Document intelligent review method and system based on multiple rule bases

Abstract

The present invention discloses a document intelligent review method and system based on a multi-rule base, and relates to the technical field of document intelligent review. The document intelligent review method and system based on a multi-rule base, through optical character recognition and format parsing technology, combined with natural language processing to identify the rule base matching features of the document content, can effectively extract the key information of the document and provide a basis for subsequent rule matching. Load and select the multi-rule base in the database for matching review. In the process of rule matching review, the document content is analyzed and classified by multi-dimensional features and rules through the random forest algorithm, so as to identify and mark the document content that does not meet the rule matching. Through the support vector machine algorithm, the marked document content is compared with the overall structure of the document, the logical conflicts in the document are identified, and they are marked again. Ensure the consistency and logic within the document, and further improve the accuracy and comprehensiveness of the review.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent document review, and in particular to a method and system for intelligent document review based on a multi-rule base. Background Art

[0002] With the rapid development of digitalization and informatization, a large number of enterprises and institutions have accumulated massive amounts of document data in their daily operations. These documents involve various types such as contracts, financial statements, and legal documents. In this context, how to efficiently and accurately review the content of documents to ensure their compliance, accuracy, and completeness has become an important requirement for improving work efficiency, reducing labor costs, and reducing human errors. Traditional manual review methods are difficult to adapt to the growing demand for document review because they are time-consuming and prone to errors. At the same time, modern society has increasingly higher requirements for data security and privacy protection. How to ensure the security of data while ensuring the efficiency of document content review has also become a key issue.

[0003] The current document review methods mainly include manual review and automated review methods based on simple rule matching. However, these methods have the following defects: Manual review: Relying on manual verification of document content one by one is not only inefficient, but also subject to the experience and professional ability of the reviewers, and prone to omissions or misjudgments. At the same time, in the face of large-scale document data, the cost of manual review is too high and it is difficult to meet actual needs. Simple rule matching: Although the automated review system based on fixed rules can improve efficiency to a certain extent, due to the static and single nature of the rules, it is difficult to adapt to complex and changeable review scenarios. For example, when it involves cross-domain and multi-conditional logical judgments, the system shows obvious deficiencies. The characteristics of delayed rule updates and difficulty in dynamic adjustment also lead to insufficient accuracy and reliability of its review results.

[0004] Therefore, there is an urgent need for a document intelligent review method and system based on multiple rule bases, which can combine multiple rule bases and intelligent algorithms, dynamically adapt to the needs of different scenarios, improve the comprehensiveness and accuracy of the review, and reduce manual participation, so as to meet the needs of efficient and secure document review. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides a document intelligent review method and system based on a multi-rule library, which solves the problems of the above-mentioned background technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a document intelligent review method based on multiple rule bases, comprising the following steps: S1. receiving the document content to be reviewed, performing optical character recognition and format analysis on the document content, identifying the rule base matching features of the document content through natural language processing, and using them to match the multiple rule bases; S2. loading and selecting multiple rule bases in the database for matching based on the rule base matching features of the document content, and performing rule matching review; S3. in the process of rule matching review, performing multi-dimensional feature analysis and rule classification on the document content through the random forest algorithm, identifying and marking the document content that does not meet the rule matching; S4. comparing the marked document content with the overall structure of the document based on the marked document content, identifying the logical conflicts in the document through the support vector machine algorithm and performing secondary marking.

[0007] Furthermore, the specific process of performing optical character recognition and format parsing on the document content is as follows: receiving the document content to be reviewed, converting the document into a digital image format, and performing optical character recognition to extract the text content in the document; performing format parsing on the extracted document content, identifying the structured information of paragraphs, titles, tables, and images in the document, and converting the identified structural information into structured data.

[0008] Furthermore, the specific process of identifying the rule base matching features of document content through natural language processing is as follows: perform word segmentation on the text content in the document, split the text into words and phrases and annotate them; extract specific entities in the document through named entity recognition and identify semantic relationships; perform syntactic analysis on the document content, build a syntactic tree and extract key grammatical structure features; match the document content with the rules in the multi-rule base according to the extracted semantic information and structural features, and extract matching features that meet the rules. The rule base matching features include semantic features, structural features, and format features.

[0009] Furthermore, according to the rule base matching features of the document content, the specific process of loading and selecting multiple rule bases in the database for matching is as follows: based on the rule base matching features extracted from the document content, the semantics, structure and format of the document are comprehensively analyzed to obtain the rule matching index; based on the rule matching index, the category of the rule base that needs to be matched is determined, and the qualified rule base is loaded from the database; the multiple rule bases include a semantic rule base, a structural rule base and a format rule base.

