Document compliance audit data management system and method based on artificial intelligence

Abstract

The invention discloses a document compliance auditing data management system and method based on artificial intelligence, and relates to the technical field of document auditing, and the management method comprises the following steps: recording the compliance auditing process of any document, and generating a corresponding auditing log; performing auditing analysis on any auditing log, generating an auditing result and feeding back the auditing result; performing abnormal feature extraction and division on the corresponding documents according to preset feature types to obtain abnormal feature sets of the corresponding documents; analyzing association conditions among different features; simulating the causal relationship between the associated features to obtain effective causal feature pairs; generating an actual causal relationship based on an audit result fed back by any effective causal feature; abnormal features existing in the actual document auditing process are recognized, and abnormal traceability is conducted on the abnormal features; the subjectivity and omission risk of manual auditing are effectively reduced, and the auditing efficiency and accuracy are greatly improved.

Description

Technical Field

[0001] The present invention relates to the technical field of document review, and in particular to an artificial intelligence-based document compliance review data management system and method. Background Art

[0002] Currently, document compliance audits rely primarily on manual review, which is inefficient and prone to subjective bias and omissions. Traditional risk assessment methods are mostly based on experience and rules, making it difficult to capture the nonlinear causal relationships between complex risk factors. Existing data analysis technologies are mostly limited to correlation analysis, making it difficult to accurately identify causal relationships. For example, simple statistical analysis can reveal a correlation between document type and violations, but it cannot clearly indicate whether the document type is the cause of the violation or the influence of other factors. There is a lack of effective tools and methods to systematically identify and quantify the causal relationship between risk factors, and thus conduct precise risk control and prevention. Summary of the Invention

[0003] The purpose of the present invention is to provide an artificial intelligence-based document compliance review data management system and method to solve the problems raised in the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an artificial intelligence-based document compliance review data management method, the management method comprising the following steps: Step S100: Record the compliance review process of any document and generate a corresponding review log; conduct review and analysis on the document data in any review log, generate the review results of the corresponding document and provide feedback; Step S200: extracting corresponding documents from any audit log, extracting and classifying abnormal features of the corresponding documents according to preset feature types, and obtaining an abnormal feature set of the corresponding documents; Step S300: Based on the audit results of any audit log, analyze the correlation between different features; based on the correlation between each feature in different audit logs, simulate the causal relationship between the correlated features; Step S400: Analyze the causal simulation of any associated feature in any audit log to obtain valid causal feature pairs; based on the audit results of any valid causal feature pairs in different audit logs, generate the actual causal relationship between each feature; Step S500: Identify abnormal features existing in the actual document review process, and trace the abnormal features based on the actual causal relationship between the abnormal features.

[0005] Furthermore, step S100 includes the following steps: Step S101: pre-configuring an audit policy database and a feature attribute library. The audit policy database pre-stores a plurality of audit policies, wherein any audit policy matches a corresponding feature attribute set. The feature attribute library pre-stores a plurality of feature attributes, wherein any feature attribute matches a feature data set. The audit policies included in the audit policy database include regulatory compliance policies, data accuracy policies, and policy compliance policies. For example, audit policies in financial reports may include financial statement integrity policies, tax compliance policies, and disclosure compliance policies. Feature attributes are keywords corresponding to various items in the audit policies, such as operating income, cost, and expenses. Step S102: Whenever a document to be reviewed is received, entity recognition and feature extraction are performed on the document data in the document to be reviewed using natural language processing technology. The recognition area of ​​each entity in the document to be reviewed is extracted, and the recognition area of ​​any feature is also extracted. If the recognition area of ​​a feature is within the recognition area of ​​a certain entity, the entity is matched with the feature to generate an entity data set for each entity. The document to be reviewed is divided into several entity data sets, and an audit log is generated. Step S103: arbitrarily select an audit strategy for generating an audit log, extract a feature attribute set and a feature data set that match the selected audit strategy, compare any feature attribute with the entity of the document to be audited, and compare any entity data with the features of the document to be audited. If all feature attributes and corresponding feature data in the selected audit strategy can be extracted from the document to be audited, then the selected audit strategy is set as the target audit strategy. Step S104: Retrieve each target audit strategy to conduct compliance audit on the document to be audited. If there is an abnormality in the selected feature, mark the selected feature as abnormal. Obtain the audit time point, audit strategy and abnormal mark of any selected feature to obtain the audit information group of the selected feature. Summarize the audit information groups of each feature in the document to be audited to obtain the audit result of the document to be audited and store it in the generated audit log.

