A method, device and apparatus for intelligent review of documents

Abstract

The present invention discloses a method, equipment and device for intelligent file review, which relates to the technical field of intelligent file review methods, and includes the following modules: a file feature data acquisition module, a file feature data analysis module, a file review model optimization module and a review file evaluation module. The method includes the following steps: collecting and extracting the first file data and field feature data from the file information library of the institution to be reviewed, establishing a file intelligent review mathematical vector machine model based on a machine learning method, and generating a file review confidence assessment coefficient; comparing and analyzing the file review confidence assessment coefficient with a preset file review confidence assessment coefficient, and controlling and adjusting abnormal comparison results; not only can the consistency of the review standard be maintained, thereby improving the accuracy of the overall review, but also the potential problems can be instantly identified and marked, and real-time feedback can be provided to help decision makers respond quickly.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent file review methods, and more specifically, to an intelligent file review method, equipment and device. Background Art

[0002] Intelligent document review refers to the analysis and review of text, images or other data types through machine learning, natural language processing (NLP) and other artificial intelligence technologies to identify key information, potential risks, compliance issues or sensitive data. The combination of intelligent document review methods and equipment provides enterprises and institutions with an efficient and accurate solution to meet compliance and risk management needs. This method not only improves the efficiency of document processing, but also reduces the error rate of manual review, providing strong support for document management in all walks of life.

[0003] For example, the Chinese invention patent with announcement number: CN118333333A discloses a method, device, equipment and storage medium for allocating review resources, including the following steps: for the target file to be reviewed, determine the members other than the collaborative editing target file from the project members as candidate reviewers; determine the evaluation feature vector corresponding to each candidate reviewer; input the evaluation feature vector of each candidate reviewer into the target model, predict the matching degree of each candidate reviewer in reviewing the target file through the target model, and output the prediction result corresponding to each candidate reviewer; according to the prediction result corresponding to each candidate reviewer, determine the target reviewer responsible for reviewing the target file from all candidate reviewers. The device can automatically allocate the target reviewer who is familiar with the content of the target file to be reviewed, effectively improving the efficiency and accuracy of allocating review resources in multi-person collaboration scenarios.

[0004] For example, the Chinese invention patent with announcement number: CN117151603A discloses a contract review method, device, storage medium and electronic device, including the following steps: obtaining the contract to be reviewed, analyzing the contract to be reviewed and obtaining key information therein, automatically judging whether the key information meets the agreed conditions, and if not, generating prompt information to remind the contract management personnel to pay attention. The method and device analyzes the contract based on text parsing technology and image recognition technology, automatically judges and compares related information, realizes automated and intelligent review of the contract, and effectively improves the efficiency of contract review.

[0005] The above disclosed technical solutions have at least the following technical problems:

[0006] On the one hand, existing natural language processing algorithms still have limitations in understanding context and semantics, especially when dealing with complex sentence structures and metaphors, which can easily lead to misjudgment. When dealing with language features and industry terms in certain specific fields, the recognition rate may be reduced because the data set used to train the model may not be comprehensive enough.

[0007] On the other hand, in the process of large-scale document review, the current system may not be able to efficiently process a large amount of information, resulting in inefficient information processing and lack of real-time performance. In addition, in some resource-limited situations, the intelligent review system consumes a lot of computing resources due to its slow processing speed, which in turn increases costs.

[0008] In view of the above problems, the present invention proposes a solution. Summary of the invention

[0009] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a method for intelligent file review, which collects and extracts first file feature data and field feature data through an institutional file information system, combines a machine learning method, establishes a mathematical model for intelligent file review, generates a file review confidence assessment coefficient, and optimizes the established mathematical model for intelligent file review; finally, the file review confidence assessment coefficient generated by the model is compared and analyzed with the preset file review confidence assessment coefficient, and abnormal analysis results are processed; so as to solve the problem that the previous system may not be able to efficiently process a large amount of information, resulting in low information processing efficiency and insufficient real-time performance.

