are concatenated, and the concatenated features are mapped to obtain the order of audit prompts; the audit prompts are combined according to the order of audit prompts to obtain an audit prompt chain.

[0064] In some embodiments, the cue word rule base and corresponding data are a combination of a structured database (such as a relational database) and a vector database, storing rule terms and cases.

[0065] In another embodiment, a pure vector knowledge base can be used. In this case, all laws, regulations, and case documents are embedded and stored in a vector database. During the review process, the most relevant knowledge fragments are dynamically retrieved and prompts are injected using retrieval enhancement generation technology to improve the flexibility of knowledge updates.

[0066] In another embodiment, if the review rules involve a large number of complex logical relationships (such as "Program A must be triggered after event B and condition C must be met"), a graph database can be used to store and traverse these rule nodes and relationships. This alternative approach can more accurately determine the dependencies and reasoning paths between rules, thereby more accurately identifying the audit prompt word chain.

[0067] Step 208: Based on the audit prompt keyword chain, audit the contents of the case file to be audited in sequence to obtain the audit results.

[0068] The review result is the matching result between the content of the case file to be reviewed and the review rules, indicating whether the content of the case file is reasonable. The review model is a neural network model with semantic processing capabilities, which can be a large model or multiple models working together. Optionally, based on the input review prompt word chain, the large model used for review is first controlled to play the role of a senior legal reviewer; then the large model is controlled to analyze and judge the content of the case file to be reviewed item by item according to the review rules in the knowledge base.

[0069] In some embodiments, the audit prompt chain reads each step requirement from the audit prompts, then searches for the corresponding content in the case file to be audited based on each step requirement; subsequently, it invokes internalized or associated audit rules, compares the corresponding content with the audit rules for compliance, and forms an audit result based on the comparison result. Optionally, the audit result may include yes / no, compliant / non-compliant, or specific modification suggestions, or it may be an intermediate conclusion for subsequent processing, or a specific score.

[0070] In some embodiments, by leveraging the contextual understanding and instruction-following capabilities of large models, complex professional rules are transformed into executable sequence tasks by reviewing the cue word chain, thereby forming corresponding control strategies.

[0071] In some embodiments, the chain of review prompts can be input into a domain-specific fine-tuning model to sequentially review the content in the case files. Using specialized corpora in the environmental and legal fields, open-source foundational models (such as LLaMA and ChatGLM) can be fine-tuned with full parameters or with high-efficiency parameters (such as LoRA) to obtain a domain-specific fine-tuning model tailored to environmental law enforcement scenarios. This may result in more accurate performance and reduced dependence on prompts in this domain, but requires additional training costs and data.

[0072] In some embodiments, in extreme cases where the evaluation logic is extremely fixed and the rules are fully structured, a traditional rule-based expert system can completely replace the large model. This replacement will provide absolute determinism and interpretability. Fixed and fully structured rules mean that all logic is expressed using "if...then..." logic.

[0073] In the aforementioned case file review method, the review model acquires the case files to be reviewed and extracts case file keywords from them. Based on these keywords, the model determines the review type of the case files. In the prompt word rule base, the model searches for corresponding review prompt words for the case files based on the review type and case file keywords, and combines these prompt words to obtain a review prompt word chain. The model then reviews the content of the case files sequentially according to the review prompt word chain to obtain the review result. Therefore, by first identifying the keywords within the documents through case file keywords, and clarifying the elements in the case files to be reviewed, the review type corresponding to the legal procedure can be determined using these elements, avoiding data anomalies caused by factors such as field errors and ensuring accuracy. Furthermore, using both review type and case file keywords as a two-dimensional strategy for obtaining review prompt words enhances the filtering dimensions of review prompt words, allowing for more accurate acquisition of review prompt words and their combination into a review prompt word chain. This establishes a sequential review relationship, and the appropriate usage order yields more accurate review prompt words. Finally, the content is reviewed sequentially using the review prompt word chain, ensuring stable and accurate review standards through the word chain under the serialized task. Moreover, this method can assist in the manual review process, ensuring high review efficiency.

[0074] In some embodiments, such as Figure 3 As shown, the review type of the case file to be reviewed is determined based on the case file keywords, including steps 302 to 304; correspondingly, in the prompt word rule base, the review prompt words corresponding to the case file to be reviewed are searched according to the review type and the case file keywords, and the review prompt words corresponding to the case file to be reviewed are combined to obtain the review prompt word chain, including steps 306 to 308. Among them:

[0075] Step 302: Based on the case file keywords, determine the case file type corresponding to the applicable procedure for the case file to be reviewed.