[0006] Furthermore, step S200 includes the following steps: Step S201: randomly selecting an audit log and extracting the audit information group of each feature in the selected audit log; randomly selecting a feature, if there is an abnormality mark in the audit information group of the selected feature, then setting the selected feature as an abnormal feature; Step S202: Obtain matching entities of any abnormal feature in the selected audit log, divide all abnormal features into a number of abnormal feature sets according to the matching entities; arbitrarily select one abnormal feature set, extract feature attributes of the entity in the selected abnormal feature set, and set the extracted feature attributes as the abnormal feature in the selected audit log; Step S203: Extract the audit time point and audit strategy from the audit information group of each feature, sort the features in the order of the audit time points, and obtain the audit time interval between any two adjacent features; if the audit strategies adopted by two adjacent features are different, set the audit time interval of the two adjacent features as the reference time interval, and calculate the average value of all reference time intervals to obtain the average time interval; by analyzing the conversion time between different audit strategies, compare the conversion time with the audit time interval in the same audit strategy, and determine whether the features belong to the same batch under the same audit strategy. Only features from different batches will have a causal relationship in the future; Step S204: If there are several adjacent and continuous features with the same audit strategy, the several features are preliminarily divided into similar feature sets, and all features are divided into several similar feature sets; any similar feature set is selected, and the audit time interval between two adjacent features in the selected similar feature set is obtained. If the audit time interval exceeds the average time interval, the selected similar feature set is divided into two feature sets based on the two adjacent features, and several feature sets are regenerated; Step S205: arbitrarily select an abnormal feature, determine the key set where the selected abnormal feature is located, and divide all abnormal features into several abnormal feature sets corresponding to the selected audit log documents according to the distribution of each abnormal feature in each feature set.

[0007] Furthermore, step S300 includes the following steps: Step S301: arbitrarily select an audit log, extract several abnormal feature sets from the selected audit log, sort the abnormal feature sets according to the chronological order of the audit time points, and arbitrarily select two abnormal feature sets, and select one abnormal feature from each of the two abnormal feature sets as two comparison features; Step S302: Pre-build a semantic association database, match the semantic associations between each feature attribute in the feature attribute library, obtain the association value between any two feature attributes, and if the association value between the two feature attributes exceeds a preset association threshold, then set the two feature attributes to have an association relationship; obtain the feature attributes corresponding to the two compared features, and if there is an association relationship between the two compared features, set the two selected abnormal features as an associated feature group, and obtain a number of associated feature groups for the selected audit log; Step S303: arbitrarily select a correlation feature group, adjust the two abnormal features in the selected correlation feature group according to the order of the audit time points, and generate a sequential correlation feature group; extract all sequential correlation feature groups in different audit logs, and arbitrarily select two sequential correlation feature groups. If the audit strategies adopted by the two sequential correlation feature groups are the same, the two sequential key feature groups are set as similar feature groups, and several similar feature groups are summarized to generate a similar feature group set. The previous abnormal feature and the next abnormal feature of each similar feature group in the similar feature group set are summarized and generated to generate a first feature set and a second feature set, and a causal relationship is set between the first feature set and the second feature set; to determine whether there is a causal relationship, it is necessary to determine whether the two abnormal features have abnormal situations in multiple audit logs at the same time.

[0008] Furthermore, step S400 includes the following steps: Step S401: arbitrarily select a similar feature group set to obtain a first feature set and a second feature set of the selected similar feature group set, and arbitrarily select an abnormal feature from each of the first feature set and the second feature set to generate a simulated feature group; Step S402: Randomly select an audit log, obtain the audit information group of each feature in the selected audit log, compare the two abnormal features in the simulated adjustment group with each feature, and if there are two features in the selected audit log that are respectively the same as the two abnormal features, obtain the audit information group of the two features. If the audit information groups of the two features both contain abnormal marks, and the order of the audit time points and the audit strategies adopted are also the same, then set the simulated feature group as a valid causal feature pair; Step S403: extract all valid causal feature pairs from the selected audit log, and arbitrarily select a group of valid causal feature pairs. Compare the selected valid causal feature pairs with the features in the remaining audit logs. If the selected valid causal feature pair also exists in the remaining audit logs, the selected valid causal feature pair is set as the actual causal feature pair. If the actual causal feature pair requires an exception in all audit logs, it means that there is an actual causal relationship. Step S404: arbitrarily select two actual causal feature pairs. If the latter feature in one actual causal feature pair and the former feature in the other actual causal feature pair generate a new actual causal feature pair, then the two selected actual causal feature pairs are connected to obtain a causal feature traceability chain, and the actual causal feature pairs are connected to each other to generate a causal feature traceability network.

