Cross-platform-based data security circulation supervision system and method

Abstract

The invention discloses a cross-platform-based data security circulation supervision system and method, and relates to the technical field of data security, and the method comprises the steps: obtaining a primary review record which classifies an archive file into a corresponding archive category in history, extracting a primary review person, determining a review person which classifies the archive file again, and carrying out the classification again to obtain a review record; performing relation extraction on the content abstract and the archive description of the archive file to obtain a reference set corresponding to each archive category; establishing neural network models, and training each neural network model according to the file description and the content abstract; and extracting the archive category, the archive description and the content abstract corresponding to the archive file to be detected, and judging whether to carry out early warning prompt on the archive category to which the archive file to be detected belongs. According to the method, whether the classification of the archive files is reasonable or not is judged by summarizing the rules of the historical operation records, so that the situation of wrong classification of the archives is reduced, and the data security and reliability are improved.

Description

Technical Field

[0001] The present invention relates to the field of data security technology, and in particular to a cross-platform data security circulation supervision system and method. Background Art

[0002] With the advent of the information age, big data has become an important tool for social research. For enterprises, the construction of a big data platform can not only safeguard the development of the enterprise, but also improve the overall management level and capabilities of the enterprise. At present, with the growth of data volume and the increase in the demand for data value mining, enterprises will generate a large number of archival files in the production process. In order to better manage archival files, they usually choose a cross-platform approach to transplant and classify archival files. Reasonable archival file classification management helps to ensure data security in a cross-platform environment.

[0003] Currently, when archival files are transplanted and classified across platforms, manual classification or semi-automatic classification based on simple rules is common. Faced with a large number of archival files on different platforms, this is not only inefficient and highly subjective, but also leads to data misclassification, increasing the risk of data leakage. For example, improper classification may allow archival files to be accessed by unauthorized personnel, resulting in problems such as the inability to guarantee data security and incomplete data security mechanisms. Summary of the Invention

[0004] The purpose of the present invention is to provide a cross-platform data security circulation supervision system and method to solve the problems raised in the prior art.

[0005] To achieve the above object, the present invention provides the following technical solutions: The cross-platform data security circulation supervision method includes the following steps: Step S100: Obtaining historical operation records for classifying archive files into corresponding archive categories as preliminary review records, extracting preliminary reviewers corresponding to the preliminary review records, and determining reviewers who reclassify the archive files from the classification personnel other than the preliminary reviewers, and recording the reviewers' operation records for reclassifying the archive files as review records; Step S200: extracting the content summary of the archive file and the archive description written by the initial reviewer and the reviewer after classifying the archive file, and extracting the relationship between them to obtain the entities and association relationships; and obtaining the benchmark set corresponding to each archive category based on the entities and association relationships in the operation record corresponding to each archive file; Step S300: establishing a corresponding neural network model for each file category, and training each neural network model based on the file description and content summary corresponding to each file category to obtain a trained neural network model; Step S400: Extract the archive category, archive description and content summary corresponding to the archive file to be detected, and obtain the entity and association relationship corresponding to the archive file to be detected. According to the benchmark set corresponding to the archive category and the neural network model, determine whether to issue an early warning prompt for the archive category to which the archive file to be detected belongs.

