A group work business data artificial intelligence comprehensive research and judgment analysis method and device

Abstract

This invention provides a method and apparatus for comprehensive AI-based analysis of organizational data, belonging to the field of AI technology. The method includes: decomposing target information according to analysis dimensions to obtain analysis threads for each dimension; acquiring analysis requirements input by the receiving unit; performing analysis task analysis on the bidirectional information of each review node based on the analysis requirements to obtain a task feature set; performing sensitivity analysis on all review nodes in each analysis thread; adjusting the sensitivity analysis results based on the task feature set to obtain a sensitivity vector; performing significant processing on the review nodes in the corresponding analysis thread; and generating a structured analysis report based on all significant processing results. This improves the efficiency, accuracy, and standardization of organizational data analysis, providing reliable data support for organizational decision-making.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and apparatus for comprehensive analysis and judgment of organizational business data using artificial intelligence. Background Technology

[0002] In the field of organizational work, the wide scope, multiple dimensions of data correlation, and high sensitivity of its business data place stringent demands on the accuracy and efficiency of analysis and judgment. However, current methods for analyzing organizational business data still suffer from numerous technical bottlenecks, making it difficult to meet practical work needs. Specific shortcomings are as follows:

[0003] Existing assessment methods rely on manual or fixed processes, lacking the automatic dimension identification and node positioning capabilities achievable by artificial intelligence. They often mix data from different dimensions together, leading to chaotic analysis and an inability to accurately locate key review nodes under each dimension. Existing technologies have weak information integration capabilities, and ignoring the correlation between review nodes can easily lead to data fragmentation, failing to form a complete assessment chain. Furthermore, existing technologies mostly rely on fixed analytical indicators and cannot analyze core elements based on the specific needs input by the user unit. This often results in over-analysis of non-critical data or omission of core requirement indicators, leading to insufficient accuracy in the generated task feature sets, which cannot support subsequent efficient assessment. In addition, existing sensitivity analysis mechanisms are simplistic, focusing only on numerical sensitivity and ignoring scenario and operational sensitivity, resulting in incomplete identification of sensitive information and causing omissions or misjudgments.

[0004] The aforementioned problems lead to a decrease in the accuracy and efficiency of analysis and judgment. Therefore, this invention proposes a method and apparatus for comprehensive analysis and judgment of organizational business data using artificial intelligence. Summary of the Invention

[0005] This invention provides a method and apparatus for comprehensive analysis and judgment of organizational business data using artificial intelligence, in order to solve the aforementioned technical problems.

[0006] This invention provides a method for comprehensive artificial intelligence analysis of organizational business data, including:

[0007] Step 1: Collect target information for each target person and break down the target information according to the judgment dimensions to obtain the analysis thread for each judgment dimension. Each analysis thread contains at least one audit node, and each audit node involves bidirectional information of several audit indicators. The bidirectional information includes the personnel permissions and audit process trajectory of the auditor, and the business data and audit result log of the auditee. The judgment dimensions include management-related dimensions, construction-related dimensions, and work-related dimensions.

[0008] Step 2: Obtain the assessment requirements input by the receiving and using unit, and perform assessment task analysis on the bidirectional information of each review node based on the assessment requirements to obtain the task feature set;

[0009] Step 3: Perform sensitivity analysis on all audit nodes in each analysis thread, and adjust the sensitivity analysis results based on the task feature set to obtain a sensitivity vector;

[0010] Step 4: Perform significant processing on the audit nodes in the corresponding analysis thread according to the sensitivity vector, and generate a structured analysis report based on all significant processing results.

[0011] Preferably, the target information is decomposed according to the judgment dimensions to obtain the analysis thread for each judgment dimension, including:

[0012] Information is extracted from the target information based on each judgment dimension in the preset dimension set to obtain bidirectional information of different target tasks under each judgment dimension;

[0013] The bidirectional information is sorted according to time sequence and task type to obtain analysis threads for corresponding judgment dimensions, wherein each target task corresponds to an audit node.

[0014] Preferably, the task feature set is obtained by analyzing the bidirectional information of each review node based on the assessment requirements, including:

[0015] The judgment requirements are semantically parsed to obtain several requirement keywords, and the time sequence of each requirement keyword is analyzed to obtain several global elements and local elements. The elements are sorted according to the time sequence of the first appearance of each element and the task coverage of the time sequence of each element to obtain an element vector.

[0016] The element vectors are matched with the vector-standard comparison table to obtain the first judgment standard for each element. At the same time, the element vectors are input into the vector analysis model to obtain the second judgment standard for each element.

[0017] The judgment accuracy of the corresponding elements is determined according to the first judgment standard and the second judgment standard, and the first element with a judgment accuracy greater than the corresponding preset accuracy is selected. At the same time, the second element with a ranking value greater than the preset value in the element vector is selected, and the first element and the second element are used as the core elements.

[0018] The structured data in the bidirectional information of each audit node is normalized and mapped into numerical features, and the unstructured data is semantically analyzed and mapped into semantic features.