[0010] Furthermore, the specific process of performing rule matching review is as follows: the rule library loaded from the database is compared with the document content one by one, and the document is reviewed according to the matching rules in the rule library; the document content is subjected to multi-dimensional feature analysis and rule classification through the random forest algorithm, and a matching operation is performed on each matching rule in the document, the matching degree value is calculated and the corresponding matching results are recorded; whether the matching degree meets the requirements is determined according to the preset threshold, and the document content that does not meet the rule matching is identified and marked.

[0011] Furthermore, the specific process of multi-dimensional feature analysis and rule classification of document content through random forest algorithm is as follows: extract semantic features, structural features and format features in the document, convert the extracted features into numerical feature vectors, construct annotated data sets, and train the random forest classification model through the annotated data sets; according to the random forest classification model, the document content to be reviewed is classified by rules, the matching value of each rule is obtained, and the classification results are generated.

[0012] Furthermore, the specific process of identifying and marking document content that does not match the rules is as follows: output the matching value of each rule through the random forest classification model to determine the matching degree between the document content and the rule; compare the calculated matching value with the preset threshold. If the matching degree is lower than the threshold, it means that the part of the document content fails to match the rule, and the document content that does not match the rule is marked.

[0013] Furthermore, the specific process of identifying logical conflicts in documents and performing secondary marking through the support vector machine algorithm is as follows: based on the marked document content, the semantic and logical relationship features in the document are extracted, the marked content is aligned with the overall structure of the document, and the structured features are extracted; the extracted logical features are converted into numerical vectors, the input data of the support vector machine model is constructed, and the document is subjected to logical conflict analysis through the trained support vector machine model; the logical conflict part in the document is identified according to the analysis results of the support vector machine model, and the identified logical conflict part is secondary marked.

[0014] The document intelligent audit system based on multiple rule bases includes the following modules: a parsing module, a rule matching module, a feature analysis module, and a logic conflict identification module; the parsing module is used to receive the document content to be audited, perform optical character recognition and format analysis on the document content, and identify the rule base matching features of the document content through natural language processing for matching multiple rule bases; the rule matching module is used to load and select multiple rule bases in the database for matching according to the rule base matching features of the document content, and perform rule matching audit; the feature analysis module is used to perform multi-dimensional feature analysis and rule classification on the document content through the random forest algorithm during the rule matching audit process, and identify and mark the document content that does not meet the rule matching; the logic conflict identification module is used to compare the marked document content with the overall structure of the document according to the marked document content, identify the logical conflicts in the document through the support vector machine algorithm, and perform secondary marking.

[0015] The present invention has the following beneficial effects:

[0016] (1) This intelligent document review method based on multiple rule bases uses optical character recognition and format parsing technology combined with natural language processing to identify the rule base matching features of document content, which can effectively extract key information of the document and provide a basis for subsequent rule matching. First, the method receives the content of the document to be reviewed and parses it to extract the semantic, structural and format features of the document. Then, based on these features, the multiple rule bases in the database are loaded and selected for matching review to ensure that the document content meets the preset rule base requirements. The automation of this process greatly improves the efficiency and accuracy of document review and reduces the need for manual intervention.

[0017] (2) In the process of rule matching and review, the document intelligent review system based on multiple rule bases uses the random forest algorithm to perform multi-dimensional feature analysis and rule classification on the document content, thereby identifying and marking the document content that does not match the rule. Furthermore, the support vector machine algorithm is used to compare the marked document content with the overall structure of the document, identify logical conflicts in the document, and mark them again. The implementation of this step helps to ensure the consistency and logic within the document, further improve the accuracy and comprehensiveness of the review, and ensure that the document not only meets the regulatory requirements, but also maintains the rationality of the structure and the rigor of the logic.

[0018] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a flow chart of the document intelligent review method based on multiple rule bases of the present invention.

[0020] Figure 2This is a flow chart of the document intelligent review system based on multiple rule bases of the present invention. DETAILED DESCRIPTION

[0021] The embodiments of the present application solve the problems of low efficiency, poor accuracy, and difficulty in adapting to complex scenario requirements in large-scale document review by using a document intelligent review method and system based on multiple rule bases. Through optical character recognition and natural language processing technology, the rule base matching features of the document are automatically extracted and analyzed, and multiple rule bases are combined for precise matching, which significantly improves the automation level and processing speed of document review. The use of random forest algorithms and support vector machine algorithms to perform multi-dimensional feature analysis and logical conflict identification on the document content further enhances the accuracy and reliability of document review, ensuring that the document not only meets regulatory requirements, but also maintains structural and logical consistency. Therefore, the document intelligent review method and system provided by the present application can effectively reduce the workload of manual review and improve the efficiency and quality of document processing.