[0010] To achieve the above object, the present invention provides the following technical solutions:

[0011] A method for intelligent document review, characterized in that it includes the following steps: based on the institution's document information system, collecting documents to be reviewed and extracting document feature data therefrom; based on a machine learning method, combining the first document feature data and the field feature data, establishing a mathematical model for intelligent document review and generating a document review confidence assessment coefficient; based on the document review results of the institution to be reviewed in previous years as historical document data, optimizing the established mathematical model for intelligent document review; comparing and analyzing the document review confidence assessment coefficient generated by the model with the preset document review confidence assessment coefficient, and processing abnormal analysis results.

[0012] In a preferred embodiment, the file feature data includes first file feature data and domain feature data; the first file feature data includes a file text feature coefficient and a file structure feature coefficient, and the domain feature data includes a domain feature coefficient.

[0013] In a preferred embodiment, the method for obtaining the file text feature coefficient is as follows: preprocess the original file obtained from the file information system, extract the document's potential subject words based on the LDA technology, and use the Word2Vec word vector model to convert each word in the file into a word vector; compare the subject word vector appearing in a specific file to be reviewed with all subject word vectors, calculate the word frequency component, form all the files to be reviewed into a set, combine the number of times the specific subject word appears in all the files to be reviewed, calculate the inverse document frequency, and further calculate the file text feature coefficient.

[0014] In a preferred embodiment, the method for obtaining the file structure characteristic coefficient is as follows: collect and record files of various types and formats from the file information system of the institution to be reviewed, perform pre-cleaning, and extract metadata in the pre-processed file content based on the text parsing library, including the number of paragraphs and words in the file; calculate the average number of words in a paragraph in the file, and further calculate the word distribution in the file; compare the number of files that meet the format specifications with the total number of files to be reviewed to obtain the file format compliance coefficient; and then combine the word distribution and the file format compliance coefficient to calculate the file structure characteristic coefficient.

[0015] In a preferred embodiment, the method for obtaining the domain characteristic coefficient is as follows: extract basic information words from the document to be reviewed through the medical document information system to be reviewed, and extract relevant domain characteristic words from the document; after removing the duplicate and missing values ​​of the basic information words in the document, normalize the number of collected basic information words; count the relevant domain characteristic words corresponding to the basic information feature category, and calculate the correlation between different types of basic information features and relevant domain features to obtain the domain characteristic coefficient.

[0016] In a preferred embodiment, the method for obtaining the file review confidence assessment coefficient is as follows: based on a machine learning method, a mathematical model for intelligent file review is established according to the file text feature coefficient, the file structure feature coefficient and the domain feature coefficient, and the error factor and the guess factor are combined to generate the file review confidence assessment coefficient.

[0017] The technical effects and advantages of the intelligent file review method, device and apparatus of the present invention are as follows:

[0018] 1. The present invention calculates the document text feature coefficient, document structure feature coefficient and domain feature coefficient by extracting feature data from the document information system of the institution to be reviewed. First, the original document is preprocessed, potential subject words are extracted based on LDA technology, and each word is converted into a word vector using the Word2Vec word vector model; by comparing the subject word vectors in the document to be reviewed, the word frequency component is calculated, and the inverse document frequency is obtained by combining the number of occurrences of the subject word in all the documents to be reviewed, and the text feature coefficient is further calculated to help us identify and evaluate the different attributes of the text; collect files in various formats, and after pre-cleaning, use the text parsing library to extract metadata, calculate the average number of paragraph words and word distribution; combine the word distribution and format compliance to calculate the document structure feature coefficient, so as to evaluate the file format compliance; remove duplicate and missing values ​​from the extracted basic information words, normalize the number, and count the correlation with the domain feature words, so as to obtain the domain feature coefficient, evaluate the recognition and processing capabilities of the document intelligent review device for domain-specific content, and make the document intelligent review method more efficient and reliable in practical applications.

[0019] 2. The present invention uses a machine learning method, combines the first file feature data and the field feature data, establishes a mathematical model for intelligent file review, and generates a file review confidence assessment coefficient; it can effectively identify potential risk files, facilitate subsequent in-depth review and monitoring, and help organizations take preventive measures in advance; the confidence assessment coefficient can quickly identify high-risk or highly suspicious files, and help reviewers prioritize important files, thereby saving time and resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 The figure is a schematic diagram of the structure of a method for intelligent file review according to the present invention.