[0076] Case file type indicates the procedural type of the case file to be reviewed during the enforcement of the Environmental Protection Law, based on the judgment criteria. Case file types include, but are not limited to, legal standards and normative standards. Case file types can also include more detailed types such as administrative inspection, administrative penalty (ordinary procedure, simplified procedure), and administrative enforcement. Using case file types can generate coarse-grained type information at the case file level, thereby forming a coarse-grained review strategy.

[0077] In some embodiments, when the case file keywords include keywords from the procedure type field, the case file type can be determined based on the procedure type field in the case file to be reviewed. When the case file keywords include procedure type keywords, the case file type can also be determined more accurately based on the types of procedure type keywords present in the case file and the frequency of those types appearing in the case file to be reviewed. For example, if the case file to be reviewed only has procedure type keywords A1 and A2, it belongs to one case file type; if the case file to be reviewed only has procedure type keywords A1, A2, and A3, it belongs to another case file type.

[0078] Step 304: Determine the content type based on the content represented by the keywords in the case file.

[0079] Content type is the value corresponding to the case file keywords. The content type is dynamically adjusted based on the information in the case file. Optionally, the content represented by the case file keywords includes the semantic content of the case file keywords, and may also include the related content between the case file keywords. The semantic content relationship is established through the related content, thereby obtaining the refined content type.

[0080] In some embodiments, the content type at the case file level can be obtained by mapping based on each case file keyword; multiple granularities of content can be determined based on the granularity of the case file keyword, and then the content at multiple granularities can be mapped to obtain the type mapping result of each granularity; the content type is determined based on the type mapping result of each granularity.

[0081] Step 306: In the matching rule base, search for matching rules for prompt words based on case file type and content type.

[0082] The matching rule base is a collection of prompt word matching rules. It includes various candidate matching rules, identified by two dimensions: case file type and content type, allowing for querying in a two-dimensional manner.

[0083] The prompt word matching rules represent the prompt word search logic for this case pending review. Since case file type and content type can form two levels of granularity, a decoupled nesting of these two levels of granularity allows for a more detailed determination of the prompt word matching rules.

[0084] In one embodiment, the prompt word matching rule includes multiple types of review prompt words arranged in a preset order, and the review prompt words corresponding to the relevant standards can be refined according to the case file type and content type. For example, if the content type of both case file type A1 and case file type A2 is content type B, then one prompt word matching rule is set for case file type A1 and content type B, and another prompt word matching rule is set for case file type A2 and content type B.

[0085] In some embodiments, the matching rule base is indexed using both case file type and content type as indexes, and using both indexes simultaneously enables efficient retrieval of corresponding prompt word matching rules. Optionally, the matching rule base may have a composite index field, allowing queries to be performed on the matching rule base based on the combined result of concatenating prompt words from case file type and content type, in order to obtain prompt word matching rules more efficiently.

[0086] Step 308: Search for review prompts according to the prompt matching rules, and combine the review prompts corresponding to the case files to be reviewed to obtain the review prompt chain.

[0087] In some embodiments, the prompt word library stores candidate review prompt words, and the corresponding review prompt words can be flexibly selected according to the prompt word matching rules to form a review prompt word chain. For example, the review prompt word chain D1 corresponding to the C1 prompt word matching rule includes review prompt words D11 and D12 in sequence; the review prompt word chain D2 corresponding to the C2 prompt word matching rule includes review prompt words D21 in sequence, and two review prompt word chains can be obtained in this way.

[0088] Optionally, the prompt word matching rules include a preset order and prompt word conversion rules; correspondingly, when the content type changes, the candidate review prompt words in the prompt word library can be searched according to the preset order in the prompt word matching rules, and the searched candidate review prompt words can be used as the first review prompt words; at the same time, prompt words can be generated according to the prompt word conversion rules to obtain the second review prompt words.