[0019] According to the assessment target type of the assessment requirements, the numerical features and semantic features are classified and analyzed, the mutual information value between each classification feature and each core element is established, and an initial weight is set for each classification feature based on the weight ratio of the core element weight, the reviewer's authority-related features and the reviewee's result features.

[0020] The initial feature set is obtained by filtering the classification features whose initial weights are greater than the preset weights.

[0021] Preferably, after obtaining the initial feature set, the process further includes:

[0022] Each remaining feature is sequentially compared with each valid feature in the set of valid features for similar historical needs to obtain a similarity set;

[0023] If the number of qualified elements in the similarity set is greater than the standard set number, the remaining features will be classified into the initial feature set to obtain the task feature set.

[0024] Preferably, sensitivity analysis is performed on all audit nodes in each analysis thread, including:

[0025] The multidimensional sensitivity baseline of each analysis thread is retrieved, and the corresponding analysis threads are aligned to obtain the multidimensional sensitivity analysis mechanism of each audit node. The multidimensional sensitivity analysis mechanism is related to numerical sensitivity, scene sensitivity, and operation sensitivity.

[0026] According to the aforementioned multidimensional sensitivity analysis mechanism, the personnel permissions and review process trajectory in the corresponding two-way information are subjected to a first comparative analysis, the business data and review result data are subjected to a second comparative analysis, and the review process trajectory and review result data are subjected to a third comparative analysis.

[0027] Based on the results of the first, second, and third control analyses, a control result matrix was constructed.

[0028] Based on the comparison result matrix and the standard range matrix of the audit nodes, a sensitive label and an initial sensitive value are set for each comparison element. The sensitive label is related to one or more of the following: numerical sensitivity, scenario sensitivity, and operational sensitivity.

[0029] Preferably, the sensitivity vector is obtained by adjusting the sensitivity analysis results based on the task feature set, including:

[0030] Input the core elements of the task feature set of the same review node and the corresponding sensitivity analysis results into the association model, and construct the feature-sensitivity association matrix by calculating the co-occurrence probability of feature items and sensitive labels;

[0031] Adjustment coefficients are set based on the weight values ​​of strongly correlated features in the feature-sensitivity correlation matrix;

[0032] If the feature term is positively correlated with the sensitive label, the initial sensitivity value is adjusted based on the first formula of the adjustment coefficient;

[0033] If the feature term is negatively correlated with the sensitive label, the initial sensitivity value is adjusted based on the second formula of the adjustment coefficient;

[0034] Determine the dependencies between review nodes in the analysis thread, calculate the influence of highly sensitive nodes on the corresponding review nodes, and accumulate it into each adjusted sensitivity value under the corresponding review node. The influence is obtained by multiplying the historical collaboration frequency between review nodes with the standard deviation of all initial sensitivity values ​​under the corresponding review node.

[0035] The sensitivity vector is obtained by arranging all the accumulated sensitivity values ​​and reference elements of each review node in the order of the analysis thread. The sensitivity vector contains the sensitivity value set of each review node in the corresponding analysis thread, and the sensitivity value set contains all the accumulated sensitivity values, sensitivity tags and reference elements under the corresponding review node.

[0036] Preferably, the review nodes in the corresponding analysis thread are significantly processed according to the sensitivity vector, including:

[0037] The sensitivity value set of each review node contained in the sensitivity vector is analyzed dimensionally to extract the comprehensive sensitivity level, the combination of sensitive types, and the identifier of the associated node.

[0038] Based on the level-combination-identifier-policy mapping table, matching identifiers are obtained and assigned according to the extraction results. During the assignment process, each sub-identifier in the matching identifier is significantly overwritten with the specified information in the corresponding bidirectional information.

[0039] Preferably, a structured analysis report is generated based on all significant processing results, including:

[0040] Dimensional analysis is performed on all significant processing results to identify the judgment dimension system. Based on the judgment dimension system, the significant processing results are divided into a dimension feature layer, an association logic layer, and a conclusion output layer. The dimension feature layer is related to the core sensitive identifier set under each judgment dimension, the association logic layer is related to the sensitive transmission and dependency relationship between review nodes, and the conclusion output layer is related to the preliminary judgment conclusion of each review node.

[0041] Extract the set of salient identifiers for each judgment dimension in the dimensional feature layer, perform semantic enhancement processing on the bidirectional information associated with each identifier, generate a sensitive identifier-associated information mapping graph, calculate the aggregation density of highly sensitive identifiers under each judgment dimension based on the mapping graph, and determine the feature compliance level of the corresponding dimension by combining the feature distribution threshold of similar judgment dimensions in history.

[0042] Based on the association logic layer, a dependency network of audit nodes is constructed. Highly sensitive propagation paths in the dependency network are mined, the sensitive impact weight of each path is calculated, and the top N9 key sensitive transmission paths with the highest impact weight are selected to generate a sensitive transmission logic summary.

[0043] The system calls upon a pre-defined rule library for expressing organizational conclusions to transform the preliminary judgments from the output layer into standardized conclusions that conform to organizational writing conventions, and marks each standardized conclusion with a prominent identifier and a source link.