[0006] Furthermore, step S100 includes: Step S110: Acquire several historical preliminary review records, which are records of cross-platform file transfers and initial file classification. Extract the preliminary reviewer number and file category automatically recorded by the computer program, as well as the file description manually recorded by the preliminary reviewer, from each preliminary review record. The file description is a record of the preliminary reviewer summarizing the content of the file and explaining the reasons for the classification after the preliminary reviewer completes the classification. Step S120: If an archive file that has been classified into archive categories is found to have an archive category classification error during the subsequent detection process, the archive file will be treated as an abnormal file; set the initial reviewer corresponding to a certain archive file F to be A, and according to the total number N of initial review records corresponding to another classification person B B , and the number of abnormal files N0, the first degree coefficient T1 of the classification personnel B as the reviewer of the file F is obtained = (N B -N0) / N B , and then obtain the first degree coefficient of each category of personnel except the initial reviewer A; Extract all keywords from the content summary of archive file F to generate a keyword set; obtain preliminary review records of several archive files corresponding to classifier B that are not abnormal archives, extract keywords from the archive description in each preliminary review record, and comprehensively obtain the number of occurrences of each keyword to generate a vocabulary list; add the number of occurrences of words in the vocabulary list that are identical to the keywords in the keyword set to obtain the total word count of classifier B; obtain the total word count of each other classifier, normalize it, and use the normalized result as the second degree coefficient T2 of classifier B as a reviewer; Step S130: The total compliance degree of the classifier B as the reviewer is obtained as T B =W1×T1+W2×T2, where W1 and W2 are the first weight and the second weight respectively; obtain the total compliance degree of each classification personnel, and take the classification personnel with the largest total compliance degree as the review personnel; and reclassify the file file F by the review personnel.

[0007] Furthermore, step S200 includes: Step S210: Obtain the file description X corresponding to the preliminary review record of a certain file file F A, File description corresponding to the review record X B And content summary X, if the archive file F has the same archive category in the initial review and the review, get the archive description X A 、File Description X B and all entities and relationships between entities in the content summary X, and establish archive description X respectively A 、File Description X B The knowledge graphs K1, K2, and K3 corresponding to the content summary X are extracted, and all entity-relationship-entity element combinations are extracted; Step S220: Get a certain element combination M1(E1 in K1 M 1,R M 1,E2 M 1) and a combination of elements in K2 M2(E1 M 2,R M 2,E2 M 2), E1 M 1. E2 M 1. E1 M 2 and E2 M 2 are entities, R M 1 and R M 2 are all related relationships, get E1 respectively M 1 and E1 M 2. R M 1 and R M 2. E2 M 1 and E2 M The word similarity S between 2 E1 、S R 、S E2 , calculate the combination similarity X between combinations M1 and M2 M =S R (S E1 +S E2 ); if X M If the value is greater than a preset first threshold, the combination M1 is first labeled to obtain all first-labeled combinations in the knowledge graph K1; then, the element combinations in the knowledge graphs K1 and K3 are compared to obtain all second-labeled combinations in the knowledge graph K1; and the element combinations that are both first-labeled and second-labeled combinations are taken as the target combinations of the archive file F; Relationship extraction and knowledge graph construction are both existing technical means in the field of large language models and will not be described in detail here; "entity-association relationship-entity" is also a common basic unit in the knowledge graph. In this scheme, it is regarded as an element combination, and the specific extraction method can be obtained through existing technology; the calculation of the combination similarity is determined based on all the entities and association relationships therein, which helps to improve the reliability of the combination similarity calculation; and the target combination obtained based on the first and second tag combinations is determined by the file description and content summary. The more target combinations there are, the greater the similarity and correlation between the file description and the content summary, that is, the more reliable the content record in the manually recorded file description, the more convincing the classification reason for the file category in the file description, and the more reliable the target combination as the classification basis for the corresponding file category, which explains the rationality of obtaining the following benchmark set.

[0008] Step S230: For each archive category, a baseline set with all elements being empty is established; based on the target combination and archive category corresponding to each archive file with the same archive category as those obtained in the initial review and the review, the target combination is added to the baseline set of the corresponding archive category, thereby obtaining a baseline set corresponding to each archive category; Furthermore, step S300 includes: taking all file descriptions and content summaries corresponding to a certain file category as relevant information, and all file descriptions and content summaries corresponding to other file categories as non-relevant information, setting the relevance of all relevant information to a certain file category to 1, and setting the relevance of non-relevant information to a certain file category to 0, and then taking a number of relevant information and non-relevant information, as well as their respective corresponding relevance levels as data sets, and substituting them into the neural network model corresponding to a certain file category for training to obtain a trained neural network model.