[0089] For example, in the case files to be reviewed, target words that do not correspond to the commonly used words in the statistical and prompt word library are counted and prompted. The word frequency of each target word is recorded in the corresponding time-retention matrix. The commonly used word library is a pre-established word library composed of non-target words and commonly used words. Non-target words include commonly used words, which refer to words that are used in various case file types. According to the preset word segmentation rules, the short sentences containing the target words are segmented to determine multiple word segmentation paths corresponding to each short sentence. The word frequency of each time-retention matrix and the joint probability corresponding to each word segmentation path are respectively input into the preset implicit semi-Markov model to calculate the joint probability corresponding to each word segmentation path. The word segmentation path with the highest joint probability among all word segmentation paths is selected to obtain the target word segmentation path. All words of the text data to be identified are obtained according to the target word segmentation path of each short sentence. The review word is retrieved for each word segmentation to obtain the second review prompt word. For a single short sentence, there are multiple segmentation paths, each with different segmentation methods. If the segmented words in each path include the target word, its frequency in the time-retention matrix can be determined. The higher the frequency, the more likely the specific word is to represent the true meaning the short sentence intends to express. The higher the accuracy of the specific word, the higher the joint probability of the segmentation paths corresponding to that word. Finally, the target segmentation path with the highest joint probability is used to represent the segmentation path that is closest to the true semantic meaning the text intends to express among the multiple segmentation paths corresponding to each short sentence.

[0090] In one exemplary embodiment, when reviewing documents, they are categorized according to their content elements; if the case file to be reviewed has an ordinary administrative penalty procedure, it is defined as an administrative penalty type, the case file content is obtained simultaneously, and prompt word matching rules are determined in the prompt word matching rule library according to the administrative penalty type and content type.

[0091] In this embodiment, a coarse-grained search dimension is first formed by case file type, and a fine-grained search dimension is formed by content type. Then, the matching rules for prompt words in the matching rule base are determined by these two dimensions, making the search rules dynamic and avoiding the situation where static rules cannot query the review prompt words. The review prompt words are obtained by the prompt word matching rules, and a review prompt word chain is constructed, forming two levels of logical refinement to construct review prompt words for environmental law enforcement records of any form, ensuring strong data availability.

[0092] In some embodiments, determining the content type based on the content represented by case file keywords includes: identifying each document in the case file to be reviewed, and classifying the content elements of each document based on the case file keywords to obtain the content elements of each document; determining the document type of each document based on the content elements. Correspondingly, in the matching rule base, searching for prompt word matching rules based on the case file type and content type includes: determining the prompt word matching rules for each document in the matching rule base based on the case file type and document type.

[0093] A document is a collection of law enforcement information related to a specific event within a case file awaiting review. A case file awaiting review is the combined result of these documents, allowing for a clear understanding of the overall law enforcement situation. Each document possesses a relatively independent format and legal effect, which can be further regulated through detailed content rules to create corresponding matching rules for each document. Content elements are the characteristics of each document's content; these elements can be content tags or content summaries.

[0094] Document type indicates the type of document in legal proceedings. Determining the document type through content elements allows for a more refined review process, moving from case file-level content verification to document-specific analysis. Even if the original document type field contains errors or anomalies, it won't mislead the reader, leading to a more accurate determination of the review type. Furthermore, different document types within the same case file to be reviewed create various dimensions of the file, necessitating targeted classification and evaluation to assess the quality of each document. Optionally, a judgment can be one document type, while a ruling can be another.

[0095] In some embodiments, each document in the case file to be reviewed is identified, and the content elements of each document are classified based on the case file keywords to obtain the content elements of each document. This includes: identifying the case file to be reviewed to obtain the initial case file content; dividing the initial case file content into a directory to obtain the case file directory; extracting case file keywords from the structured fields in each document according to the case file directory, and mapping the extracted case file keywords in each document to obtain the content elements.

[0096] In some embodiments, the review prompts are different for each document. For example, different document scoring rules are defined; a case file may contain 10 documents, an ordinary administrative penalty case file may also require 20 legality standards, and 10 documents may correspond to 50 normative standards. These correspondences are pre-defined and can be found by knowing the case file type and document type.

[0097] In this embodiment, the content elements of each document in the case file to be reviewed are categorized based on the case file keywords to obtain the content elements of each document. Then, the document type of each document is determined according to the content elements, so that each document not only has matching content at the case file level, but also has refined matching rules at the document level, so as to conform to the document's own refined review standards. Subsequently, in the matching rule base, the prompt word matching rules of each document are determined based on the case file type and document type, so as to more efficiently determine the required review prompt words.

[0098] In some embodiments, extracting case file keywords from the case file to be reviewed includes: identifying the case file to be reviewed to obtain initial case file content; dividing the initial case file content into a directory to obtain a case file directory; and selecting case file keywords from the structured fields under the case file directory.