[0009] Furthermore, step S500 includes the following steps: Step S501: Real-time monitoring of the document review process, entity recognition and feature extraction of document data in the real-time document, comparison of the review policy database and the feature attribute database, and generation of review information groups for each feature in the real-time document; obtaining abnormality flags for each feature of the review information group, and obtaining a number of abnormal features of the real-time document; Step S502: arbitrarily select two abnormal features from the plurality of abnormal features, extract the audit time points in the audit information group of the two selected abnormal features, generate the expected causal feature group according to the order of the audit time points, and obtain the expected causal feature groups of the real-time document. Step S503: The causal feature traceability network is retrieved and compared with the plurality of expected causal feature groups. If the plurality of expected causal feature groups are continuously connected in the causal feature traceability network, an expected traceability chain is generated, the traceability features of each abnormal feature are obtained, and an abnormal repair reminder is issued for each traceability feature. The abnormal feature recognition is performed again on the repaired real-time document until the real-time document is reviewed.

[0010] In order to better implement the above method, a document compliance audit data management system is also proposed. The management system includes a document data acquisition module, a document feature recognition module, a causal association simulation module, a causal association generation module and an audit anomaly capture module; The document data collection module is used to record the compliance review process of any document and generate the corresponding review log; it reviews and analyzes the document data in any review log, generates the review results of the corresponding document, and provides feedback; The document feature recognition module is used to extract the corresponding documents of any audit log, extract and classify the abnormal features of the corresponding documents according to the preset feature types, and obtain the abnormal feature set of the corresponding documents; The causal association simulation module is used to analyze the correlation between different features based on the audit results of any audit log; based on the correlation between each feature in different audit logs, it simulates the causal relationship between the related features; The causal association generation module is used to analyze the causal simulation of any associated features in any audit log to obtain valid causal feature pairs; based on the audit results of any valid causal feature pairs in different audit logs, it generates the actual causal relationship between each feature; The audit anomaly capture module is used to identify abnormal features that exist in the actual document audit process and trace the abnormal features based on the actual causal relationship between the abnormal features.

[0011] Furthermore, the document data collection module includes a data collection and processing unit and an audit result feedback unit; The data collection and processing unit is used to record the compliance review process of any document and generate the corresponding audit log; the audit result feedback unit is used to review and analyze the document data in any audit log, generate the audit results of the corresponding document and provide feedback.

[0012] Furthermore, the causal association simulation module includes a feature association analysis unit and a causal relationship simulation unit; The feature correlation analysis unit is used to analyze the correlation between different features based on the audit results of any audit log; the causal relationship simulation unit is used to simulate the causal relationship between related features based on the correlation between various features in different audit logs.