[0044] The feature compliance level, sensitive transmission logic summary, standardized conclusions, and significant identifier traceability links are automatically filled into the report template to generate a structured analysis report.

[0045] This invention provides an artificial intelligence-based comprehensive analysis and judgment device for organizational business data, comprising:

[0046] The thread parsing module is used to collect target information for each target person and decompose the target information according to the judgment dimensions to obtain the analysis thread for each judgment dimension. Each analysis thread contains at least one audit node, and each audit node involves bidirectional information of several audit indicators. The bidirectional information includes the personnel permissions and audit process trajectory of the auditing party, and the business data and audit result log of the audited party. The judgment dimensions include management-related dimensions, construction-related dimensions, and work-related dimensions.

[0047] The feature set acquisition module is used to acquire the judgment requirements input by the receiving and using unit, and to perform judgment task analysis on the bidirectional information of each review node based on the judgment requirements to obtain the task feature set.

[0048] The sensitivity analysis module is used to perform sensitivity analysis on all audit nodes in each analysis thread, and adjust the sensitivity analysis results based on the task feature set to obtain a sensitivity vector;

[0049] The report generation module is used to perform significant processing on the audit nodes in the corresponding analysis thread according to the sensitivity vector, and generate a structured assessment report based on all significant processing results.

[0050] Compared with the prior art, the beneficial effects of this application are as follows:

[0051] By systematically collecting two-way information on organizational work, breaking down and analyzing threads according to three dimensions—management, construction, and work—and dynamically extracting precise task characteristics based on the needs of user units, the system identifies key sensitive nodes through multi-dimensional sensitivity analysis and dynamic adjustment. Ultimately, it generates structured analysis reports that conform to organizational work document standards, significantly improving the efficiency, accuracy, and standardization of organizational work data analysis and providing reliable data support for organizational work decision-making.

[0052] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0053] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0054] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0055] Figure 1 This is a flowchart of an artificial intelligence-based comprehensive analysis method for organizational business data in an embodiment of the present invention;

[0056] Figure 2 This is a structural diagram of an artificial intelligence-based comprehensive analysis and judgment device for organizational business data in an embodiment of the present invention. Detailed Implementation

[0057] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0058] This invention provides a method for comprehensive artificial intelligence analysis of organizational business data, such as... Figure 1 As shown, it includes:

[0059] Step 1: Collect target information for each target person and break down the target information according to the judgment dimensions to obtain the analysis thread for each judgment dimension. Each analysis thread contains at least one audit node, and each audit node involves bidirectional information of several audit indicators. The bidirectional information includes the personnel permissions and audit process trajectory of the auditor, and the business data and audit result log of the auditee. The judgment dimensions include management-related dimensions, construction-related dimensions, and work-related dimensions.

[0060] Step 2: Obtain the assessment requirements input by the receiving and using unit, and perform assessment task analysis on the bidirectional information of each review node based on the assessment requirements to obtain the task feature set;

[0061] Step 3: Perform sensitivity analysis on all audit nodes in each analysis thread, and adjust the sensitivity analysis results based on the task feature set to obtain a sensitivity vector;

[0062] Step 4: Perform significant processing on the audit nodes in the corresponding analysis thread according to the sensitivity vector, and generate a structured analysis report based on all significant processing results.

[0063] In this embodiment, the target personnel are specific individuals who need to be analyzed and evaluated in the organization business, and are related to the core organization business.

[0064] In this embodiment, the target information refers to all information related to the assessment generated by the target personnel throughout the entire organizational work process, including basic identity information and business-related data. This data is collected in batches from a pre-stored organizational work business-related database. For example, basic identity information includes gender, age, education, and current position; business information includes annual performance appraisal forms, appointment documents, and major event report forms for the past three years; basic information for target personnel applying to work as community workers includes place of origin and years of service; business information includes application form, training registration form, and community participation record; basic information for target personnel applying to work as engineers includes their professional field and graduating institution; business information includes patent authorization certificates, project research and development reports, and unit recommendation opinions. It should be noted that the relevant information originates from entities with corresponding permissions, and data anonymization and other processing can be performed on the relevant information according to actual needs.

[0065] In this embodiment, the management-related dimension focuses on the assessment of selection, evaluation, appointment, and supervision, with an emphasis on core aspects such as qualifications, work performance, and compliance. The construction-related dimension focuses on the assessment of development, education, management, and services, with an emphasis on the compliance of development processes, participation in education, and the effectiveness of their roles. The work-related dimension focuses on the assessment of recruitment, recognition, training, and services, with an emphasis on the compliance of application conditions, the effectiveness of project implementation, and the implementation of service guarantees.

[0066] In this embodiment, the analysis thread is a single judgment dimension. The related data under this dimension in the target information are sorted according to the business process time sequence and task type to form a complete analysis link. Each link corresponds to the judgment logic of a dimension. For example, the order of the promotion analysis thread under the management of management-related dimensions includes: tenure review node - annual performance review node - major event report review node - record review node, etc.