[0099] The initial case file content is the complete content of the case file to be reviewed, which can be obtained through natural language processing. The initial case file content can be obtained by recognizing images or PDFs based on natural language processing (NLP). The initial case file content may include the results of named entity recognition and the type of those results.

[0100] The case file catalog is a collection of content structure fields for the initial case content. Optionally, the case file catalog can be divided according to documents, with each document's identifier or preset fields serving as a structured field within the catalog. Structured fields are used to populate the case file's keyword settings; from an overall perspective, these structured fields collectively represent the content structure of the initial case file content.

[0101] In some embodiments, the initial case file content can be detected sequentially according to the preset field order of the program fields; the content matching the preset field order is arranged to obtain the structured fields of the case file directory.

[0102] In one exemplary embodiment, natural language processing is first used to categorize images or PDFs into directories. Then, named entity recognition (NER) and text classification are used to extract key information from the structured fields of the case file directory as case file keywords. For example, key information from structured fields such as "name of the party concerned," "time and place of violation," "legal provision number on which the penalty is based," "amount of fine," and "date of the decision" are identified and extracted as case file keywords.

[0103] In this embodiment, the overall information of the case file to be reviewed is first obtained by extracting the initial case file content. This overall information is then abstracted into a structured case file directory, avoiding the problem of abnormal case file keyword recognition caused by anomalies in the original directory. After narrowing down the scope of case file keyword extraction using the case file directory, the case file keywords are extracted from the structured fields under the case file directory to ensure the accuracy of the case file keywords.

[0104] In some embodiments, the contents of the case file to be reviewed are reviewed sequentially according to the review prompt word chain to obtain the review result, including: in the case file to be reviewed, according to each review prompt word in the review prompt word chain, the content to be reviewed corresponding to each review prompt word and the review rule corresponding to each content to be reviewed are determined in sequence; a first review result is generated according to the review rules that each content to be reviewed meets; a second review result is generated according to the review rules that each content to be reviewed does not meet; the first review result and the second review result are different.

[0105] Audit rules are knowledge rules associated with audit prompts, used to determine whether the content to be audited matches the audit rules, thereby determining the compliance of the content to be audited. Optionally, audit rules can be inclusion rules, logical matching rules, clause matching rules, or date interval rules.

[0106] The first review result indicates that the review rule has been approved, and it is expressed as "yes," "compliant," or "matches." The first and second review results can be represented by preset values, and they can be divided according to the degree of compliance. When the degree of matching between a piece of content to be reviewed and the review rule is greater than or equal to the matching threshold, a first review result is generated for that content. When the degree of matching between a piece of content to be reviewed and the review rule is less than the matching threshold, a second review result is generated for that content. The first and second review results can be intermediate conclusions or final results.

[0107] In some embodiments, each audit prompt includes at least four steps, each step used to determine the audit rules corresponding to the content to be audited. For example, "Step 1 is used to verify whether the document contains a complete list of legal elements such as the full name and address of the parties, which is an audit rule of inclusion; Step 2 is used to verify whether the description of 'illegal facts' and the 'list of evidence' section logically correspond, which is an audit rule of logical matching; Step 3 is used to determine whether the application of Article XX of the Air Pollution Prevention and Control Law is accurate, which is an audit rule of clause matching; Step 4 is used to calculate whether the interval between 'notification date' and 'decision date' meets the legal requirement of not less than X days...", which is an audit rule of date interval.

[0108] In one embodiment, the large model, acting as a senior legal auditor based on the input chain of audit prompts, analyzes and judges each case to be reviewed according to the audit rules in the knowledge base. The analysis and judgment steps include: reading each requirement in the audit prompts; determining the content to be reviewed corresponding to the audit prompts in the case; and then comparing the internalized or associated audit rules based on the content to be reviewed to generate a "yes / no," "compliant / non-compliant," or specific modification suggestions. Thus, by leveraging the large model's contextual understanding and instruction-following capabilities, complex professional rules are transformed into a sequence of executable tasks. Furthermore, the large model is generated based on the classification or regression of audit prompts and case file information, exhibiting probability variations. This prompting engineering guides the model to output structured and professional review opinions.