[0022] The overall idea of ​​the solution in the embodiments of this application is as follows:

[0023] Receive the document content to be reviewed, perform optical character recognition and format analysis on the document content, and identify the rule base matching features of the document content through natural language processing for matching multiple rule bases.

[0024] According to the rule base matching characteristics of the document content, multiple rule bases in the database are loaded and selected for matching, and rule matching review is performed.

[0025] During the rule matching review process, the random forest algorithm is used to perform multi-dimensional feature analysis and rule classification on the document content to identify and mark document content that does not match the rules.

[0026] According to the marked document content, the marked document content is compared with the overall structure of the document, and the logical conflicts in the document are identified and re-marked through the support vector machine algorithm.

[0027] See also Figure 1The embodiment of the present invention provides a technical solution: a document intelligent review method based on a multi-rule base, comprising the following steps: S1. receiving the document content to be reviewed, performing optical character recognition and format analysis on the document content, identifying the rule base matching features of the document content through natural language processing, and matching the multi-rule base; S2. loading and selecting the multi-rule base in the database for matching according to the rule base matching features of the document content, and performing rule matching review; S3. in the rule matching review process, performing multi-dimensional feature analysis and rule classification on the document content through the random forest algorithm, identifying and marking the document content that does not meet the rule matching; S4. comparing the marked document content with the overall structure of the document according to the marked document content, identifying the logical conflict in the document through the support vector machine algorithm and performing secondary marking.

[0028] In this implementation scheme, step S1: the system receives the document content to be reviewed. These document contents include contracts, financial statements, legal documents and other types of documents. Then, the system performs optical character recognition and format analysis on the document content: Optical Character Recognition (OCR): by scanning paper documents or pictures, identify the text content therein and convert it into an editable digital text format. Format analysis: This step analyzes the structure of the document, identifies the title, paragraph, table and other contents in the document, and provides the necessary document format information for subsequent rule matching. Then, the system identifies the "rule base matching features" in the document content through natural language processing (NLP) technology. Natural language processing is a branch of artificial intelligence that mainly studies how computers understand, generate and process human language. Here, it is used to analyze the document content and extract feature information related to rule matching. These rule base matching features will be used as input to match different rule bases (such as semantic rule bases, structural rule bases and format rule bases) to determine whether the document content meets the preset audit standards. Step S2: Based on the rule base matching features extracted in step S1, the system loads multiple rule bases from the database and selects an appropriate rule base for matching according to the characteristics of the document content. The rule matching review process performed can check whether the document content meets the requirements in different rule bases: Semantic rule base: used to review whether the meaning of the document is correct and whether it conforms to the logic of the context. Structural rule base: Check whether the organizational structure of the document is standardized, such as whether there is correct chapter division, title format, etc. Format rule base: Ensure that the document format meets the regulations, such as font, font size, line spacing, etc. Step S3: During the rule matching review process, the system uses the random forest algorithm to perform multi-dimensional feature analysis and rule classification. Random forest is an ensemble learning method that generates multiple decision trees and votes on their results to improve the accuracy and robustness of predictions. In this step, the random forest algorithm analyzes the multi-dimensional features of the document content (such as semantics, structure, format, etc.), and classifies the document content based on the analysis results, and finally generates a matching value for each rule. Matching value: indicates the degree of matching between a document and a certain rule base. The higher the matching value, the more consistent the document content is with the requirements of the rule base. If the document content does not match the rule, the system will identify and mark the part that does not meet the rule. Step S4: After identifying and marking the document content that does not comply with the rules, the system compares the marked document content with the overall structure of the document. Next, the system identifies logical conflicts in the document through the support vector machine algorithm and re-marks these conflicting parts. Support vector machine (SVM) is a machine learning algorithm commonly used in classification and regression analysis. It divides data into different categories by finding the best separating hyperplane. In this step, SVM is used to check whether there are logical inconsistencies or conflicts in the content of the document, and additionally mark these conflicting parts for further review.

[0029] Specifically, the specific process of optical character recognition and format parsing of document content is as follows: receiving the document content to be reviewed, converting the document into a digital image format, and performing optical character recognition to extract the text content in the document; performing format parsing on the extracted document content, identifying the structured information of paragraphs, titles, tables, and images in the document, and converting the identified structural information into structured data.