[0021] Figure 2 The present invention is a schematic diagram of the structure of a file intelligent review device. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] Embodiment 1, Figure 1 The present invention provides a file intelligent review method, equipment and device, which include the following modules: a file feature data acquisition module, a file feature data analysis module, a file review model optimization module and a review file evaluation module.

[0024] It should be noted that in this implementation case, medical institution documents are selected as the specific scenario for the application of the methods, equipment and devices for document review, but it is not limited to this scenario.

[0025] S10, file feature data collection module: based on the hospital institution file information system, collect the patient's medical records, notice text, diagnosis information and other key data, and extract file feature data therefrom, the file feature data includes file feature first data and domain feature data, the file feature first data includes file text feature coefficient and file structure feature coefficient, the domain feature data includes domain feature coefficient.

[0026] The text feature coefficient is an indicator or parameter used to quantify and describe the characteristics of text content. It is mainly used in the fields of text analysis, natural language processing, and information retrieval. It can help identify and evaluate the different attributes of the text, thus providing a basis for subsequent processing and analysis. It has the following functions in the intelligent file review method equipment:

[0027] Feature-based classification: By calculating text feature coefficients (such as TF-IDF, word frequency, etc.), smart review equipment is able to classify documents into predefined categories such as medical, legal, financial, etc. This helps in quickly screening and archiving documents.

[0028] Content auditing: Text feature extraction can help identify sensitive information and compliance issues. For example, through feature extraction, audit equipment can automatically detect documents containing personal information (PII), health information (PHI), or other protected information.

[0029] Information retrieval: By calculating the text feature coefficients, the device can optimize the search engine and provide more accurate search results. For example, in medical documents, after extracting features, relevant case or drug information can be quickly found through keywords.

[0030] Content comparison: When dealing with a large number of documents, text feature coefficients can help assess the similarity between documents. This is very useful for duplicate checking, plagiarism detection, and version management.

[0031] The specific method for obtaining the text feature coefficient is as follows:

[0032] Preprocess the original medical file data, delete irrelevant symbols, punctuation and special characters, reduce redundancy, and ensure data quality and consistency.

[0033] LDA technology is used to extract potential topic words of the document, and the Word2Vec word vector model is used to convert each word into a vector. The number of times the topic word appears in the i-th document to be reviewed is recorded as , record all word vectors that appear in the document to be reviewed as , calculate the word frequency component in the document to be reviewed. The specific calculation formula is as follows:

[0034]

[0035] All the files to be reviewed are formed into a set, denoted as , , is a positive integer, and the inverse document frequency is calculated. The specific calculation formula is as follows:

[0036]

[0037] Combine the word frequency component and inverse document frequency in the document to be reviewed to calculate the text feature coefficient of the specific document to be reviewed. The specific calculation formula is as follows:

[0038] .

[0039] The file structure characteristic coefficient refers to the characteristic parameters used to quantify and evaluate different structural elements in the file in the file intelligent review method, equipment and device. It can effectively help understand the overall layout and information organization of the file, identify and evaluate the file content and its potential risks or compliance. Thereby improving the efficiency and accuracy of document processing; the specific role of the file intelligent review method, equipment and device is as follows:

[0040] Enhanced review accuracy: Structural feature coefficients can help analyze the organization of file content, such as paragraph layout, title, image location, etc., to determine the rationality and compliance of the file; by extracting file structural feature coefficients, the type, format and internal structure of the file can be effectively identified. This is crucial for distinguishing normal files from potentially harmful files (such as viruses, Trojans, etc.).

[0041] Improve efficiency: The structural feature coefficient can help the review system quickly locate abnormal structures or non-compliant content in the file, reducing the time and effort of manual inspection; in the automatic review process, the device can quickly classify and filter files according to the structural feature coefficient, improving review efficiency.

[0042] Support intelligent decision-making: The extracted feature coefficients can be used as inputs to machine learning models to support intelligent review decisions. During the review process, real-time analysis of file structure feature coefficients can provide users with instant feedback and suggest whether further review or other measures are needed.

[0043] Improve security: By analyzing the structural characteristics of files, potential security risks can be discovered, such as hidden malicious code or inappropriate information, so as to prevent security risks in advance; file structure characteristic coefficients can also be used for compliance checks to ensure that files comply with relevant regulations and industry standards and reduce legal risks.