[0109] In this embodiment, audit prompts are used to select the content to be audited and the corresponding audit rules from the case files to be audited. During the audit process of the audit model, different contents to be audited in the same case file have their own corresponding audit rules, which transforms the differentiated professional rules into a sequence of tasks that the model can execute. Furthermore, based on the contents to be audited and their differentiated audit rules, two audit results are constructed to more meticulously determine whether the contents to be audited are compliant.

[0110] In some embodiments, the method further includes: statistically analyzing each first review result and each second review result according to the evaluation dimensions to determine the branch score; and combining the branch scores to obtain the case file score.

[0111] Evaluation dimensions are the dimensions used to combine audit results. When using evaluation dimensions, the independent audit results of each item to be audited can be transformed into organized branches, forming an intermediate dimension between the global and individual points, which facilitates multi-dimensional data auditing.

[0112] Branch scores are the review result scores for each evaluation dimension. By combining branch scores, the processing dimensions of the review rules can be combined according to the evaluation dimensions to clarify the review dimensions that need to be adjusted and the corresponding requirements.

[0113] The case file score is a combination of the scores from each branch. The score can be the sum of the branch scores or a weighted combination of the branch scores according to preset weights. Using branch scoring allows for a more precise determination of the review outcome for the case file to be reviewed.

[0114] In some embodiments, determining the branch score based on each first review result and each second review result includes: in each evaluation dimension, obtaining a first review result score based on the number of first review results and a second review result score based on the number of second review results; calculating the difference between the first review result score and the second review result score for the same dimension to obtain the branch score for each evaluation dimension.

[0115] In one embodiment, the first and second review results, including itemized statistics, yield an initial review report with evaluation conclusions and quantitative scores. The dimensions in the initial review report are then combined, and each dimension is evaluated to obtain a case file score. For example, branch scores may include element completeness and accuracy of legal application; element completeness is worth 10 points, with a deduction of 2 points for missing contact information of the parties involved; accuracy of legal application is worth 8 points, with a deduction of 2 points for inaccurate cited clauses. This process continues to obtain the case file score, with a final total score of 85 points.

[0116] In one exemplary embodiment, a separate, rule-based scoring module calculates the final score based on a summary of qualitative conclusions and a formula. This alternative scoring process is more transparent and controllable, and the scoring logic is separated from the large model's reasoning process.

[0117] In one embodiment, the format of the initial review report is adjusted to generate an easy-to-read final review report, which may include specific rules or similar case references from the knowledge base. Simultaneously, the results of each first review, each second review, each branch score, and the case file score (including deductions and issues) are fed back to the knowledge base for subsequent model optimization or rule updates.

[0118] In this embodiment, when using evaluation dimensions, the independent review results of each item to be reviewed can be converted into organized branches, forming an intermediate dimension between the global and individual points, so that the review results of each dimension can be evaluated separately. Based on this, the branch scores are combined into a case file score to form an overall evaluation score. Thus, by using a single review result, branch scores, and case file scores to form multiple granular evaluation results, the case files to be reviewed can be evaluated more accurately.

[0119] In one embodiment, such as Figure 4 As shown, the content of the embodiment is discussed from an overall perspective. This system adopts a typical three-layer architecture: client / user layer 402, intelligent evaluation service layer 404, and data and knowledge layer 406.

[0120] The data and knowledge layer 406 stores the aforementioned law enforcement case file database, procedural normative rule library, document element normative library, and typical problem case library. The function of the data and knowledge layer 406 is to provide the review engine with "fuel" (data to be reviewed) and "benchmark" (review rules), and it is typically deployed on a server or cloud storage. The intelligent review service layer 404 is deployed on a high-performance server or cloud, responsible for the main calculations and logical processing, used to execute steps 202-208 and corresponding embodiments to achieve case file review. The client / user layer 402 is executed by the user terminal, which can be a computer browser on an intranet; it includes a review task submission interface and a review result display interface. The review task submission interface is used for users to upload case file files or select cases to be reviewed and start the review; the review result display interface is used to display the total review score, scores for each dimension (such as legality, standardization, and completeness), a list of problems (including specific locations, rule basis, and modification suggestions), and case file quality trend analysis charts in a clear and intuitive dashboard or report page.

[0121] The aforementioned intelligent review service layer 404 includes a case file analysis and information extraction module, a dynamic review prompt construction module, a large-scale model review reasoning and scoring module, and a review result generation and feedback module.