[0013] Furthermore, the causal association generation module includes a relationship simulation verification unit and a feature causal generation unit; The relationship simulation verification unit is used to analyze the causal simulation of any associated feature in any audit log to obtain a valid causal feature pair; the feature causal generation unit is used to generate the actual causal relationship between each feature based on the audit results fed back by any valid causal feature pair in different audit logs.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. This application can effectively reduce the subjectivity and omission risk of manual review by automatically recording, extracting features, and identifying abnormal features during the document review process, significantly improving review efficiency and accuracy, and also enhancing the intelligence level of document compliance review. 2. Through the causal reasoning mechanism, the present invention can mine the causal relationship between features from multi-dimensional and multi-log data, thereby constructing a causal feature traceability network, which can achieve systematic identification and tracing of the root causes of anomalies, providing more in-depth and scientific decision-making support for risk prevention and control; 3. The present invention helps staff control the accuracy of document compliance audits and improve audit efficiency by monitoring abnormal features in real time during the audit process and providing intelligent reminders and repair suggestions based on historical causal networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 Schematic diagram of steps for an AI-based document compliance review data management approach; Figure 2 This is a structural diagram of the AI-based document compliance review data management system. 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: Figures 1 to 2 As shown, the present invention provides a document compliance review data management method based on artificial intelligence, and the management method includes the following steps: Step S100: Record the compliance review process of any document and generate a corresponding review log; conduct review and analysis on the document data in any review log, generate the review results of the corresponding document and provide feedback; Wherein, step S100 includes the following steps: Step S101: pre-configuring an audit policy database and a feature attribute library, wherein the audit policy database pre-stores a plurality of audit policies, wherein any audit policy is matched with a corresponding feature attribute set, and the feature attribute library pre-stores a plurality of feature attributes, wherein any feature attribute is matched with a feature data set; Step S102: Whenever a document to be reviewed is received, entity recognition and feature extraction are performed on the document data in the document to be reviewed using natural language processing technology. The recognition area of ​​each entity in the document to be reviewed is extracted, and the recognition area of ​​any feature is also extracted. If the recognition area of ​​a feature is within the recognition area of ​​a certain entity, the entity is matched with the feature to generate an entity data set for each entity. The document to be reviewed is divided into several entity data sets, and an audit log is generated. Step S103: arbitrarily select an audit strategy for generating an audit log, extract a feature attribute set and a feature data set that match the selected audit strategy, compare any feature attribute with the entity of the document to be audited, and compare any entity data with the features of the document to be audited. If all feature attributes and corresponding feature data in the selected audit strategy can be extracted from the document to be audited, then the selected audit strategy is set as the target audit strategy. Step S104: Retrieve each target audit strategy and conduct compliance audit on the document to be audited. If the selected feature has an anomaly, mark the selected feature as an anomaly. Obtain the audit time point, audit strategy, and anomaly mark for any selected feature to obtain an audit information group for the selected feature. Summarize the audit information groups for each feature in the document to be audited to obtain the audit result of the document to be audited and store it in the generated audit log. Example 1: Set up a compliance audit of an enterprise's annual financial report. The pre-configured audit policy database includes a financial statement integrity policy, a tax compliance policy, and a disclosure compliance policy. Among them, the financial statement integrity policy matches a feature attribute set including operating income, net profit, etc. The system uses NLP technology to identify the existence of entities such as "operating income" and "net profit" in the report, which are all the same as the feature attribute set of the financial statement integrity policy. The financial statement integrity policy is then called to check whether the financial statement is complete. If there is data missing in the net profit area, an audit log is generated and the feature is marked as an abnormality.

[0018] Step S200: extracting corresponding documents from any audit log, extracting and classifying abnormal features of the corresponding documents according to preset feature types, and obtaining an abnormal feature set of the corresponding documents; Wherein, step S200 includes the following steps: Step S201: randomly selecting an audit log and extracting the audit information group of each feature in the selected audit log; randomly selecting a feature, if there is an abnormality mark in the audit information group of the selected feature, then setting the selected feature as an abnormal feature; Step S202: Obtain matching entities of any abnormal feature in the selected audit log, divide all abnormal features into a number of abnormal feature sets according to the matching entities; arbitrarily select one abnormal feature set, extract feature attributes of the entity in the selected abnormal feature set, and set the extracted feature attributes as the abnormal feature in the selected audit log; Step S203: Extract the audit time points and audit strategies from the audit information groups of each feature, sort the features in the order of the audit time points, and obtain the audit time intervals between any two adjacent features; if the audit strategies adopted by two adjacent features are different, set the audit time interval between the two adjacent features as the reference time interval, and calculate the average of all reference time intervals to obtain the average time interval; Step S204: If there are several adjacent and continuous features with the same audit strategy, the several features are preliminarily divided into similar feature sets, and all features are divided into several similar feature sets; any similar feature set is selected, and the audit time interval between two adjacent features in the selected similar feature set is obtained. If the audit time interval exceeds the average time interval, the selected similar feature set is divided into two feature sets based on the two adjacent features, and several feature sets are regenerated; Step S205: arbitrarily select an abnormal feature, determine the key set where the selected abnormal feature is located, and divide all abnormal features into several abnormal feature sets corresponding to the selected audit log documents according to the distribution of each abnormal feature in each feature set.