[0030] In this embodiment, the system receives the content of the document to be reviewed. In order for the computer to process these documents, they first need to be converted into a digital image format, such as JPEG or PNG format. This step is to ensure that the document content can be read and analyzed by the computer system. The text in the digital image is extracted using optical character recognition (OCR) technology. OCR is a technology that converts text in an image into text. It can extract letters, numbers, punctuation marks and other information from the image or scan of the document and convert it into a text format that can be processed by the computer. This step is a key link in the intelligent review of documents because it enables the system to obtain the text information in the document, laying the foundation for subsequent rule matching and analysis. Once the text content is extracted, the system begins format parsing. The purpose of this process is to identify different structural elements in the document, such as paragraphs, titles, tables, images, etc. Format parsing can help the system understand the layout and structure of the document for subsequent processing. By identifying the position and hierarchical relationship of each element, the system can analyze the document more accurately. The structural information obtained by parsing will be converted into structured data. Structured data is the layout, content and format of the document converted into a standardized data format, which makes the subsequent intelligent review process more efficient and accurate. For example, information such as data in tables, image locations, and hierarchical relationships of titles will all be presented in a structured form, providing accurate data support for subsequent rule matching and review work.

[0031] Specifically, the specific process of identifying the rule base matching features of document content through natural language processing is as follows: perform word segmentation on the text content in the document, split the text into words and phrases and annotate them; extract specific entities in the document through named entity recognition and identify semantic relationships; perform syntactic analysis on the document content, build a syntactic tree and extract key grammatical structure features; match the document content with the rules in the multi-rule base based on the extracted semantic information and structural features, and extract matching features that meet the rules. The rule base matching features include semantic features, structural features, and format features.

[0032] In this embodiment, in natural language processing, word segmentation is the process of splitting continuous text into meaningful words or phrases. Through word segmentation, the system can identify the basic units in the text, such as words or phrases. Word segmentation not only helps the system understand each individual word, but also provides a basis for subsequent annotation and analysis. Named entity recognition is a technology in natural language processing that is used to identify specific entities in a document, such as names, place names, dates, company names, etc. In this step, the system automatically extracts important entities in the document through the NER algorithm and identifies their categories. For example, the name, amount, date and other information of the company may be extracted in the financial statement. Semantic relationship recognition refers to identifying the association and relationship between different entities in the document, such as the "contract" in "Company A and Company B's contract" is the relationship between Company A and Company B. By analyzing these semantic relationships, the system can understand the deep meaning of the document content, thereby performing more accurate rule matching. Syntactic analysis is to understand the role and relationship of words in a sentence by analyzing the grammatical structure of the sentence. The syntax tree is a hierarchical structure that shows the grammatical relationship between the various components of a sentence (such as subject, predicate, object, etc.). By constructing the syntax tree, the system can extract key grammatical features, such as noun phrases, verb phrases, etc., which helps to understand the grammatical structure in the document and further provide information for rule matching. After completing steps such as word segmentation, NER, and syntactic analysis, the system will match the document content with the rules in the multi-rule base based on the extracted semantic information and structural features. The rules in the rule base include semantic rules (such as the occurrence or association of certain specific entities), structural rules (such as specific syntactic patterns or paragraph structures), and format rules (such as matching of table or image formats). Through these rules, the system can extract matching features that meet specific rules, thereby providing a basis for subsequent review and analysis.

[0033] Specifically, according to the rule base matching features of the document content, the specific process of loading and selecting multiple rule bases in the database for matching is as follows: based on the rule base matching features extracted from the document content, the semantics, structure and format of the document are comprehensively analyzed to obtain the rule matching index; based on the rule matching index, the category of the rule base that needs to be matched is determined, and the qualified rule base is loaded from the database; the multiple rule bases include a semantic rule base, a structural rule base and a format rule base.