[0067] In this embodiment, the audit node is a key step in the analysis thread to audit and judge specific business processes. Each node corresponds to a specific audit task and is the smallest execution unit for judgment. The audit indicator is the specific standard used in the audit node to judge whether the business process is compliant and meets the standards. It is the core basis for audit judgment. According to the organizational documents and business specifications, the judgment requirements of each audit node are transformed into quantifiable and verifiable indicators to form an audit node-audit indicator comparison table, which serves as the execution standard for audit operations.

[0068] In this embodiment, bidirectional information is collected from the organization and personnel business system, specifically from the information of the reviewer and the reviewee. Personnel permissions are based on job responsibilities and regulations, and are pre-defined roles with different permissions for different positions within the business system, such as whether they have appointment review permissions. The review process trajectory is a record of all operations performed by the reviewer at each review node, including operation time, operation content, and operation object, such as file retrieval time and appointment document verification time. Business data refers to specific information generated by the target personnel in the organization and personnel business that is directly related to the review node, such as talent application business data. The review result log is structured information recording the review conclusion, reasons, basis, and remarks, such as pass or fail.

[0069] In this embodiment, the user unit refers to the organization-related part or institution that proposes the need for analysis and judgment of organizational business data. The analysis and judgment need is the specific requirement proposed by the user unit based on business work needs to analyze and judge specific organizational data.

[0070] In this embodiment, the assessment task analysis is a process of breaking down the assessment requirements into specific, executable analysis tasks, clarifying the objectives, required data, and completion standards for each task. For example, the task analysis for assessing the qualifications for a proposed promotion is broken down into three tasks: 1. Extract the appointment information and assessment data of 50 candidates for promotion; 2. Review each candidate's tenure to ensure it meets the required duration and that all assessment scores are qualified or above; 3. Verify that all operators at each review node have the corresponding permissions and count any nodes with violations. The feature task set is a collection of feature information selected from bidirectional information that provides core support for the complete assessment task. It includes numerical and semantic features and must align with the key assessment requirements. For example, the feature set for assessing the qualifications for a proposed promotion includes: numerical features: tenure, number of qualified assessments in the past 3 years; semantic features: description of reviewer permissions, keywords in review opinions. The feature set for assessing talent project applications includes: numerical features: patent authorization time, project funding; semantic features: results verification opinions, recommendations from recommending units.

[0071] In this embodiment, sensitivity analysis is the process of detecting bidirectional information at each review node in the analysis thread to identify sensitive issues such as data anomalies, scenario violations, and operational non-compliance. It is implemented based on a pre-established three-dimensional sensitivity analysis mechanism of value-scenario-operation. For example, it identifies anomalies by comparing business data with the baseline; scenario sensitivity: it identifies missing materials and missing processes by comparing with business specifications; operation sensitivity: it verifies the permissions and operation records of reviewers to identify unauthorized access. Sensitive issues are identified through bidirectional information comparison and analysis using this mechanism.

[0072] In this embodiment, the sensitivity analysis result is an information record obtained after sensitivity analysis, which includes the sensitivity type, initial sensitivity value and basis of each audit node.

[0073] In this embodiment, after the sensitivity analysis results of the sensitivity vector set task feature set are adjusted, the structured data sequence arranged in the order of the review nodes in the analysis thread includes the adjusted sensitivity value, sensitivity label and control element of each node.

[0074] In this embodiment, the salience processing involves marking highly sensitive review nodes with supplementary explanations or related supporting materials based on sensitive information in the sensitivity vector. The salience-processed node information is then organized into a structured result table according to the structure of review node name - sensitivity level - salience mark - supplementary explanation - material link. The structured analysis report integrates all salience processing results into a formal document in accordance with the organizational work business writing specifications and preset templates.

[0075] The beneficial effects of the above technical solution are as follows: by systematically collecting two-way information on organizational work, breaking down and analyzing threads according to the three dimensions of management, construction, and work, dynamically extracting accurate task characteristics in combination with the needs of user units, and locating key sensitive nodes through multi-dimensional sensitivity analysis and dynamic adjustment, a structured analysis report that conforms to the norms of organizational work documents is finally generated, which significantly improves the efficiency, accuracy, and standardization of organizational work data analysis and provides reliable data support for organizational work decision-making.

[0076] This invention provides a method for comprehensive AI-based analysis of organizational business data, which decomposes the target information according to analysis dimensions to obtain analysis threads for each analysis dimension, including:

[0077] Information is extracted from the target information based on each judgment dimension in the preset dimension set to obtain bidirectional information of different target tasks under each judgment dimension;

[0078] The bidirectional information is sorted according to time sequence and task type to obtain analysis threads for corresponding judgment dimensions, wherein each target task corresponds to an audit node.

[0079] In this embodiment, the preset dimension set is a set of information extracted in advance based on the core areas of organizational business and related specifications, which serves as the basis for judging the dimensions.

[0080] In this embodiment, information extraction involves extracting information content related to the judgment dimension from the target information.