[0122] The input data for the case file analysis and information extraction module consists of case files awaiting review, retrieved from the environmental enforcement database. These case files are the original documents of the files to be reviewed. The module uses Natural Language Processing (NLP) technology to generate a case file catalog from images or PDFs. Then, using NLP techniques such as Named Entity Recognition (NER) and text classification, it automatically extracts key information from the case files, obtaining case file keywords. For example, it identifies and extracts structured fields such as "name of the party involved," "time and location of the violation," "legal provision number on which the penalty is based," "amount of the fine," and "date of the decision." The output is then the case file keywords, which are the structured case file information, which can be in JSON format.

[0123] The input data for the dynamic review prompt construction module consists of the aforementioned case file information and the review rules retrieved from the knowledge base, i.e., the relevant review rules. Instead of using fixed prompt words, the module dynamically combines and assembles targeted, multi-step review prompt word chains from the knowledge base based on the parsed case file type (e.g., simplified procedure, general procedure) and content type. For example, when reviewing an "Administrative Penalty Decision," the prompt word chain might sequentially include: "Step 1: Verify whether the document contains a complete list of legally required elements such as the full name and address of the party concerned; Step 2: Verify whether the description of the 'illegal facts' and the 'list of evidence' section logically correspond; Step 3: Determine whether the application of Article XX of the 'Air Pollution Prevention and Control Law' is accurate; Step 4: Calculate whether the interval between the 'notification date' and the 'decision date' meets the legal requirement of no less than X days..." Thus, the dynamic review prompt construction module outputs a carefully constructed review prompt word chain that guides the large model through step-by-step, item-by-item reviews.

[0124] The input data for the large-scale model review reasoning and scoring module consists of dynamically generated review prompts, structured case file information, and a large language model as its foundation. This large language model can be an LLM or a model fine-tuned for the environmental field. The large-scale model review reasoning and scoring module acts as a senior legal reviewer, sequentially reviewing the content of the case file according to the review prompt chain based on the foundational large model. It analyzes and judges the case file information item by item according to rules in the knowledge base. Its reasoning process is as follows: the review prompt chain reads each step of the requirements in the review prompts, then searches for the corresponding content in the case file based on each step; subsequently, it calls the internalized or associated review rules, compares the corresponding content with the review rules for compliance, and forms the review result based on the comparison result. Optionally, the review result can include yes / no, compliant / non-compliant, or specific modification suggestions, or it can be an intermediate conclusion for subsequent processing, or a specific score.

[0125] The large-scale model's review reasoning and scoring module leverages the model's contextual understanding and instruction-following capabilities to transform complex professional rules into a sequence of executable tasks. Furthermore, the large-scale model is generated based on review prompts and the case file information to be reviewed, through classification or regression. Thus, through prompting engineering, the model outputs structured and professional review opinions, which can be an initial review report containing sub-item review conclusions and quantitative scores. For example: "Element completeness: 10 points (deduction: missing contact information for the parties involved, deduct 2 points); accuracy of legal application: 8 points (deduction: inaccurate cited clauses, deduct 2 points)... Total case file score: 85 points."

[0126] The input data for the evaluation result generation and feedback module is the initial evaluation report generated by the audit model (large model). This module formats the initial report to generate an easy-to-read final evaluation report, which can include specific rules or similar case references from the knowledge base. Simultaneously, this module feeds the evaluation results (including deductions and issues) back to the knowledge base for subsequent model optimization or rule updates. For example, the final standardized intelligent evaluation report (HTML / Excel format).

[0127] In one exemplary embodiment, the above method can be applied to the daily case file quality review process of the Legal Affairs Department. First, Legal Affairs Department staff (users) upload 10 completed environmental administrative penalty case files to the server via a client. Next, the case file parsing module on the server automatically performs OCR and NLP processing on the 10 documents, extracting the structured information of each case file, i.e., case file keywords. Then, the rule matching and prompt construction steps are executed; that is, the dynamic review prompt construction module generates corresponding review prompt word chains based on the case file type. For example, if the case file type is "General Procedure Penalty Decision," the corresponding "General Procedure Penalty Decision Full-Element Review Rules" are retrieved from the knowledge base to dynamically generate a unique, multi-step review prompt for each case file. Then, the large-scale model review reasoning module sequentially receives the "structured information" and "dynamic prompts" of each case file, performing step-by-step reasoning to complete model reasoning and scoring. For example, the model first determines whether the "Case Filing Approval Form" exists, then verifies whether the facts in the "Investigation Report" are clear, and finally checks whether the delivery time of the "Penalty Notice" meets the time limit. After each judgment is completed, the model internally calculates the score according to the preset deduction rules, obtaining the score for each branch and / or the case file score. Next, the review result generation module summarizes the model's reasoning results for each case file, generating a comprehensive review report that includes the total score, ranking, and common issues. Finally, the report is displayed to staff through the client interface, where staff can click on specific deduction items to view the "problem description" and "legal basis" prompted by the system, and then proceed with further processing.