[0034] In this embodiment, the extraction of rule base matching features of document content: In the previous steps, the system has extracted semantic features, structural features and format features in the document through natural language processing. Each feature represents a different aspect of the document content. For example, semantic features may include specific entities (such as company name, time), structural features may include the organization of paragraphs in the document, and format features may include the existence and format of tables, images or lists. These extracted features will be used as basic input for the subsequent rule matching process. Comprehensive analysis of the semantics, structure and format of the document: The system conducts a comprehensive analysis of the document content based on the extracted features. Semantic analysis can help the system understand the keywords and important concepts in the document; structural analysis helps the system identify the organization of the document, such as the title, the level of paragraphs, chapters, etc.; format analysis helps identify whether there are special formats such as tables and images in the document. Based on the comprehensive analysis, the system will calculate a "rule matching index", which indicates the degree of matching between the document and each rule base. Obtaining the rule matching index: The rule matching index is a value calculated by the system based on the degree of matching between the document content and the rules in the rule base. The higher the value, the better the match between the document content and a rule base. This index is a comprehensive evaluation result obtained by comparing the semantic features, structural features and format features of the document with the rules in the rule base. The rule matching index includes the semantic matching index, structural matching index and format matching index. : The semantic matching index measures the degree of matching between the document content and the semantic rule base, which is achieved by extracting key entities, terms or semantic relations in the document and comparing them with the semantic rules in the rule base. ; The number of semantic entities extracted from the document. : indicates the Semantic Entity Corresponding rules in the rule base The similarity between them. Obtained by cosine similarity. Structural matching index : The structural matching index measures the degree of matching between the document structure (such as paragraphs, titles, chapters) and the rules in the structural rule base. The structure of a document is usually obtained by parsing the document's layout format (such as chapter hierarchy, paragraphs, tables, etc.). ; The number of structured parts in the document. : indicates the Structured part (such as paragraphs, titles, etc.) and structural rules in the rule library The degree of match between them. Usually calculated using regular expressions or pattern matching algorithms. Format matching index : The format matching index measures the degree of matching between the document’s layout, fonts, images and other format features and the rules in the format rule library. The format analysis includes the document’s visual elements, such as fonts, table layout, image position, etc. ; p represents the number of formatting elements in the document, : indicates the Format Features (fonts, colors, images) and formatting rules in the rule library The matching degree between them is obtained by calculating the similarity between the font, color and image format features in the document and the corresponding format rules in the rule library. Determine the category of rule libraries that need to be matched: According to the semantic matching index, structural matching index and format matching index, the system can determine which rule libraries are most suitable for the review of the current document. For example, if the document contains a large number of specific named entities (such as company names, product models, etc.), the system may give priority to the semantic rule library. If the document has a complex chapter structure or charts, the system may choose the structural rule library or the format rule library. The matching index helps the system intelligently select the most relevant rule library to ensure the efficiency and accuracy of document review. Once the category of rule libraries that need to be matched is determined, the system will load these rule libraries from the database. The rule library in the database usually includes multiple categories, such as: Semantic rule library: contains rules related to named entities and semantic relationships in the document, which are used to ensure the semantic consistency and legality of the document content. Structural rule library: contains rules related to document format and organizational structure, such as chapter format, paragraph level, etc. Format rule library: contains specifications for elements such as tables, images, special formats, etc. in the document, which are used to check whether the format of the document meets the requirements. The system loads and selects these multiple rule bases to conduct a comprehensive matching audit based on the different characteristics of the document content (semantics, structure, format). Each rule base plays a different role in the matching process, helping the system to more accurately determine whether the document meets the corresponding specifications or rules.

[0035] Specifically, the specific process of performing rule matching review is as follows: compare the rule library loaded from the database with the document content one by one, and review the document according to the matching rules in the rule library; use the random forest algorithm to perform multi-dimensional feature analysis and rule classification on the document content, and perform matching operations on each matching rule in the document, calculate the matching value and record the corresponding matching results; determine whether the matching degree meets the requirements based on the preset threshold, and identify and mark the document content that does not meet the rule matching.

[0036] In this implementation scheme, the rule base is compared with the document content: the system first compares the rule base loaded from the database with the document content to be reviewed one by one. Each rule base contains multiple specific rules, which define the standards that various types of content in the document should meet. The system checks the document content item by item according to these rules to determine whether the document meets the predetermined standards. Matching rule execution: During the comparison process, the system will perform rule matching operations on each content in the document to evaluate whether the document content meets the standards in the rule base. These rules may cover aspects such as the format, structure, and semantics of the document. Matching value calculation: For each rule matching operation, the system will calculate a matching value. This matching value reflects the degree of compliance of the document content with the corresponding rule in the rule base. The higher the matching value, the better the degree of compliance of the document content with the rule. Determine whether the matching degree meets the requirements: the system will preset a matching threshold. Only when the matching value exceeds this threshold, the document is considered to meet the requirements of the rule. If the matching value is lower than the threshold, the system will consider that the part of the content does not meet the rule. Identify and mark content that does not conform to the rules: The system marks the document content based on the matching results and highlights the parts that do not conform to the rules. These contents that do not conform to the rules need further manual review or processing to ensure that the document finally meets the requirements.

[0037] Specifically, the specific process of multi-dimensional feature analysis and rule classification of document content through random forest algorithm is as follows: extract semantic features, structural features and format features in the document, convert the extracted features into numerical feature vectors, construct annotated data sets, and train the random forest classification model through the annotated data sets; according to the random forest classification model, the document content to be reviewed is classified by rules, the matching value of each rule is obtained, and the classification results are generated.