[0081] In this embodiment, the target task is the specific work task that needs to be completed to achieve the assessment target under the current dimension. Each target task corresponds to a set of clear review requirements and work content. For example, if the assessment dimension is the talent assessment review sub-dimension, the target tasks can be divided into application condition review tasks.

[0082] In this embodiment, the task type is classified according to business attributes. Reflecting the core review direction of the target task is another basis for sorting. It must be consistent with the stage division of the organization business process, such as preliminary review, secondary review, and final review.

[0083] In this embodiment, sorting is a process of organizing and arranging bidirectional information of all target tasks under the same judgment dimension by combining time sequence and task type. First, the tasks are grouped by task type, and then arranged in time sequence within each group, ultimately forming a logically clear information sequence.

[0084] The beneficial effects of the above technical solution are as follows: by extracting target information around a preset dimension set and obtaining bidirectional information of target tasks under each judgment dimension, and then sorting them by time sequence and task type to form an analysis thread containing review nodes, the extraction direction and organization logic of organizational business data are standardized, avoiding the blindness of information extraction and data chaos, and constructing a structured information link that fits the actual business process. This makes the bidirectional information of each review link traceable and the review standards clear, laying a solid data foundation for subsequent sensitive analysis, accurate location of risk points and generation of standardized judgment reports.

[0085] This invention provides an artificial intelligence-based comprehensive analysis method for organizational business data. Based on the analysis requirements, it performs analysis of the bidirectional information of each review node to obtain a task feature set, including:

[0086] The judgment requirements are semantically parsed to obtain several requirement keywords, and the time sequence of each requirement keyword is analyzed to obtain several global elements and local elements. The elements are sorted according to the time sequence of the first appearance of each element and the task coverage of the time sequence of each element to obtain an element vector.

[0087] The element vectors are matched with the vector-standard comparison table to obtain the first judgment standard for each element. At the same time, the element vectors are input into the vector analysis model to obtain the second judgment standard for each element.

[0088] The judgment accuracy of the corresponding elements is determined according to the first judgment standard and the second judgment standard, and the first element with a judgment accuracy greater than the corresponding preset accuracy is selected. At the same time, the second element with a ranking value greater than the preset value in the element vector is selected, and the first element and the second element are used as the core elements.

[0089] The structured data in the bidirectional information of each audit node is normalized and mapped into numerical features, and the unstructured data is semantically analyzed and mapped into semantic features.

[0090] According to the assessment target type of the assessment requirements, the numerical features and semantic features are classified and analyzed, the mutual information value between each classification feature and each core element is established, and an initial weight is set for each classification feature based on the weight ratio of the core element weight, the reviewer's authority-related features and the reviewee's result features.

[0091] The initial feature set is obtained by filtering the classification features whose initial weights are greater than the preset weights.

[0092] In this embodiment, constructing the feature vector includes:

[0093] Calculate the ranking value for each feature:

[0094] ;

[0095] in, Let be the sorting value of the i-th element; The sum of task weights under the task coverage area spanned by the i-th element; For the number of elements; The length of the time sequence spanned by the i-th element; This represents the total timing length. This is the overall sequence number of the timing sequence; The time sequence number of the first occurrence of the i-th element; Let be the feature attribute of the i-th feature, and the feature attribute can be either a global attribute or a local attribute. When it is a global attribute, The value is 1, when it is a local attribute. The value is 0.8;

[0096] Sort the values ​​from largest to smallest to form the element vector.

[0097] In this embodiment, the accuracy of the analysis of the corresponding elements is determined:

[0098] ;

[0099] in, The accuracy of the assessment of the i-th element; , These represent the standard precision of the i-th element based on the first and second judgment criteria, respectively. Let be the average historical assessment accuracy of the i-th element.

[0100] In this embodiment, The first standard is the number of accurate cases divided by the total number of cases. The second standard is the number of accurate cases divided by the total number of cases. The weight of each task within the task coverage of the i-th element over the past three years is calculated as the number of accurate cases over the past three years / the total number of cases over the past three years. The importance of each task is pre-defined, and the task weight can be obtained by directly summing and calculating the weights. .

[0101] In this embodiment, the demand keywords are words that represent the core content of the demand assessment and are selected from the semantic parsing results. They can be directly extracted.

[0102] In this embodiment, the "through-time sequence" refers to the coverage of the demand keyword across the entire analysis process timeline, that is, the time span from the start to the end of the business steps associated with the keyword, reflecting the duration of the keyword's impact on the analysis process. Global elements are the core elements that permeate the entire analysis process timeline and play a decisive role in achieving the analysis objective. Local elements are those that only permeate a portion of the analysis process timeline and provide support for specific analysis steps.

[0103] In this embodiment, the first occurrence of the time sequence point is the specific link and corresponding sequence number of the element when it first appears on the time sequence axis of the business process analysis, reflecting the order in which the element intervenes in the analysis.

[0104] In this embodiment, the task coverage is the sum of the number of specific tasks included in the analysis time sequence of the element penetration and the importance weight of each task.

[0105] In this embodiment, the element vector is an ordered set formed by arranging all elements in descending order of their sorting values, such as: {Proposed promotion 0.25, tenure requirement 0.19, compliance of review process 0.18, compliance of reviewer authority 0.17, standardization of result formula 0.11}.