[0044] Multi-dimensional analysis: The file structure feature coefficient can be combined with other features (such as text content, metadata, etc.) to form a multi-dimensional comprehensive analysis to improve the comprehensiveness of the review; by comparing the structural feature coefficients of normal files, the system can more easily identify abnormal files and conduct in-depth analysis.

[0045] The specific steps for obtaining the structural characteristic coefficients are as follows:

[0046] First, data pre-cleaning is performed on medical documents of various types and formats, such as electronic medical records, medical imaging reports, prescriptions, and diagnostic records, to remove redundant information and format inconsistent data, thereby achieving data standardization for subsequent processing;

[0047] Extract the preprocessed document content and metadata based on text parsing libraries (such as docx and pdfminer in Python). The extracted metadata includes the number of paragraphs and words in the document.

[0048] Convert the text into vectors based on GloVe, capture the number of paragraphs and words in the file, and record the number of paragraphs as ,in = , is an integer; the number of words captured in the paragraph is recorded as , calculate the average number of words in the file, denoted as , further calculate the word distribution in the file, the specific calculation formula is as follows:

[0049] Z

[0050] Extract digital data indicators such as font size and list symbols in the document, compare and confirm whether the review document meets the format specifications, and record the digital data that meets the indicators as , all digital data to be reviewed are recorded as , calculate the conformity coefficient of the file format, the specific calculation formula is as follows:

[0051]

[0052] Based on the word distribution in the document and the conformity coefficient of the file format, the structural feature coefficient of the file is calculated. The specific calculation method is as follows:

[0053] .

[0054] The domain feature coefficient refers to an indicator used to quantify and evaluate the importance or influence of features related to a specific field and expertise; in the document intelligent review method device, it is usually used in machine learning and data analysis to measure the consistency and matching degree between the features involved in the beginning and end of the document. In the document intelligent review method device, the specific functions of the domain feature coefficient are as follows:

[0055] Improve classification accuracy: In the intelligent review process, domain feature coefficients can help improve the accuracy of the classification model. By weighting the features, the review system can better distinguish between normal files and potential violation files.

[0056] Supporting semantic understanding: Domain feature coefficients can help the system understand the semantic context of text. In intelligent review, understanding the meaning of text is often more important than simply identifying keywords. By analyzing domain features, the system can better capture the implicit meaning and sentiment in the text, thereby making more reasonable review judgments.

[0057] Handling diversity and complexity: In multi-domain or cross-domain document review, the domain feature coefficient can effectively handle the diversity and complexity of the content. By identifying features in different domains, the system can better handle different types of documents and reduce misjudgments and missed judgments.

[0058] Enhanced explainability: Based on the domain feature coefficients, the decision-making process of the review system can become more explainable. By providing feature importance analysis, reviewers can understand why the system makes specific judgments, thereby enhancing trust in the results of intelligent review.

[0059] The specific steps for obtaining the domain characteristic coefficient are as follows:

[0060] Obtain relevant medical data, including medical records, imaging examination results, and other data, through the medical document information system to be reviewed; extract basic information data from the documents to be reviewed, including the patient's blood pressure and blood sugar, etc., and extract relevant field feature words from the documents, including feature words for describing patient symptoms, corresponding words for treatment plans, etc.

[0061] The number of basic information words collected is recorded as , Indicates the basic information feature category, , is a positive integer; after removing the duplicate and missing values ​​of the basic information words in the file, the number of basic information words collected is normalized, and the normalization method is as follows:

[0062]

[0063] in Indicates The minimum number of basic information words under the category, Indicates The maximum number of basic information words in each category;

[0064] The relevant field feature words corresponding to the basic information feature category are extracted and counted, and the number of occurrences of the relevant field feature words is recorded as , calculate the correlation between different types of basic information features and related domain features, that is, the domain feature coefficient. The specific calculation formula is as follows:

[0065]

[0066] It can be seen from the above formula that by detecting the domain characteristic coefficient, the file intelligent review device can improve its ability to identify and process domain-specific content, making the file intelligent review method more efficient and reliable in practical applications.

[0067] S20, file feature data analysis module: based on machine learning methods, combined with the first file feature data and the domain feature data, a mathematical model for intelligent file review is established to generate a file review confidence assessment coefficient.