[0038] In this implementation scheme, during the document review process, the specific process of multi-dimensional feature analysis and rule classification of document content by random forest algorithm is as follows: Feature extraction: First, semantic features, structural features and format features are extracted from the document. Semantic features may include key entities (such as company name, product name, etc.) and semantic relationships (such as logical relationships between sentences) in the document; structural features include the organizational structure of the document, such as chapters, paragraphs, titles, lists, etc.; format features include typesetting format, fonts, colors, images and other information in the document. Feature vectorization: The extracted semantic, structural and format features are converted into numerical feature vectors. These feature vectors provide a standardized representation for each document, which is convenient for training and prediction of machine learning models. Construct annotated data sets: Use annotated data sets to train random forest classification models. These annotated data sets include document content and corresponding rule labels. Through these annotated data, the system can learn the relationship between documents and different rules. Train random forest classification model: Based on the annotated data set, train the random forest classification model. Random forest is an ensemble learning method that makes classification decisions by combining multiple decision trees. Each decision tree will classify the document and ultimately decide whether the document meets a certain rule by majority voting. Rule classification and matching value acquisition: Once training is completed, the random forest model can be applied to new documents to be reviewed. The model classifies the rules based on the extracted feature vectors to determine whether the document content meets the requirements in the rule base. For each rule, the model calculates a matching value to indicate the degree of compliance of the document content with the rule. The calculation formula for obtaining the matching value of each rule: In the application of the random forest classification model, the calculation of the matching value is usually based on the output of the model, indicating the degree of match between the document and the rule. The matching value of each rule is calculated by the following formula: ;in: It is The matching value of a rule, is the number of positive examples predicted by the model to conform to the rule (TruePositive, the number of correct matches); It is the number of false negatives (the number of incorrect matches) predicted by the model as not meeting the rule. For each rule, the system uses the random forest model to give a matching value, which indicates the degree of matching between the document content and the rule base. Finally, the system will classify the document content as "compliant with the rule" or "not compliant with the rule" based on these matching values.

[0039] Specifically, the specific process of identifying and marking document content that does not match the rules is as follows: output the matching value of each rule through the random forest classification model to determine the matching degree between the document content and the rule; compare the calculated matching value with the preset threshold. If the matching degree is lower than the threshold, it means that the part of the document content fails to meet the rules, and the document content that does not match the rules is marked.

[0040] In this implementation scheme, the random forest classification model outputs a matching value: the system processes the document content through the trained random forest classification model and outputs the matching value of each rule. These matching values ​​indicate the degree of matching between the document content and each rule in the rule base. The higher the matching value, the more consistent the document content is with the rule. Comparison of the matching value with the preset threshold: the system sets a matching threshold to determine whether the document content complies with the rule. When the matching value of the document content is lower than the threshold, it means that the content of this part of the document fails to meet the requirements of the rule and there is a situation of non-compliance with the rule. Marking the document content that does not comply with the rule: when the matching value of the document is lower than the preset threshold, the system will automatically mark these parts of the document content that do not comply with the rule. The marked content is usually highlighted or distinguished from the rest of the document by other means, so as to facilitate further inspection and processing by the auditor. Through this process, the system can effectively identify the parts of the document that do not comply with the rules and mark them in time, ensuring that the auditor can efficiently focus on dealing with potential errors or non-compliant content. This automated marking process greatly improves the efficiency and accuracy of document review, reduces the need for manual intervention, and reduces the risk of human error.

[0041] Specifically, the specific process of identifying logical conflicts in documents and performing secondary marking through the support vector machine algorithm is as follows: based on the marked document content, the semantic and logical relationship features in the document are extracted, the marked content is aligned with the overall structure of the document, and the structured features are extracted; the extracted logical features are converted into numerical vectors, the input data of the support vector machine model is constructed, and the document is subjected to logical conflict analysis through the trained support vector machine model; the logical conflict part in the document is identified according to the analysis results of the support vector machine model, and the identified logical conflict part is secondary marked.