[0009] Furthermore, step S400 includes: Step S410: The archive file to be detected is a file that has completed cross-platform transmission and completed the initial archive category classification. The archive category of the archive file to be detected is C0. The archive description and content summary of the preliminary review record of the archive file to be detected are extracted, and all element combinations therein are obtained. The number of element combinations is Y. The combination similarity between the yth element combination and each combination in the benchmark set of archive category C0 is calculated. The maximum value of the combination similarity is used as the target value of the yth element combination, and the target value of each element combination is obtained, thereby obtaining the category fit degree of the archive file to be detected. , where e is the natural logarithm, q is the adjustment factor coefficient, and Sy is the target value of the y-th element combination; It should be noted that the function h=1-e -xWhen x is x≥0, h is [0,1), which is a function in which h increases as x increases. When x is small, the increase in h is larger than when x is large. In this scheme, the target value is determined based on the combination of the archive file to be detected and the benchmark set. When the target value is small, it is sufficient to indicate that the category fit between the two is large, so the function h=1-e is used here. -x The design is carried out, and the adjustment factor coefficient q is used as the adjustment factor for the degree of category fit, and its specific value should be determined according to actual conditions.

[0010] Step S420: Substitute the file description and content summary of the preliminary review record of the file to be tested into each neural network model to obtain the degree of correlation between the file to be tested and each file category. If the category fit degree Z is less than the preset degree threshold or the correlation degree of the file category C0 is not the maximum, an early warning is issued for the file category to which the file to be tested belongs, and relevant personnel are prompted to handle the matter. In this scheme, when a certain archive file is classified across platforms, the basis for judging whether the classification is reasonable is whether it conforms to the category and whether it is the most appropriate compared with other categories.

[0011] A cross-platform data security circulation supervision system, including a reviewer determination module, a benchmark set establishment module, a neural network model training module, and an early warning prompt module; Reviewer determination module: used to obtain historical operation records of classifying archive files into corresponding archive categories as preliminary review records, extract the preliminary reviewers corresponding to the preliminary review records, and determine the reviewers who will reclassify the archive files from the classification personnel other than the preliminary reviewers, and use the reviewers' operation records of reclassifying the archive files as review records; Benchmark set establishment module: This module is used to extract the content summaries of archival documents and the archival descriptions written by the initial reviewers and reviewers after classifying the archival documents, and perform relationship extraction to obtain the entities and associations therein; and obtain the benchmark set corresponding to each archival category based on the entities and associations in the operation records corresponding to each archival document; Neural network model training module: used to establish a corresponding neural network model for each file category, train each neural network model according to the file description and content summary corresponding to each file category, and obtain a trained neural network model; Early warning prompt module: used to extract the archive category, archive description and content summary corresponding to the archive file to be detected, and obtain the entities and association relationships corresponding to the archive file to be detected. According to the benchmark set corresponding to the archive category and the neural network model, it determines whether to issue an early warning prompt for the archive category to which the archive file to be detected belongs.

[0012] Furthermore, the reviewer determination module includes a preliminary review record analysis unit, a degree coefficient calculation unit, and a reviewer determination unit; Initial review record analysis unit: used to obtain several historical initial review records, extract the initial reviewer number and file category automatically recorded by the computer program in each initial review record, and the file description manually recorded by the initial reviewer; Degree coefficient calculation unit: used to set abnormal files; obtain the total number of preliminary review records corresponding to the classification personnel, and the number of files among them that are abnormal files, to obtain the first degree coefficient of the classification personnel as the file review personnel; extract all keywords in the content summary of the file to generate a keyword set; obtain the preliminary review records of several files corresponding to the classification personnel that are not abnormal files, and combine and analyze them to obtain the second degree coefficient of the classification personnel as the review personnel; Reviewer determination unit: used to obtain the total compliance level of the classified personnel as reviewers, and according to the total compliance level of each classified personnel, the classified personnel with the largest total compliance level will be used as the reviewer.