[0068] The specific calculation formula of the document review confidence assessment coefficient is as follows:

[0069]

[0070] In the formula, is the error factor, is the guessing factor, and its value is arrive between, is the weight factor of the file structure feature coefficient, is the weight factor of the document text feature coefficient, is the weight factor of the domain characteristic coefficient, and .

[0071] It should be noted that in the mathematical model of intelligent document review, the preset error factors and the preset guessing factors are mainly used to compensate for the errors of the model in practical applications. Specifically, the error factors are used to reflect the systematic deviations that may occur when the model analyzes the text features, structural features and domain features of the document. Due to the complexity of the document content and the limitations of the text analysis algorithm, the model may have certain errors when identifying and extracting keywords, and the error factors can effectively correct these deviations. On the other hand, the guessing factors are used to quantify the reasonable speculations made by the model when it cannot be completely determined. These speculations may come from the uncertainty in the feature extraction process or from the ambiguity of domain knowledge. By introducing the guessing factors, the model can more flexibly cope with the diversity of text features and ensure the robustness and accuracy of the document review confidence assessment results.

[0072] In terms of preset content, the error factor can be set based on the error distribution of historical data analysis and text feature recognition, while the guess factor can be defined based on the uncertainty of domain knowledge and the model's tolerance for non-completely matching text structures. These two factors work together to ensure that the document intelligent review model maintains high reliability and adaptability in a variety of actual situations.

[0073] In this implementation case, feature data is extracted from the document information system of the institution to be reviewed, and the document text feature coefficient, document structure feature coefficient and domain feature coefficient are calculated. First, the original document is preprocessed, potential subject words are extracted based on LDA technology, and each word is converted into a word vector using the Word2Vec word vector model; by comparing the subject word vectors in the document to be reviewed, the word frequency component is calculated, and the inverse document frequency is obtained by combining the number of occurrences of the subject word in all the documents to be reviewed, and the text feature coefficient is further calculated to help us identify and evaluate the different attributes of the text; collect files in various formats, and after pre-cleaning, use the text parsing library to extract metadata, calculate the average number of words in a paragraph and the word distribution; combine the word distribution and format compliance to calculate the document structure feature coefficient to evaluate the file format compliance; remove duplicate and missing values ​​from the extracted basic information words, normalize the number, and count the correlation with the domain feature words, so as to obtain the domain feature coefficient, evaluate the recognition and processing capabilities of the document intelligent review device for domain-specific content, and make the document intelligent review method more efficient and reliable in practical applications.

[0074] Based on machine learning methods, combined with the first file feature data and field feature data, a mathematical model for intelligent file review is established to generate a file review confidence assessment coefficient. This model can effectively identify potential risk files, facilitate subsequent in-depth review and monitoring, and help institutions take preventive measures in advance. The confidence assessment coefficient can quickly identify high-risk or highly suspicious files, helping reviewers prioritize important files, thereby saving time and resources.

[0075] S30, document review model optimization module: based on the document review results of the institution to be reviewed in previous years as historical document data, the established document intelligent review mathematical model is optimized.

[0076] In this step, the document review data of previous years of the review agency are collected and sorted, and the collected historical document data are cleaned to remove missing values ​​and outliers in the data, and the processed data are formed into a historical document data set. ,in , is a positive integer. Substitute the processed historical file data into the file review mathematical vector machine model for calculation, compare the calculated confidence assessment results with the historical file review results, and based on the comparison results, modify the model parameters, reduce the error value, further test and train, and optimize the file review mathematical vector machine model.

[0077] Specifically, based on the mathematical model of intelligent file review, a comparison is made with the actual historical file review results, and the text feature coefficients of the files to be reviewed in the file sample output by the model, the file structure feature coefficients and the domain feature coefficients are compared with the file text feature coefficients, file structure feature coefficients and the domain feature coefficients in the actual file sample. If the specific coefficients in the actual file sample are too different from the specific coefficients in the file sample output by the model, it is necessary to adjust and optimize the model parameters in the mathematical model of intelligent file review. If the adjusted and optimized model still cannot reach the specific coefficients in the actual file sample, it is necessary to collect and organize the file data to be reviewed again, check the correctness of the collected file data, and verify the accuracy and correctness of the processing process to ensure the reliability of the data output by the mathematical model of intelligent file review.