[0019] Step S300: Based on the audit results of any audit log, analyze the correlation between different features; based on the correlation between each feature in different audit logs, simulate the causal relationship between the correlated features; Wherein, step S300 includes the following steps: Step S301: arbitrarily select an audit log, extract several abnormal feature sets from the selected audit log, sort the abnormal feature sets according to the chronological order of the audit time points, and arbitrarily select two abnormal feature sets, and select one abnormal feature from each of the two abnormal feature sets as two comparison features; Step S302: Pre-build a semantic association database, match the semantic associations between each feature attribute in the feature attribute library, obtain the association value between any two feature attributes, and if the association value between the two feature attributes exceeds a preset association threshold, then set the two feature attributes to have an association relationship; obtain the feature attributes corresponding to the two compared features, and if there is an association relationship between the two compared features, set the two selected abnormal features as an associated feature group, and obtain a number of associated feature groups for the selected audit log; Step S303: arbitrarily select a correlation feature group, adjust the two abnormal features in the selected correlation feature group according to the order of the audit time points, and generate a sequential correlation feature group; extract all sequential correlation feature groups from different audit logs, and arbitrarily select two sequential correlation feature groups. If the audit strategies adopted by the two sequential correlation feature groups are the same, then the two sequential key feature groups are set as similar feature groups, and several similar feature groups are aggregated to generate a similar feature group set. The previous abnormal feature and the next abnormal feature of each similar feature group in the similar feature group set are aggregated to generate a first feature set and a second feature set, and a causal relationship is set between the first feature set and the second feature set; Example 2: Suppose that multiple financial report audit logs are analyzed and it is found that "operating income anomalies" and "income tax anomalies" appear at the same time. Then, through analysis through the semantic association library, it is found that there is a correlation between operating income and income tax, and the two features are set as the associated feature group; at the same time, in different audit logs, it is found that "operating income anomalies" are audited first and "income tax anomalies" are audited later, and the same audit strategy is adopted. Then, the causal relationship of "income anomalies may lead to tax anomalies" is preliminarily simulated.

[0020] Step S400: Analyze the causal simulation of any associated feature in any audit log to obtain valid causal feature pairs; based on the audit results of any valid causal feature pairs in different audit logs, generate the actual causal relationship between each feature; Step S400 includes the following steps: Step S401: arbitrarily select a similar feature group set to obtain a first feature set and a second feature set of the selected similar feature group set, and arbitrarily select an abnormal feature from each of the first feature set and the second feature set to generate a simulated feature group; Step S402: Randomly select an audit log, obtain the audit information group of each feature in the selected audit log, compare the two abnormal features in the simulated adjustment group with each feature, and if there are two features in the selected audit log that are respectively the same as the two abnormal features, obtain the audit information group of the two features. If the audit information groups of the two features both contain abnormal marks, and the order of the audit time points and the audit strategies adopted are also the same, then set the simulated feature group as a valid causal feature pair; Step S403: extract all valid causal feature pairs from the selected audit log, and arbitrarily select a group of valid causal feature pairs. Compare the selected valid causal feature pairs with the features in the remaining audit logs. If the selected valid causal feature pair also exists in the remaining audit logs, the selected valid causal feature pair is set as the actual causal feature pair. Step S404: arbitrarily select two actual causal feature pairs. If the latter feature in one actual causal feature pair and the former feature in the other actual causal feature pair generate a new actual causal feature pair, then connect the two selected actual causal feature pairs to obtain a causal feature traceability chain. Connect each actual causal feature pair to generate a causal feature traceability network. Example 3: Assume that "income anomaly → tax anomaly" is selected from the simulated causal pairs. If a search in the historical logs reveals that these two anomalies appear simultaneously in multiple logs with the same time sequence, then this pair is marked as an "actual causal feature pair." If "cost anomaly → income anomaly" and "income anomaly → tax anomaly" are found at the same time, then the two are connected into a causal chain: "cost anomaly → income anomaly → tax anomaly," thereby constructing a causal feature tracing network.