[0106] In this embodiment, the vector-standard lookup table is a table of correspondence between element vectors and judgment standards, established in advance based on organizational documents, business specifications, and historical judgment experience. It includes fields such as element name, applicable scenario, and standard content. For example, element name: Required years of service, applicable scenario: promotion, standard content: ≥3 years of service in the current equivalent position, based on the date of organization document issuance; element name: Compliant reviewer permissions, applicable scenario: promotion review, standard content: reviewers need to have the authority to review the selection and appointment of candidates, and the permission configuration is based on the job description of the organization department.

[0107] In this embodiment, the first judgment criterion is to match each element in the element vector with the vector-standard lookup table and extract the corresponding standard. For example, the first judgment criterion is: the element is the length of service meets the standard. The first judgment criterion is that promotion requires the current position at the same level to have served for ≥3 years. The length of service is calculated from the date of the organization's document issuance to the date of judgment initiation.

[0108] In this embodiment, historical analysis data of organizational work over the past 3 to 5 years is collected, and effective cases containing element vectors, primary standards, actual execution standards, and analysis results are selected. The neural network model is trained with element vectors as input and actual execution standards as output to obtain a vector analysis model.

[0109] In this embodiment, the second judgment criterion is the judgment criterion output by the model after the element vector is input into the vector analysis model.

[0110] In this embodiment, the preset accuracy is a threshold for judgment accuracy set in advance based on the requirements of the organizational business for the accuracy of the judgment. It is used to screen the elements that meet the accuracy standard. Different elements can be set with different thresholds according to the importance of the business. The basic preset accuracy is set in combination with the risk level of the combined business. The threshold is adjusted according to the degree of influence of the element on the judgment result to form an element preset accuracy table. For example, the preset accuracy of the core element of meeting the tenure standard in the promotion judgment is 1.5.

[0111] The first element is the element whose judgment accuracy is greater than the preset accuracy. The preset value is a threshold value for ranking values ​​that is determined in advance based on the distribution of ranking values ​​in the element vector. It is used to filter elements that meet the importance criteria. For example, if the ranking value distribution of the promotion judgment element vector is 0.25, 0.19, 0.18, 0.17, and 0.11, the preset value is set to 0.16, which covers the first four elements.

[0112] In this embodiment, the core element is the union of the first element and the second element.

[0113] In this embodiment, structured data is data with a fixed format that can be directly quantified in bidirectional information. It usually exists in the form of tables, database fields, etc., including types such as numerical values, dates, and codes. The normalization process is a process of uniformly converting structured data into 0 to 1.

[0114] In this embodiment, numerical features are features presented in numerical form after structured data has been normalized.

[0115] In this embodiment, unstructured data refers to data in two-way information that has no fixed format and cannot be directly quantified. Unstructured data is extracted from the two-way information of the review node. Image data is processed by OCR to convert it into text; audio data is processed by speech-to-text and uniformly converted into text format; meaningless text is removed and core content is retained.

[0116] In this embodiment, semantic analysis is a process of extracting keywords, understanding semantics, and judging sentiment from unstructured text data to uncover core information and logical relationships within the text. For example, semantic analysis is performed on the review opinion that the appointment document is authentic and valid, the length of service is accurately calculated, and the basic conditions for promotion are met, extracting keywords such as: authentic appointment document, accurate length of service, and meeting the promotion conditions. For the project summary report abstract: the project has achieved a breakthrough in core technology and won the second prize of municipal science and technology progress, extracting keywords such as: breakthrough in core technology and second prize of municipal science and technology progress.

[0117] In this embodiment, semantic features are features presented in the form of keywords or phrases after unstructured data is semantically analyzed. For example, the semantic features of the audit opinion are: the appointment document is authentic, the length of service is accurate, and the conditions for promotion are met. The semantic features of the major event report are: holding shares in Company A, holding 5% of the shares, and applying in December 2023. The semantic features of the project summary are: breakthrough in core technology and second prize of municipal science and technology progress award.

[0118] In this embodiment, the target type is determined by clarifying the core target category to be achieved in the assessment work based on the assessment requirements, and determining the direction of feature classification. Common types in organizational work include: qualification compliance assessment, process compliance assessment, material authenticity assessment, and result rationality assessment.

[0119] In this embodiment, a rule is established to correspond to the target type, feature category, and feature type. Based on the rule, numerical features and semantic features are automatically classified. For example, qualification compliance assessment - qualification compliance features - features containing keywords such as job title, assessment, and education. The numerical features and semantic features are automatically classified according to the rule. The classified features are the feature sets that belong to the same feature category after classification analysis. Each classified feature corresponds to a specific target type and is the basic unit for subsequent weight setting and feature selection.

[0120] In this embodiment, the mutual information value is a quantitative indicator that measures the degree of correlation between classification features and core elements, and its value ranges from 0 to 1. The calculation formula is:

[0121] Where X is the classification feature and Y is the core element.