[0078] S40, review document assessment module: compare and analyze the document review confidence assessment coefficient generated by the model with the preset document review confidence assessment coefficient, and process abnormal analysis results.

[0079] The specific method for comparing and analyzing the document review confidence assessment coefficient generated by the model with the preset document review confidence assessment coefficient is as follows:

[0080] When the generated document review confidence assessment coefficient is greater than or equal to the preset accuracy confidence threshold, the review document sample is marked as a high accuracy sample;

[0081] When the generated document review confidence assessment coefficient is less than the accuracy confidence threshold and greater than the information security threshold, the review document sample is marked as a medium accuracy sample;

[0082] When the generated document review confidence assessment coefficient is less than or equal to the information security threshold, the review document sample will be marked as a low-accuracy sample; and the relevant personnel will be informed to mark these low-confidence samples and review the data set.

[0083] S50, the process of handling abnormal analysis results is continuously tested and analyzed, and the results of the re-examination are evaluated.

[0084] The process of handling abnormal analysis results is to continuously detect and analyze data and evaluate the results of the re-examination, specifically:

[0085] When reviewing the data set, collect several file review confidence assessment coefficients during the review to establish a data set

[0086] Calculate the standard deviation of the document review confidence assessment coefficient using the document review confidence assessment coefficient in the data set and the average of the document review confidence assessment coefficients , and the standard deviation of the document review confidence assessment coefficient and the average of the document review confidence assessment coefficients The standard deviation reference thresholds are set and the pre-set reference threshold of the document review confidence assessment coefficient Conduct comparative analysis;

[0087] like , then issue a warning and review again;

[0088] like and , then continuous testing is performed;

[0089] like and , then re-examine to confirm that it is correct.

[0090] Figure 2 A device for a file intelligent review method includes a file feature data acquisition module, a file feature data analysis module, a file review model optimization module, a review file evaluation module and a re-file review module, and there are connections between the modules;

[0091] A document feature data collection module, which is used to collect documents to be reviewed based on the document information system of the institution to be reviewed and extract document feature data therefrom;

[0092] The file feature data analysis module is used to establish a mathematical model for intelligent file review based on machine learning methods and combined with file feature data to generate a file review confidence assessment coefficient;

[0093] The document review model optimization module is used to optimize the established document intelligent review mathematical model based on the document review results of the institution to be reviewed in previous years as historical document data, so as to obtain the optimized document intelligent review mathematical model;

[0094] The review document assessment module is used to compare and analyze the document review confidence assessment coefficient generated by the optimized document intelligent review mathematical model with the preset document review confidence assessment coefficient, and to process abnormal analysis results.

[0095] The re-file review module is used to continuously detect and analyze data during the process of handling abnormal analysis results and evaluate the results of the re-review.

[0096] The present application also includes an electronic device, the electronic device comprising:

[0097] at least one processor; and,

[0098] a memory communicatively connected to the at least one processor; wherein,

[0099] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to perform a document intelligent review method.