[0021] Step S500: Identify abnormal features in the actual document review process, and trace the abnormal features based on the actual causal relationship between the abnormal features; Wherein, step S500 includes the following steps: Step S501: Real-time monitoring of the document review process, entity recognition and feature extraction of document data in the real-time document, comparison of the review policy database and the feature attribute database, and generation of review information groups for each feature in the real-time document; obtaining abnormality flags for each feature of the review information group, and obtaining a number of abnormal features of the real-time document; Step S502: arbitrarily select two abnormal features from the plurality of abnormal features, extract the audit time points in the audit information group of the two selected abnormal features, generate the expected causal feature group according to the order of the audit time points, and obtain the expected causal feature groups of the real-time document. Step S503: The causal feature traceability network is retrieved and compared with the plurality of expected causal feature groups. If the plurality of expected causal feature groups are continuously connected in the causal feature traceability network, an expected traceability chain is generated, the traceability features of each abnormal feature are obtained, and an abnormal repair reminder is issued for each traceability feature. The abnormal feature recognition is performed again on the repaired real-time document until the real-time document is reviewed.

[0022] A document compliance audit data management system, the management system includes a document data acquisition module, a document feature recognition module, a causal association simulation module, a causal association generation module and an audit anomaly capture module; The document data collection module is used to record the compliance review process of any document and generate the corresponding review log; it reviews and analyzes the document data in any review log, generates the review results of the corresponding document, and provides feedback; The document feature recognition module is used to extract the corresponding documents of any audit log, extract and classify the abnormal features of the corresponding documents according to the preset feature types, and obtain the abnormal feature set of the corresponding documents; The causal association simulation module is used to analyze the correlation between different features based on the audit results of any audit log; based on the correlation between each feature in different audit logs, it simulates the causal relationship between the related features; The causal association generation module is used to analyze the causal simulation of any associated features in any audit log to obtain valid causal feature pairs; based on the audit results of any valid causal feature pairs in different audit logs, it generates the actual causal relationship between each feature; The audit anomaly capture module is used to identify abnormal features that exist in the actual document audit process and trace the abnormal features based on the actual causal relationship between the abnormal features.

[0023] Among them, the document data collection module includes a data collection and processing unit and an audit result feedback unit; The data collection and processing unit is used to record the compliance review process of any document and generate the corresponding audit log; the audit result feedback unit is used to review and analyze the document data in any audit log, generate the audit results of the corresponding document and provide feedback.

[0024] Among them, the causal association simulation module includes a feature association analysis unit and a causal relationship simulation unit; The feature correlation analysis unit is used to analyze the correlation between different features based on the audit results of any audit log; the causal relationship simulation unit is used to simulate the causal relationship between related features based on the correlation between various features in different audit logs.

[0025] Among them, the causal association generation module includes a relationship simulation verification unit and a feature causal generation unit; The relationship simulation verification unit is used to analyze the causal simulation of any associated feature in any audit log to obtain a valid causal feature pair; the feature causal generation unit is used to generate the actual causal relationship between each feature based on the audit results fed back by any valid causal feature pair in different audit logs.

[0026] 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. An artificial intelligence-based document compliance review data management method, characterized by: The management method comprises the following steps: Step S100: Record the compliance review process of any document and generate a corresponding review log; conduct review and analysis on the document data in any review log, generate the review results of the corresponding document and provide feedback; Step S200: extracting corresponding documents from any audit log, extracting and classifying abnormal features of the corresponding documents according to preset feature types, and obtaining an abnormal feature set of the corresponding documents; Step S300: Based on the audit results of any audit log, analyze the correlation between different features; based on the correlation between each feature in different audit logs, simulate the causal relationship between the correlated features; Step S400: Analyze the causal simulation of any associated feature in any audit log to obtain valid causal feature pairs; based on the audit results of any valid causal feature pairs in different audit logs, generate the actual causal relationship between each feature; Step S500: Identify abnormal features existing in the actual document review process, and trace the abnormal features based on the actual causal relationship between the abnormal features.