[0122] In this embodiment, the core element weight is an importance weight set according to the degree of influence of the core element on the assessment target. It is implemented through a multi-round anonymous scoring method, and its value ranges from 0 to 1.

[0123] In this embodiment, the features related to the reviewer's authority are those that are related to the reviewer's authority in the classification features. They reflect the compliance of the review process and cover the reviewer's authority, the scope of operation authority, the rationality of authority configuration, etc. Features containing keywords such as review authority, authority configuration, and operation authority can be selected from the classification features and labeled as review authority-related features.

[0124] In this embodiment, the auditee's result features are the categorized features that are related to the auditee's business results. They reflect the qualifications or authenticity of the materials of the subject being assessed and cover content such as appointment results, assessment results, and effectiveness of results. Features containing keywords such as appointment results, assessment results, effectiveness of results, and authenticity of materials are selected from the categorized features and can be labeled as auditee result features.

[0125] In this embodiment, the initial weight = mutual information value × core element weight × feature category weight coefficient.

[0126] In this embodiment, the preset weights are pre-set weight thresholds based on the distribution of the initial weights of the classification features and the accuracy requirements of the analysis. These thresholds are used to filter out features that provide core support for the analysis, eliminate secondary or redundant features, and calculate the average and median of the initial weights of all classification features. The average value is used as the initial reference for the preset weights. The thresholds are adjusted in conjunction with the number of features required for the analysis to ensure that the selected features cover the main analysis directions without excessive redundancy. The final preset weights are determined after confirmation by business personnel.

[0127] In this embodiment, the set of classification features with initial weights greater than preset weights is the core data foundation for the subsequent sensitivity analysis and judgment report generation.

[0128] The beneficial effects of the above technical solution are as follows: by semantic parsing, element sorting and standard matching of the assessment requirements, core elements are accurately extracted; then, by feature transformation, classification and weight setting of the two-way information of the review nodes, an initial feature set that highly matches the assessment target is selected to ensure the authority of the standard, and the flexibility is improved by combining historical data and model analysis. At the same time, the accuracy and importance of the features are guaranteed by quantitative means such as mutual information value and weight ratio.

[0129] This invention provides a method for comprehensive artificial intelligence analysis of organizational business data, which, after obtaining an initial feature set, further includes:

[0130] Each remaining feature is sequentially compared with each valid feature in the set of valid features for similar historical needs to obtain a similarity set;

[0131] If the number of qualified elements in the similarity set is greater than the standard set number, the remaining features will be classified into the initial feature set to obtain the task feature set.

[0132] In this embodiment, the remaining features are the classification features whose initial weights are less than the preset weights and have not been included in the initial feature set. Historical similar needs refer to the features accumulated under the same and similar needs in the past. Similar needs meet the same judgment dimensions and have the same core judgment objectives. The effective feature set is the set of classification features that have been verified to have a practical supporting role in the judgment conclusion during the judgment process of historical similar needs. And the effective feature is a single feature in the effective feature set.

[0133] In this embodiment, similarity analysis is based on calculating similarity using the Euclidean distance method.

[0134] In this embodiment, for a single remaining feature, a similarity analysis is performed on it one by one with all valid features in a set of valid features for a certain historical similar demand, forming a similarity value set. Each element in the set corresponds to the degree of similarity between the remaining feature and a valid feature. Elements with similarity values ​​greater than or equal to a preset qualified threshold are considered qualified elements. The preset qualified threshold is 0 or 6. For example, in the similarity set {0.95, 0.12, 0.08, 0.1}, only the element corresponding to 0.95 is a qualified element.

[0135] The method for determining the standard setting quantity is as follows:

[0136] ;

[0137] in, The standard number of matches to the corresponding similarity set; This represents the total number of elements present in the corresponding similarity set. The average number of elements in all valid feature sets of similar historical requirements; The constant value is 2.7; To round up, NB is dynamically calculated based on the current remaining features. and The threshold is adaptively adjusted to avoid the rigidity of fixed values, while accurately filtering out the remaining features that are similar to a sufficient number of historically effective features.

[0138] In this embodiment, the task feature set is obtained by incorporating the remaining features with a number of elements greater than the standard set into the initial feature set.

[0139] The beneficial effects of the above technical solution are: by performing similarity analysis between the remaining features and the effective features of similar historical requirements, and by comparing the number of qualified elements with the standard setting to screen valuable features, a more accurate and comprehensive data foundation is provided for subsequent sensitive analysis of audit nodes, sensitive vector adjustment, and generation of structured judgment reports.

[0140] This invention provides a comprehensive AI-based analysis method for organizational data, which performs sensitivity analysis on all review nodes in each analysis thread, including:

[0141] The multidimensional sensitivity baseline of each analysis thread is retrieved, and the corresponding analysis threads are aligned to obtain the multidimensional sensitivity analysis mechanism of each audit node. The multidimensional sensitivity analysis mechanism is related to numerical sensitivity, scene sensitivity, and operation sensitivity.