[0100] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0101] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0102] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0103] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0104] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0105] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for intelligent file review, characterized in that: The following steps are involved: Based on the document information system of the institution to be examined, the document to be examined is collected, and document feature data is extracted therefrom, wherein the document feature data includes first document feature data and field feature data; The first file feature data includes a file text feature coefficient and a file structure feature coefficient, and the domain feature data includes a domain feature coefficient; The method for obtaining the file text feature coefficient is as follows: Preprocess the original documents obtained from the document information system and extract the potential subject words of the documents based on LDA technology; Use the Word2Vec word vector model to convert each topic word in the document to be reviewed into a topic word vector; Compare the subject word vectors appearing in each file to be reviewed with the subject word vectors of all files to be reviewed, and calculate the word frequency component; A collection of all the documents to be reviewed is constructed, and the inverse document frequency is calculated based on the number of times a specific subject word appears in all the documents to be reviewed; The document text feature coefficient is calculated by combining the word frequency component and the inverse document frequency; The method for obtaining the file structure characteristic coefficient is as follows: Collect and pre-clean documents of various types and formats from the document information system of the institution to be reviewed; Extract metadata from the preprocessed file content based on the text parsing library, the metadata includes the number of paragraphs and words in the file; Calculate the average number of words in a paragraph in the file based on the metadata, and calculate the word distribution in the file; Compare the number of files that meet the preset format specifications with the total number of files to be reviewed, and calculate the file format compliance coefficient; Combining the word distribution and file format conformity coefficient, data analysis is performed to calculate the file structure feature coefficient; The specific method for obtaining the domain characteristic coefficient is as follows: The number of basic information words collected is recorded as , Indicates the basic information feature category, , is a positive integer; after removing the duplicate and missing values ​​of the basic information words in the file, the number of basic information words collected is normalized, and the normalization method is as follows: in Indicates The minimum number of basic information words under the category, Indicates The maximum number of basic information words in each category; The relevant field feature words corresponding to the basic information feature category are extracted and counted, and the number of occurrences of the relevant field feature words is recorded as , calculate the correlation between different types of basic information features and related domain features, that is, the domain feature coefficient , the specific calculation formula is as follows: Based on machine learning methods and combined with file feature data, a mathematical model for intelligent file review is established; Based on the previous years' document review results of the institution to be reviewed as historical document data, the established document intelligent review mathematical model is optimized to obtain an optimized document intelligent review mathematical model; Compare and analyze the document review confidence assessment coefficient generated by the optimized document intelligent review mathematical model with the preset accurate confidence threshold, and process the abnormal analysis results. The document review confidence assessment coefficient is calculated based on the document text feature coefficient, the document structure feature coefficient and the domain feature coefficient, and combined with the preset error factor and the preset guess factor; The process of handling abnormal analysis results is continuously monitored and data analyzed to evaluate the results of the re-examination.

2. A method for intelligent file review according to claim 1, characterized in that: The specific calculation formula of the word frequency component is as follows: The specific calculation formula of the inverse document frequency is as follows: The specific calculation formula of the file text characteristic coefficient is as follows: In the formula, is the number of times the subject word appears in the i-th document to be reviewed, represents the i-th file to be reviewed, where , is a positive integer; is the word frequency component, is the inverse document frequency, is the document text feature coefficient, are all word vectors that appear in the document to be reviewed.

3. A method for intelligent file review according to claim 2, characterized in that: The process of handling abnormal analysis results is to continuously detect and analyze data and evaluate the results of the re-examination, specifically: Re-examine the data set for abnormal analysis results, and collect several file review confidence assessment coefficients during the re-examination to establish a data set Calculate the standard deviation of the document review confidence assessment coefficient using the document review confidence assessment coefficient in the data set and the average of the document review confidence assessment coefficients , and the standard deviation of the document review confidence assessment coefficient and the average of the document review confidence assessment coefficients The standard deviation reference thresholds are set and a preset confidence threshold Conduct comparative analysis; like , then issue a warning and review again; like and , then continuous testing is performed; like and , then re-examine to confirm that it is correct.

4. A device using the intelligent file review method according to any one of claims 1 to 3, characterized in that: It includes a file feature data collection module, a file feature data analysis module, a file review model optimization module, a review file evaluation module and a re-file review module, and there are connections between the modules; A document feature data collection module, which is used to collect documents to be reviewed based on the document information system of the institution to be reviewed and extract document feature data therefrom; The file feature data analysis module is used to establish a mathematical model for intelligent file review based on machine learning methods and combined with file feature data; The document review model optimization module is used to optimize the established document intelligent review mathematical model based on the document review results of the institution to be reviewed in previous years as historical document data, so as to obtain the optimized document intelligent review mathematical model; A review file assessment module is used to compare and analyze the file review confidence assessment coefficient generated by the optimized file intelligent review mathematical model with a preset accurate confidence threshold, and process abnormal analysis results. The file review confidence assessment coefficient is calculated based on the file text feature coefficient, the file structure feature coefficient and the domain feature coefficient, combined with a preset error factor and a preset guessing factor; The re-file review module is used to continuously detect and analyze data during the process of handling abnormal analysis results and evaluate the results of the re-review.

5. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute a document intelligent review method as described in any one of claims 1 to 3.

Images