[0013] Furthermore, the early warning prompt module includes a category fit degree calculation unit and an early warning prompt unit; Category fit calculation unit: used to extract the file description and content summary of the preliminary review record of the file to be tested, and obtain the category fit of the file to be tested; Early warning prompt unit: used to determine whether to issue an early warning prompt for the archive category to which the archive file to be detected belongs based on the neural network model corresponding to the archive category and the degree of category fit.

[0014] Compared with the prior art, the present invention has the following beneficial effects: the present invention provides a cross-platform data security circulation supervision system and method, including: obtaining historical preliminary review records that classify archival files into corresponding archival categories, extracting preliminary review personnel, determining review personnel who reclassify the archival files into archival categories, and reclassifying to obtain review records; extracting relationships between the content summary and archival description of the archival files to obtain a benchmark set corresponding to each archival category; establishing a neural network model and training each neural network model based on the archival description and content summary; extracting the archival category, archival description and content summary corresponding to the archival files to be detected, and determining whether to issue an early warning prompt for the archival category to which the archival files to be detected belong. The present invention summarizes the rules of historical operation records and determines whether the classification of archival files is reasonable, which helps to reduce the occurrence of archival misclassification and improves data security and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 Schematic diagram of the process of the cross-platform data security circulation supervision method of the present invention; Figure 2 This is a structural diagram of the cross-platform data security circulation supervision system of the present invention. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0017] Example: Figure 1 As shown, the present invention provides a technical solution for a cross-platform data security circulation supervision method, comprising the following steps: Step S100: Obtaining historical operation records for classifying archive files into corresponding archive categories as preliminary review records, extracting preliminary reviewers corresponding to the preliminary review records, and determining reviewers who reclassify the archive files from the classification personnel other than the preliminary reviewers, and recording the reviewers' operation records for reclassifying the archive files as review records; Step S110: Acquire several historical preliminary review records, which are records of cross-platform file transfers and initial file classification. Extract the preliminary reviewer number and file category automatically recorded by the computer program, as well as the file description manually recorded by the preliminary reviewer, from each preliminary review record. The file description is a record of the preliminary reviewer summarizing the content of the file and explaining the reasons for the classification after the preliminary reviewer completes the classification. Step S120: If an archive file that has been classified into archive categories is found to have an archive category classification error during the subsequent detection process, the archive file will be treated as an abnormal file; set the initial reviewer corresponding to a certain archive file F to be A, and according to the total number N of initial review records corresponding to another classification person B B , and the number of abnormal files N0, the first degree coefficient T1 of the classification personnel B as the reviewer of the file F is obtained = (N B -N0) / N B , and then obtain the first degree coefficient of each category of personnel except the initial reviewer A; Extract all keywords from the content summary of archive file F to generate a keyword set; obtain preliminary review records of several archive files corresponding to classifier B that are not abnormal archives, extract keywords from the archive description in each preliminary review record, and comprehensively obtain the number of occurrences of each keyword to generate a vocabulary list; add the number of occurrences of words in the vocabulary list that are identical to the keywords in the keyword set to obtain the total word count of classifier B; obtain the total word count of each other classifier, normalize it, and use the normalized result as the second degree coefficient T2 of classifier B as a reviewer; The first degree coefficient is determined based on the historical classification of the classifier, which is equivalent to the reliability of the classification by the classifier. The larger the first degree coefficient is, the better the classifier is as a reviewer, but it should also be determined based on the second degree coefficient. The second degree coefficient is determined by the current archival file and the archival file historically classified by the classifier. Generally speaking, the greater the similarity between the current archival file and the archival file historically classified by the classifier, the more experience the classifier has in classifying such archival files. The more such experience is, the more authoritative the classifier is in classifying archival files. By comprehensive comparison, a reviewer can be obtained.