[0042] In this embodiment, the semantic and logical relationship features in the document are extracted: relevant semantic information and logical relationship features are extracted from the document content that has been marked as non-compliant. These features usually include entities involved in the document (such as people, company names, dates, etc.), the order of events, contextual relationships (such as causal relationships, conditional relationships, etc.), and other logical structures (such as contradictions, repetitions, etc.). These features can help the system identify potential logical problems in the document. Align the marked content with the overall structure of the document: align the marked document content with the overall structure of the document. This process involves analyzing elements such as the paragraph structure, chapter titles, table and chart positions of the document, and locating the marked content (such as the part that does not comply with the rules) to the overall structure of the document. This helps to determine the position and context of the marked content in the document, so as to better understand its logical relationship with other content. Convert logical features into numerical vectors: When performing logical analysis on the document, the extracted semantic and logical features are converted into numerical vectors. This step is to convert complex text information into a digital format that can be understood by the machine learning model. It includes using the bag of words model (Bag of Words) and TF-IDF (term frequency-inverse document frequency) to convert logical features in the document into numerical values. Constructing the input data of the support vector machine (SVM) model: Use the converted numerical vector as the input data of the support vector machine (SVM) model. SVM is a supervised learning algorithm, which is usually used for classification problems. Here, the goal of the model is to learn to identify which parts of the document content have logical conflicts based on the samples in the training data. SVM can help the system distinguish between logical and illogical parts in the document, and learn and classify based on the labeled training data. Using the trained SVM model for logical conflict analysis: The trained SVM model is used to perform logical conflict analysis on the content of the document to be reviewed. The support vector machine analyzes the logical structure of the document and identifies parts that may have logical conflicts. Inconsistencies, repeated information, or unreasonable causal relationships in the document can all be identified as logical conflicts by the system. Secondary marking of logical conflict parts: Based on the analysis results of the SVM model, the system will secondary mark the identified logical conflict parts. These conflicts may be caused by contradictory content or due to logically incoherent structures in the document. When marking, the system usually highlights or otherwise marks these parts for further review and correction by the reviewer.

[0043] See also Figure 2The document intelligent review system based on multiple rule bases includes the following modules: parsing module, rule matching module, feature analysis module, and logic conflict identification module; the parsing module is used to receive the document content to be reviewed, perform optical character recognition and format analysis on the document content, and identify the rule base matching features of the document content through natural language processing for matching multiple rule bases; the rule matching module is used to load and select multiple rule bases in the database for matching according to the rule base matching features of the document content, and perform rule matching review; the feature analysis module is used to perform multi-dimensional feature analysis and rule classification on the document content through the random forest algorithm during the rule matching review process, and identify and mark the document content that does not meet the rule matching; the logic conflict identification module is used to compare the marked document content with the overall structure of the document according to the marked document content, identify the logical conflicts in the document through the support vector machine algorithm and perform secondary marking.

[0044] In this implementation scheme, parsing module: The main function of this module is to receive the document content to be reviewed and perform preliminary processing on the document. First, the document is converted into editable text content through optical character recognition (OCR) technology; second, format parsing is performed to identify structural elements such as paragraphs, titles, tables, and images in the document; finally, the semantic features and rule matching features of the document are extracted through natural language processing (NLP) technology to provide basic data for subsequent rule matching. Rule matching module: This module loads the relevant multi-rule library in the database according to the rule library matching features of the document content, and performs matching review. The rule library classifies the document according to the semantic, structural and format features of the document, and determines whether the document complies with the corresponding regulations by comparing the document content with the rules, and marks the parts that do not comply with the rules. Feature analysis module: During the rule matching review process, the feature analysis module conducts in-depth analysis of the document content through the random forest algorithm. The algorithm classifies the rules and identifies the document content that does not match the rules by analyzing multiple dimensional features in the document (such as semantics, structure, format, etc.). Marking the parts that do not comply with the rules is convenient for subsequent processing and manual review. Logical conflict identification module: This module uses the support vector machine (SVM) algorithm to further analyze the content of the marked document and identify the logical conflicts in the document. By comparing the overall structure of the document with the marked content, it detects whether the document has inconsistencies, inconsistent information or logical incoherence, and re-marks these logical conflicting parts to ensure the logic and consistency of the document.

[0045] In summary, this application has at least the following effects:

[0046] The intelligent document review method and system based on multiple rule bases greatly reduces the workload and time of manual review and improves the overall efficiency of document review through intelligent matching and automated processing of multiple rule bases. The system combines multi-dimensional analysis of semantic, structural and format features to ensure a high degree of match between document content and rules, effectively reducing review errors caused by human negligence or rule omissions. Through automated rule matching and logical conflict identification, the system can efficiently identify document content and logical conflicts that do not conform to the rules, reduce manual intervention, and thus reduce the occurrence of human errors. Using advanced technologies such as natural language processing, random forest algorithm and support vector machine algorithm, the system can intelligently select rule bases according to the characteristics of different documents and conduct a comprehensive review of document content, meeting the needs of modern digitalization and intelligence. During the document review process, the system can efficiently process document data, ensure data security and privacy protection, and meet corporate and legal compliance requirements.