2. The method for managing document compliance review data based on artificial intelligence according to claim 1, characterized in that: The step S100 includes the following steps: Step S101: pre-configuring an audit policy database and a feature attribute library, wherein the audit policy database pre-stores a plurality of audit policies, wherein any audit policy is matched with a corresponding feature attribute set, and the feature attribute library pre-stores a plurality of feature attributes, wherein any feature attribute is matched with a feature data set; Step S102: Whenever a document to be reviewed is received, entity recognition and feature extraction are performed on the document data in the document to be reviewed using natural language processing technology. The recognition area of ​​each entity in the document to be reviewed is extracted, and the recognition area of ​​any feature is also extracted. If the recognition area of ​​a feature is within the recognition area of ​​a certain entity, the entity is matched with the feature to generate an entity data set for each entity. The document to be reviewed is divided into several entity data sets, and an audit log is generated. Step S103: arbitrarily select an audit strategy for generating an audit log, extract a feature attribute set and a feature data set that match the selected audit strategy, compare any feature attribute with the entity of the document to be audited, and compare any entity data with the features of the document to be audited. If all feature attributes and corresponding feature data in the selected audit strategy can be extracted from the document to be audited, then the selected audit strategy is set as the target audit strategy. Step S104: Retrieve each target audit strategy to conduct compliance audit on the document to be audited. If there is an abnormality in the selected feature, mark the selected feature as abnormal. Obtain the audit time point, audit strategy and abnormal mark of any selected feature to obtain the audit information group of the selected feature. Summarize the audit information groups of each feature in the document to be audited to obtain the audit result of the document to be audited and store it in the generated audit log.

3. The method for managing document compliance review data based on artificial intelligence according to claim 2, characterized in that: The step S200 includes the following steps: Step S201: randomly selecting an audit log and extracting the audit information group of each feature in the selected audit log; randomly selecting a feature, if there is an abnormality mark in the audit information group of the selected feature, then setting the selected feature as an abnormal feature; Step S202: Obtain matching entities of any abnormal feature in the selected audit log, divide all abnormal features into a number of abnormal feature sets according to the matching entities; arbitrarily select one abnormal feature set, extract feature attributes of the entity in the selected abnormal feature set, and set the extracted feature attributes as the abnormal feature in the selected audit log; Step S203: Extract the audit time points and audit strategies from the audit information groups of each feature, sort the features in the order of the audit time points, and obtain the audit time intervals between any two adjacent features; if the audit strategies adopted by two adjacent features are different, set the audit time interval between the two adjacent features as the reference time interval, and calculate the average of all reference time intervals to obtain the average time interval; Step S204: If there are several adjacent and continuous features with the same audit strategy, the several features are preliminarily divided into similar feature sets, and all features are divided into several similar feature sets; any similar feature set is selected, and the audit time interval between two adjacent features in the selected similar feature set is obtained. If the audit time interval exceeds the average time interval, the selected similar feature set is divided into two feature sets based on the two adjacent features, and several feature sets are regenerated; Step S205: arbitrarily select an abnormal feature, determine the key set where the selected abnormal feature is located, and divide all abnormal features into several abnormal feature sets corresponding to the selected audit log documents according to the distribution of each abnormal feature in each feature set.

4. The method for managing document compliance review data based on artificial intelligence according to claim 3, characterized in that: The step S300 includes the following steps: Step S301: arbitrarily select an audit log, extract several abnormal feature sets from the selected audit log, sort the abnormal feature sets according to the chronological order of the audit time points, and arbitrarily select two abnormal feature sets, and select one abnormal feature from each of the two abnormal feature sets as two comparison features; Step S302: Pre-build a semantic association database, match the semantic associations between each feature attribute in the feature attribute library, obtain the association value between any two feature attributes, and if the association value between the two feature attributes exceeds a preset association threshold, then set the two feature attributes to have an association relationship; obtain the feature attributes corresponding to the two compared features, and if there is an association relationship between the two compared features, set the two selected abnormal features as an associated feature group, and obtain a number of associated feature groups for the selected audit log; Step S303: arbitrarily select a correlation feature group, adjust the two abnormal features in the selected correlation feature group according to the order of the audit time points, and generate a sequential correlation feature group; extract all sequential correlation feature groups in different audit logs, and arbitrarily select two sequential correlation feature groups. If the audit strategies adopted by the two sequential correlation feature groups are the same, the two sequential key feature groups are set as similar feature groups, and several similar feature groups are summarized to generate a similar feature group set. The previous abnormal feature and the next abnormal feature of each similar feature group in the similar feature group set are summarized to generate a first feature set and a second feature set, and a causal relationship is set between the first feature set and the second feature set.