[0018] Step S130: The total compliance degree of the classifier B as the reviewer is obtained as T B =W1×T1+W2×T2, where W1 and W2 are the first weight and the second weight respectively; obtain the total compliance degree of each classification personnel, and take the classification personnel with the largest total compliance degree as the review personnel; and reclassify the file file F by the review personnel.

[0019] Step S200: Extract the content summary of the archive file and the archive description written by the initial reviewer and the reviewer after classifying the archive file, and extract the relationships to obtain the entities and association relationships therein; and obtain the benchmark set corresponding to each archive category based on the entities and association relationships in the operation records corresponding to each archive file.

[0020] Step S210: Obtain the file description X corresponding to the preliminary review record of a certain file file F A , File description corresponding to the review record X B And content summary X, if the archive file F has the same archive category in the initial review and the review, get the archive description X A 、File Description X B and all entities and relationships between entities in the content summary X, and establish archive description X respectively A 、File Description X B The knowledge graphs K1, K2, and K3 corresponding to the content summary X are extracted, and all entity-relationship-entity element combinations are extracted; Step S220: Get a certain element combination M1(E1 in K1 M 1,R M 1,E2 M 1) and a combination of elements in K2 M2(E1 M 2,R M 2,E2 M 2), E1 M 1. E2 M 1. E1 M 2 and E2 M 2 are entities, R M 1 and R M 2 are all related relationships, get E1 respectively M 1 and E1 M 2. R M 1 and R M 2. E2 M 1 and E2 M The word similarity S between 2 E1 、S R 、S E2 , calculate the combination similarity X between combinations M1 and M2 M =S R (S E1 +S E2 ); if X M If the value is greater than a preset first threshold, the combination M1 is first labeled to obtain all first-labeled combinations in the knowledge graph K1; then, the element combinations in the knowledge graphs K1 and K3 are compared to obtain all second-labeled combinations in the knowledge graph K1; and the element combinations that are both first-labeled and second-labeled combinations are taken as the target combinations of the archive file F; By analogy with the element combinations in knowledge graphs K1 and K3, all the second tag combinations in knowledge graph K1 are obtained as follows: Get a combination of elements M1(E1 in K1 M 1,R M 1,E2 M 1) and an element combination M3(E1 in K3 M 3,R M 3,E2 M 3), E1 M 1. E2 M 1. E1 M 3 and E2 M 3 are all entities, R M 1 and R M 3 are all related relationships, get E1 respectively M 1 and E1 M 3. R M 1 and R M 3. E2M 1 and E2 M The word similarity S between 3 E3 、S R1 、S E4 , calculate the combination similarity X between combinations M1 and M3 M1 =S R1 (S E3 +S E4 ); if X M1 If it is greater than a preset second threshold, the combination M1 is second-marked to obtain all second-marked combinations in the knowledge graph K1.

[0021] Step S230: Establish a baseline set with all elements empty for each archive category; according to the target combination and archive category corresponding to each archive file with the same archive category after preliminary review and review, add the target combination to the baseline set of the corresponding archive category, and thus obtain the baseline set corresponding to each archive category.

[0022] Step S300: establishing a corresponding neural network model for each file category, and training each neural network model based on the file description and content summary corresponding to each file category to obtain a trained neural network model; All file descriptions and content summaries corresponding to a certain file category are regarded as relevant information, and all file descriptions and content summaries corresponding to other file categories are regarded as irrelevant information. The relevance of all relevant information to a certain file category is set to 1, and the relevance of irrelevant information to a certain file category is set to 0. Then, several relevant information and irrelevant information, as well as their corresponding relevance levels, are used as data sets and substituted into the neural network model corresponding to a certain file category for training to obtain the trained neural network model.

[0023] The input of the trained neural network model is the archive description and content summary, and the output is the correlation between the archive file and the archive category corresponding to the neural network model. The correlation value range is [0,1].