[0047] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0048] The present invention is described with reference to flowcharts and / or block diagrams of systems, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0049] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0050] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0051] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0052] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A document intelligent review method based on a multi-rule base, characterized in that: The following steps are involved: S1. Receive the document content to be reviewed, perform optical character recognition and format analysis on the document content, and identify the rule base matching features of the document content through natural language processing for matching multiple rule bases; S2. According to the rule base matching features of the document content, load and select multiple rule bases in the database for matching, and perform rule matching review; S3. During the rule matching review process, the document content is analyzed in multiple dimensions and classified by the random forest algorithm to identify and mark the document content that does not meet the rule matching requirements; S4. According to the marked document content, the marked document content is compared with the overall structure of the document, and the logical conflicts in the document are identified by a support vector machine algorithm and re-marked; The specific process of matching the rule base according to the document content and loading and selecting multiple rule bases in the database for matching is as follows: Based on the matching features of the rule base extracted from the document content, the semantics, structure and format of the document are comprehensively analyzed to obtain the rule matching index; According to the rule matching index, determine the category of the rule base that needs to be matched, and load the rule base that meets the conditions from the database; The multiple rule bases include a semantic rule base, a structural rule base and a format rule base; The specific process of executing rule matching review is as follows: The rule base loaded from the database is compared with the document content one by one, and the document is reviewed according to the matching rules in the rule base; The document content is analyzed in multiple dimensions and classified by the random forest algorithm. The matching operation is performed on each matching rule in the document, the matching degree value is calculated, and the corresponding matching results are recorded. Determine whether the matching degree meets the requirements based on the preset threshold, and identify and mark the document content that does not meet the rule matching; The specific process of performing multi-dimensional feature analysis and rule classification on document content by using the random forest algorithm is as follows: Extract semantic features, structural features, and format features from documents, convert the extracted features into numerical feature vectors, construct annotated datasets, and train the random forest classification model through the annotated datasets; According to the random forest classification model, the document content to be reviewed is classified by rules, the matching value of each rule is obtained, and the classification result is generated; The specific process of identifying and marking document content that does not match the rule is as follows: The matching degree value of each rule is output through the random forest classification model to determine the matching degree between the document content and the rule; The calculated matching value is compared with the preset threshold. If the matching value is lower than the threshold, it means that the content of the document fails to meet the rule, and the document content that does not meet the rule matching is marked; The specific process of identifying logical conflicts in documents and performing secondary marking by using the support vector machine algorithm is as follows: According to the marked document content, the semantic and logical relationship features in the document are extracted, the marked content is aligned with the overall structure of the document, and the structural features are extracted; Transform the extracted logical features into numerical vectors, construct the input data of the support vector machine model, and perform logical conflict analysis on the document through the trained support vector machine model; The logical conflict parts in the document are identified according to the analysis results of the support vector machine model, and the identified logical conflict parts are re-marked.

2. The document intelligent review method based on multiple rule bases according to claim 1 is characterized in that: The specific process of optical character recognition and format parsing of document content is as follows: Receive the document content to be reviewed, convert the document into a digital image format, and perform optical character recognition to extract the text content in the document; Perform format analysis on the extracted document content, identify the structured information of paragraphs, titles, tables, and images in the document, and convert the identified structural information into structured data.

3. The document intelligent review method based on multiple rule bases according to claim 2 is characterized in that: The specific process of identifying the rule base matching features of document content through natural language processing is as follows: Perform word segmentation on the text content in the document, split the text into words and phrases and annotate them; Extract specific entities in documents through named entity recognition and identify semantic relationships; Perform syntactic analysis on the document content, build a syntactic tree and extract key grammatical structure features; According to the extracted semantic information and structural features, the document content is matched with the rules in the multi-rule base, and matching features that meet the rules are extracted; The rule base matching features include semantic features, structural features, and format features.

4. A document intelligent review system based on a multi-rule base, applied to a document intelligent review method based on a multi-rule base as claimed in any one of claims 1 to 3, characterized in that: It includes the following modules: parsing module, rule matching module, feature analysis module, and logic conflict identification module; The parsing module is used to receive the document content to be reviewed, perform optical character recognition and format parsing on the document content, and identify the rule base matching features of the document content through natural language processing for matching multiple rule bases; The rule matching module is used to load and select multiple rule bases in the database for matching according to the rule base matching features of the document content, and perform rule matching review; The feature analysis module is used to perform multi-dimensional feature analysis and rule classification on the document content through the random forest algorithm during the rule matching review process, and identify and mark the document content that does not meet the rule matching; The logic conflict identification module is used to compare the marked document content with the overall structure of the document according to the marked document content, identify the logic conflicts in the document through the support vector machine algorithm and perform secondary marking.

Images