5. The method for managing document compliance review data based on artificial intelligence according to claim 4, characterized in that: The step S400 includes the following steps: Step S401: arbitrarily select a similar feature group set to obtain a first feature set and a second feature set of the selected similar feature group set, and arbitrarily select an abnormal feature from each of the first feature set and the second feature set to generate a simulated feature group; Step S402: Randomly select an audit log, obtain the audit information group of each feature in the selected audit log, compare the two abnormal features in the simulated adjustment group with each feature, and if there are two features in the selected audit log that are respectively the same as the two abnormal features, obtain the audit information group of the two features. If the audit information groups of the two features both contain abnormal marks, and the order of the audit time points and the audit strategies adopted are also the same, then set the simulated feature group as a valid causal feature pair; Step S403: extract all valid causal feature pairs from the selected audit log, and arbitrarily select a group of valid causal feature pairs. Compare the selected valid causal feature pairs with the features in the remaining audit logs. If the selected valid causal feature pair also exists in the remaining audit logs, the selected valid causal feature pair is set as the actual causal feature pair. Step S404: arbitrarily select two actual causal feature pairs. If the latter feature in one actual causal feature pair and the former feature in the other actual causal feature pair generate a new actual causal feature pair, then the two selected actual causal feature pairs are connected to obtain a causal feature traceability chain, and the actual causal feature pairs are connected to each other to generate a causal feature traceability network.

6. The method for managing document compliance review data based on artificial intelligence according to claim 5, characterized in that: The step S500 includes the following steps: Step S501: Real-time monitoring of the document review process, entity recognition and feature extraction of document data in the real-time document, comparison of the review policy database and the feature attribute database, and generation of review information groups for each feature in the real-time document; obtaining abnormality flags for each feature of the review information group, and obtaining a number of abnormal features of the real-time document; Step S502: arbitrarily select two abnormal features from the plurality of abnormal features, extract the audit time points in the audit information group of the two selected abnormal features, generate the expected causal feature group according to the order of the audit time points, and obtain the expected causal feature groups of the real-time document. Step S503: The causal feature traceability network is retrieved and compared with the plurality of expected causal feature groups. If the plurality of expected causal feature groups are continuously connected in the causal feature traceability network, an expected traceability chain is generated, the traceability features of each abnormal feature are obtained, and an abnormal repair reminder is issued for each traceability feature. The abnormal feature recognition is performed again on the repaired real-time document until the real-time document is reviewed.

7. A document compliance review data management system, configured to execute the document compliance review data management method based on artificial intelligence according to any one of claims 1 to 6, characterized in that: The management system includes a document data acquisition module, a document feature recognition module, a causal association simulation module, a causal association generation module and an audit anomaly capture module; The document data collection module is used to record the compliance review process of any document and generate a corresponding review log; perform review and analysis on the document data in any review log, generate the review results of the corresponding document and provide feedback; The document feature recognition module is used to extract the corresponding documents of any audit log, extract and classify the abnormal features of the corresponding documents according to the preset feature types, and obtain the abnormal feature set of the corresponding documents; The causal association simulation module is used to analyze the correlation between different features based on the audit results of any audit log; based on the correlation between each feature in different audit logs, simulate the causal relationship between the associated features; The causal association generation module is used to analyze the causal simulation of any associated feature in any audit log to obtain a valid causal feature pair; based on the audit results fed back by any valid causal feature pair in different audit logs, generate the actual causal relationship between each feature; The audit anomaly capture module is used to identify abnormal features existing in the actual document audit process and trace the abnormal features based on the actual causal relationship between the abnormal features.

8. A document compliance review data management system according to claim 7, characterized in that: The document data acquisition module includes a data acquisition processing unit and an audit result feedback unit; The data collection and processing unit is used to record the compliance review process of any document and generate a corresponding review log; the review result feedback unit is used to review and analyze the document data in any review log, generate the review results of the corresponding document and provide feedback.

9. The document compliance review data management system according to claim 7, characterized in that: The causal association simulation module includes a feature association analysis unit and a causal relationship simulation unit; The feature association analysis unit is used to analyze the association between different features based on the audit results of any audit log; the causal relationship simulation unit is used to simulate the causal relationship between associated features based on the association between each feature between different audit logs.

10. The document compliance review data management system according to claim 7, characterized in that: The causal association generation module includes a relationship simulation verification unit and a feature causal generation unit; The relationship simulation verification unit is used to analyze the causal simulation of any associated feature in any audit log to obtain a valid causal feature pair; the feature causal generation unit is used to generate the actual causal relationship between each feature based on the audit results fed back by any valid causal feature pair in different audit logs.

Images