[0024] Step S400: Extract the archive category, archive description and content summary corresponding to the archive file to be detected, and obtain the entity and association relationship corresponding to the archive file to be detected. According to the benchmark set corresponding to the archive category and the neural network model, determine whether to issue an early warning prompt for the archive category to which the archive file to be detected belongs.

[0025] Step S410: The archive file to be detected is a file that has completed cross-platform transmission and completed the initial archive category classification. The archive category of the archive file to be detected is C0. The archive description and content summary of the preliminary review record of the archive file to be detected are extracted, and all element combinations therein are obtained. The number of element combinations is Y. The combination similarity between the yth element combination and each combination in the benchmark set of archive category C0 is calculated. The maximum value of the combination similarity is used as the target value of the yth element combination, and the target value of each element combination is obtained, thereby obtaining the category fit degree of the archive file to be detected. , where e is the natural logarithm, q is the adjustment factor coefficient, and Sy is the target value of the y-th element combination; Step S420: Substitute the archive description and content summary of the preliminary review record of the archive file to be detected into each neural network model to obtain the degree of correlation between the archive file to be detected and each archive category. If the category fit degree Z is less than the preset degree threshold or the correlation degree of the archive category C0 is not the maximum, an early warning will be issued for the archive category to which the archive file to be detected belongs, and relevant personnel will be prompted to handle it.

[0026] This solution also provides a cross-platform data security circulation supervision system, as shown in the attached Figure 2 As shown, including: Reviewer determination module: used to obtain historical operation records of classifying archive files into corresponding archive categories as preliminary review records, extract the preliminary reviewers corresponding to the preliminary review records, and determine the reviewers who will reclassify the archive files from the classification personnel other than the preliminary reviewers, and use the reviewers' operation records of reclassifying the archive files as review records; Benchmark set establishment module: This module is used to extract the content summaries of archival documents and the archival descriptions written by the initial reviewers and reviewers after classifying the archival documents, and perform relationship extraction to obtain the entities and associations therein; and obtain the benchmark set corresponding to each archival category based on the entities and associations in the operation records corresponding to each archival document; Neural network model training module: used to establish a corresponding neural network model for each file category, train each neural network model according to the file description and content summary corresponding to each file category, and obtain a trained neural network model; Early warning prompt module: used to extract the archive category, archive description and content summary corresponding to the archive file to be detected, and obtain the entities and association relationships corresponding to the archive file to be detected. According to the benchmark set corresponding to the archive category and the neural network model, it determines whether to issue an early warning prompt for the archive category to which the archive file to be detected belongs.

[0027] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above and that the invention can be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. A cross-platform data security circulation supervision method, characterized by: The following steps are involved: Step S100: Obtaining historical operation records for classifying archive files into corresponding archive categories as preliminary review records, extracting preliminary reviewers corresponding to the preliminary review records, and determining reviewers who reclassify the archive files from the classification personnel other than the preliminary reviewers, and recording the reviewers' operation records for reclassifying the archive files as review records; Step S200: extracting the content summary of the archive file and the archive description written by the initial reviewer and the reviewer after classifying the archive file, and extracting the relationship between them to obtain the entities and association relationships; and obtaining the benchmark set corresponding to each archive category based on the entities and association relationships in the operation record corresponding to each archive file; Step S300: establishing a corresponding neural network model for each file category, and training each neural network model based on the file description and content summary corresponding to each file category to obtain a trained neural network model; Step S400: Extract the archive category, archive description and content summary corresponding to the archive file to be detected, and obtain the entity and association relationship corresponding to the archive file to be detected. According to the benchmark set corresponding to the archive category and the neural network model, determine whether to issue an early warning prompt for the archive category to which the archive file to be detected belongs.