[0128] In one exemplary embodiment, the above method can be applied to the real-time assistance process for law enforcement officers when drafting documents. First, frontline law enforcement officers draft a "Decision to Order Correction of Illegal Acts," upload it to the server, and begin review, completing data generation and input. Then, the draft text is sent in real-time to the intelligent review service layer 404. The process is the same as above—parsing -> prompt construction -> model reasoning—but the rule base focuses on real-time document standardization checks, which can include essential clauses, standardized terminology, and other dimensions. Then, within seconds, the service layer returns the review results (such as "missing 'method of performance' description" or "incomplete legal name cited") to the terminal. Finally, on the terminal interface, the problem points are marked in the corresponding positions of the document draft to assist law enforcement officers in refining the document on-site.

[0129] In one exemplary embodiment, the overall process follows a strict sequential sequence: user triggers review -> client uploads data to server -> (server-side) case file parsing module starts -> parsing results are passed to dynamic prompt building module -> this module generates prompt words in conjunction with the knowledge base and passes them along with case file information to the large model review and reasoning module -> the model completes reasoning and passes the results to the review result generation module -> this module formats the report and stores it in the database and returns it to the client -> the client interface updates and displays the review report.

[0130] Based on this, relying on a structured legal standards and procedural rules knowledge base, combined with dynamically constructed multi-step review prompts, the large-scale model is precisely guided to locate flaws in case files, automatically generating a list of issues with specific rule basis and modification suggestions. This not only improves the accuracy of case file problem identification but also helps standardize document preparation, strengthen the level of law enforcement standardization, and ensures accuracy. Moreover, it overcomes the limitations of manual experience-based reviews. The large-scale model strictly follows unified legal standards and procedural rules for review, completely avoiding fluctuations in review standards caused by differences in human experience, emotions, and cognition, achieving "consistent judgments for similar cases," ensuring the objectivity and fairness of review results, meeting the compliance requirements of environmental supervision, and ensuring consistency. Furthermore, it achieves a shift from "fully manual" to "automated," enabling the system to complete the automatic initial review of large batches of case files in a short time, significantly improving review efficiency and processing capacity. At the same time, it frees legal supervisors from tedious and repetitive basic inspection work, allowing them to focus on the analysis of complex and difficult cases and the improvement of rules, comprehensively improving the quality and credibility of environmental law enforcement and ensuring high efficiency.

[0131] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0132] Based on the same inventive concept, this application also provides a case file review device for implementing the case file review method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more case file review device embodiments provided below can be found in the limitations of the case file review method described above, and will not be repeated here.

[0133] In one exemplary embodiment, such as Figure 5 As shown, a case file review device is provided, applied to a review model, including:

[0134] Extraction module 502 is used to obtain the case files to be reviewed and extract case file keywords from the case files to be reviewed;

[0135] The determination module 504 is used to determine the review type of the case file to be reviewed based on the case file keywords;

[0136] The query module 506 is used to search for the audit prompt words corresponding to the case file to be audited in the prompt word rule base according to the audit type and case file keywords, and combine the audit prompt words corresponding to the case file to be audited to obtain the audit prompt word chain;

[0137] The review module 508 is used to review the contents of the case file to be reviewed in sequence according to the review prompt word chain, and obtain the review result.

[0138] In one embodiment, the determining module 504 is configured to:

[0139] Based on the aforementioned case file keywords, determine the case file type corresponding to the applicable procedure for the case file to be reviewed;

[0140] The content type is determined based on the content represented by the keywords in the case file;

[0141] In the matching rule base, based on the case file type and the content type, search for matching rules for the prompt words;

[0142] The review prompt words are searched according to the prompt word matching rules, and the review prompt words corresponding to the case files to be reviewed are combined to obtain the review prompt word chain.

[0143] In one embodiment, the determining module 504 is configured to:

[0144] Identify each document in the case file to be reviewed, and classify the content elements based on the case file keywords of each document to obtain the content elements of each document;

[0145] The document type of each document is determined based on the content elements described;

[0146] In the matching rule base, the prompt word matching rules for each of the documents are determined based on the case file type and the document type.