2. The cross-platform data security circulation supervision method according to claim 1 is characterized in that: Step S100 includes: Step S110: Acquire several historical preliminary review records, which are records of cross-platform file transfers and initial file classification. Extract the preliminary reviewer number and file category automatically recorded by the computer program, as well as the file description manually recorded by the preliminary reviewer, from each preliminary review record. The file description is a record of the preliminary reviewer's summary of the content in the file and the explanation of the reasons for the classification after the preliminary reviewer completes the classification. Step S120: If an archive file that has been classified into archive categories is found to have an archive category classification error during the subsequent detection process, the archive file will be treated as an abnormal file; set the initial reviewer corresponding to a certain archive file F to be A, and according to the total number N of initial review records corresponding to another classification person B, B , and the number of abnormal files N0, the first degree coefficient T1 of the classification personnel B as the reviewer of the file F is obtained = (N B -N0) / N B , and then obtain the first degree coefficient of each category of personnel except the initial reviewer A; Extract all keywords from the content summary of archive file F to generate a keyword set; obtain preliminary review records of several archive files corresponding to classifier B that are not abnormal archives, extract keywords from the archive description in each preliminary review record, and comprehensively obtain the number of occurrences of each keyword to generate a vocabulary list; add the number of occurrences of words in the vocabulary list that are identical to the keywords in the keyword set to obtain the total word count of classifier B; obtain the total word count of each other classifier, normalize it, and use the normalized result as the second degree coefficient T2 of classifier B as a reviewer; Step S130: The total compliance degree of the classifier B as the reviewer is obtained as T B =W1×T1+W2×T2, where W1 and W2 are the first weight and the second weight respectively; obtain the total compliance degree of each classification personnel, and take the classification personnel with the largest total compliance degree as the review personnel; and reclassify the file file F by the review personnel.

3. The cross-platform data security circulation supervision method according to claim 1 is characterized in that: Step S200 includes: Step S210: Obtain the file description X corresponding to the preliminary review record of a certain file file F A , File description corresponding to the review record X B And content summary X, if the archive file F has the same archive category in the initial review and the review, get the archive description X A 、File Description X B and all entities and relationships between entities in the content summary X, and establish archive description X respectively A 、File Description X B The knowledge graphs K1, K2, and K3 corresponding to the content summary X are extracted, and all entity-relationship-entity element combinations are extracted; Step S220: Get a certain element combination M1(E1 in K1 M 1,R M 1,E2 M 1) and a combination of elements in K2 M2(E1 M 2,R M 2,E2 M 2), E1 M 1. E2 M 1. E1 M 2 and E2 M 2 are entities, R M 1 and R M 2 are all related relationships, get E1 respectively M 1 and E1 M 2. R M 1 and R M 2. E2 M 1 and E2 M The word similarity S between 2 E1 、S R 、S E2 , calculate the combination similarity X between combinations M1 and M2 M =S R (S E1 +S E2 ); if X M If the value is greater than a preset first threshold, the combination M1 is first labeled to obtain all first-labeled combinations in the knowledge graph K1; then, the element combinations in the knowledge graphs K1 and K3 are compared to obtain all second-labeled combinations in the knowledge graph K1; and the element combinations that are both first-labeled and second-labeled combinations are taken as the target combinations of the archive file F; Step S230: Establish a baseline set with all elements empty for each archive category; according to the target combination and archive category corresponding to each archive file with the same archive category after preliminary review and review, add the target combination to the baseline set of the corresponding archive category, and thus obtain the baseline set corresponding to each archive category.

4. The cross-platform data security circulation supervision method according to claim 3 is characterized in that: Step S300 includes: taking all file descriptions and content summaries corresponding to a certain file category as relevant information, and all file descriptions and content summaries corresponding to other file categories as non-relevant information, setting the relevance of all relevant information to the certain file category to 1, and setting the relevance of non-relevant information to the certain file category to 0, and then taking a number of relevant information and non-relevant information, as well as their respective corresponding relevance levels as a data set, and substituting them into a neural network model corresponding to a certain file category for training to obtain a trained neural network model.

5. The cross-platform data security circulation supervision method according to claim 4 is characterized in that: Step S400 includes: Step S410: The archive file to be detected is a file that has completed cross-platform transmission and completed the initial archive category classification. The archive category of the archive file to be detected is C0. The archive description and content summary of the preliminary review record of the archive file to be detected are extracted, and all element combinations therein are obtained. The number of element combinations is Y. The combination similarity between the yth element combination and each combination in the benchmark set of archive category C0 is calculated. The maximum value of the combination similarity is used as the target value of the yth element combination, and the target value of each element combination is obtained, thereby obtaining the category fit degree of the archive file to be detected. , where e is the natural logarithm, q is the adjustment factor coefficient, and Sy is the target value of the y-th element combination; Step S420: Substitute the archive description and content summary of the preliminary review record of the archive file to be detected into each neural network model to obtain the degree of correlation between the archive file to be detected and each archive category. If the category fit degree Z is less than the preset degree threshold or the correlation degree of the archive category C0 is not the maximum, an early warning will be issued for the archive category to which the archive file to be detected belongs, and relevant personnel will be prompted to handle it.

6. A data security circulation supervision system, configured to execute the cross-platform data security circulation supervision method according to any one of claims 1 to 5, characterized in that: The system includes a reviewer determination module, a benchmark set establishment module, a neural network model training module and an early warning prompt module; Reviewer determination module: used to obtain historical operation records of classifying archive files into corresponding archive categories as preliminary review records, extract the preliminary reviewers corresponding to the preliminary review records, and determine the reviewers who will reclassify the archive files from the classification personnel other than the preliminary reviewers, and use the reviewers' operation records of reclassifying the archive files as review records; Benchmark set establishment module: This module is used to extract the content summaries of archival documents and the archival descriptions written by the initial reviewers and reviewers after classifying the archival documents, and perform relationship extraction to obtain the entities and associations therein; and obtain the benchmark set corresponding to each archival category based on the entities and associations in the operation records corresponding to each archival document; Neural network model training module: used to establish a corresponding neural network model for each file category, train each neural network model according to the file description and content summary corresponding to each file category, and obtain a trained neural network model; Early warning prompt module: used to extract the archive category, archive description and content summary corresponding to the archive file to be detected, and obtain the entities and association relationships corresponding to the archive file to be detected. According to the benchmark set corresponding to the archive category and the neural network model, it determines whether to issue an early warning prompt for the archive category to which the archive file to be detected belongs.

7. The data security circulation supervision system according to claim 6, characterized in that: The reviewer determination module includes a preliminary review record analysis unit, a degree coefficient calculation unit and a reviewer determination unit; Initial review record analysis unit: used to obtain several historical initial review records, extract the initial reviewer number and file category automatically recorded by the computer program in each initial review record, and the file description manually recorded by the initial reviewer; Degree coefficient calculation unit: used to set abnormal files; obtain the total number of preliminary review records corresponding to the classification personnel, and the number of files among them that are abnormal files, to obtain the first degree coefficient of the classification personnel as the file review personnel; extract all keywords in the content summary of the file to generate a keyword set; obtain the preliminary review records of several files corresponding to the classification personnel that are not abnormal files, and combine and analyze them to obtain the second degree coefficient of the classification personnel as the review personnel; Reviewer determination unit: used to obtain the total compliance level of the classified personnel as reviewers, and according to the total compliance level of each classified personnel, the classified personnel with the largest total compliance level will be used as the reviewer.

8. The data security circulation supervision system according to claim 6, characterized in that: The early warning prompt module includes a category fit degree calculation unit and an early warning prompt unit; Category fit calculation unit: used to extract the file description and content summary of the preliminary review record of the file to be tested, and obtain the category fit of the file to be tested; Early warning prompt unit: used to determine whether to issue an early warning prompt for the archive category to which the archive file to be detected belongs based on the neural network model corresponding to the archive category and the degree of category fit.

Images