[0147] In one embodiment, the extraction module 502 is used to:

[0148] The case files to be reviewed are identified to obtain the initial case file content;

[0149] The initial case file content is divided into a catalog to obtain a case file catalog;

[0150] Select case file keywords from the structured fields under the case file directory.

[0151] In one embodiment, the audit module 508 is configured to:

[0152] In the case file to be reviewed, the content to be reviewed corresponding to each of the review prompt words in the review prompt word chain and the review rules corresponding to each of the review prompt words are determined in sequence;

[0153] Based on the review rules that each of the items to be reviewed conforms to, a first review result is generated;

[0154] A second review result is generated based on the review rules that each of the items to be reviewed does not comply with; the first review result is different from the second review result.

[0155] In one embodiment, the audit module 508 is configured to:

[0156] According to the evaluation dimensions, the first review results and the second review results are statistically analyzed to determine the score for each branch;

[0157] The case file score is obtained by combining the scores from the aforementioned branches.

[0158] Each module in the aforementioned case file review device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0159] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a case review method.

[0160] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0161] In one embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above method embodiments.

[0162] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the steps in the above method embodiments.

[0163] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0164] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0165] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0166] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0167] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for reviewing case files, characterized in that, Performed by the audit model, the method includes: Obtain case files to be reviewed, and extract case file keywords from the case files to be reviewed; The review type of the case file to be reviewed is determined based on the keywords in the case file. In the prompt word rule base, the prompt words corresponding to the case file to be reviewed are found according to the review type and case file keywords, and the prompt words corresponding to the case file to be reviewed are combined to obtain the review prompt word chain; Based on the aforementioned audit prompt keyword chain, the contents of the case file to be audited are audited sequentially to obtain the audit results.

2. The method according to claim 1, characterized in that, The process of determining the review type of the case file to be reviewed based on the case file keywords includes: Based on the aforementioned case file keywords, determine the case file type corresponding to the applicable procedure for the case file to be reviewed; The content type is determined based on the content represented by the keywords in the case file; In the matching rule base, based on the case file type and the content type, search for matching rules for the prompt words; The review prompt words are searched according to the prompt word matching rules, and the review prompt words corresponding to the case files to be reviewed are combined to obtain the review prompt word chain.

3. The method according to claim 2, characterized in that, The determination of content type based on the content represented by the case file keywords includes: Identify each document in the case file to be reviewed, and classify the content elements based on the case file keywords of each document to obtain the content elements of each document; The document type of each document is determined based on the content elements described; The step of searching for matching rules for prompt words in the matching rule base based on the case file type and the content type includes: In the matching rule base, the prompt word matching rules for each of the documents are determined based on the case file type and the document type.

4. The method according to claim 1, characterized in that, The extraction of case file keywords from the case files to be reviewed includes: The case files to be reviewed are identified to obtain the initial case file content; The initial case file content is divided into a catalog to obtain a case file catalog; Select case file keywords from the structured fields under the case file directory.

5. The method according to claim 1, characterized in that, The step of sequentially reviewing the content of the case file to be reviewed based on the review prompt word chain to obtain the review result includes: In the case file to be reviewed, the content to be reviewed corresponding to each of the review prompt words in the review prompt word chain and the review rules corresponding to each of the review prompt words are determined in sequence; Based on the review rules that each of the items to be reviewed conforms to, a first review result is generated; A second review result is generated based on the review rules that each of the items to be reviewed does not comply with; the first review result is different from the second review result.

6. The method according to claim 5, characterized in that, The method further includes: According to the evaluation dimensions, the first review results and the second review results are statistically analyzed to determine the score for each branch; The case file score is obtained by combining the scores from the aforementioned branches.

7. A case file review device, characterized in that, Applied to the audit model, the device includes: The extraction module is used to obtain case files to be reviewed and extract case file keywords from the case files to be reviewed; The determination module is used to determine the review type of the case file to be reviewed based on the case file keywords; The query module is used to search for the audit prompt words corresponding to the case file to be audited in the prompt word rule base according to the audit type and case file keywords, and combine the audit prompt words corresponding to the case file to be audited to obtain the audit prompt word chain; The review module is used to review the contents of the case file to be reviewed in sequence according to the review prompt word chain, and obtain the review result.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images