[0142] According to the aforementioned multidimensional sensitivity analysis mechanism, the personnel permissions and review process trajectory in the corresponding two-way information are subjected to a first comparative analysis, the business data and review result data are subjected to a second comparative analysis, and the review process trajectory and review result data are subjected to a third comparative analysis.

[0143] Based on the results of the first, second, and third control analyses, a control result matrix was constructed.

[0144] Based on the comparison result matrix and the standard range matrix of the audit nodes, a sensitive label and an initial sensitive value are set for each comparison element. The sensitive label is related to one or more of the following: numerical sensitivity, scenario sensitivity, and operational sensitivity.

[0145] In this embodiment, the multidimensional sensitivity baseline is a set of three sensitivity judgment benchmarks—numerical, scenario, and operation—determined in advance based on organizational guidelines and historical sensitive cases for each analysis thread. It is the core reference for identifying sensitive issues. Each dimension of the analysis thread corresponds to a specific baseline, which is obtained by matching from a pre-stored dimension-thread-sensitivity type-benchmark standard comparison table. For example, the multidimensional sensitivity baseline for the promotion analysis thread is as follows: the numerical sensitivity baseline is ≥3 years of service and a qualified performance evaluation; the scenario sensitivity baseline is that only designated personnel can participate in the initial qualification review and the public announcement period is ≥5 working days; and the operation sensitivity baseline is that the review process must include three steps: material retrieval, verification, and opinion entry, with an interval of ≤24 hours.

[0146] In this embodiment, the analysis thread alignment process refers to matching and associating the review nodes and bidirectional information of the analysis thread with the sensitivity types and judgment criteria of the multidimensional sensitive baseline. This ensures that the information of each review node corresponds to the specific sensitivity judgment rules in the baseline, eliminating the dimensional deviation between the information and the baseline. For example, during the alignment process, the length of service data corresponds to the baseline value sensitivity: length of service ≥ 3 years; the personnel permissions correspond to the baseline operation sensitivity: only designated personnel can conduct the initial review; and the initial review stage corresponds to the baseline scenario sensitivity: the initial review must be completed within 7 days after the promotion is initiated.

[0147] In this embodiment, the multidimensional sensitivity analysis mechanism, after alignment processing, determines a comprehensive analysis rule system covering numerical sensitivity, scenario sensitivity, and operational sensitivity for each review node. This clarifies which information needs analysis, what standards are used to determine sensitivity, and how sensitivity types are categorized. This serves as the basis for subsequent comparative analysis. For example:

[0148] Multi-dimensional sensitivity analysis mechanism for promotion authority verification nodes:

[0149] Numerical sensitivity analysis rules: The reviewer's permission level must be ≥3; otherwise, the data will be deemed sensitive.

[0150] Scene sensitivity analysis rules: Access control must be carried out after the materials have been reviewed and approved; if it is carried out earlier or later, the scene will be deemed sensitive.

[0151] Operation sensitivity analysis rules: The review process must include access control table retrieval, operation log verification, and feedback entry. If any step is missing, the operation is deemed sensitive.

[0152] In this embodiment, numerical sensitivity is achieved through... To determine sensitivity, if the deviation is ≥10%, the numerical sensitivity threshold is triggered. Scenario sensitivity refers to a mismatch between the business scenario of the audit node and the scenario standards of the multi-dimensional sensitivity baseline. Operational sensitivity refers to inconsistencies between the auditer's operational behavior and the operational standards of the multi-dimensional sensitivity baseline, reflecting compliance issues in the audit process.

[0153] In this embodiment, the first comparative analysis compares the reviewer's permissions with the review process trajectory to determine whether the reviewer possesses the corresponding permissions and whether the operation trajectory meets the permission requirements, thus verifying the operational compliance of the review process. For example, the first comparative analysis of the promotion permission verification node is as follows: Personnel permissions: Possesses permission verification permissions; Review process trajectory: 2024-05-15 10:00 Conducted permission verification → 10:30 Submitted results; Comparison result: Permissions match the trajectory, no operational sensitivity. The first comparative analysis of the voting node is as follows: Personnel permissions: Unrelated personnel do not have voting review permissions; Review process trajectory: 2024-06-20 09:30 Participated in voting operations; Comparison result: Permissions contradict the trajectory, triggering operational sensitivity.

[0154] The second comparative analysis involves comparing the auditee's business data with the audit result data to determine whether the business data supports the audit conclusion and whether the conclusion is consistent with the data. The core of this analysis is to verify the authenticity of the results. For example, in the second comparative analysis of the initial review stage for promotion qualifications: Business data: 2.5 years of service; Audit result data: Qualified conclusion; Comparison result: The 2.5 years of service does not support the qualified conclusion, triggering numerical sensitivity.

[0155] The third comparative analysis involves comparing the audit process trajectory with the audit result data to determine whether the operation trajectory fully supports the audit conclusion, whether the conclusion generation conforms to the trajectory process, and the correlation between the core verification process and the result. For example: Audit process trajectory: only material retrieval operations, no data verification or opinion entry operations; Audit result data: conclusion is qualified; Comparison result: the trajectory is missing key links, cannot support the conclusion, and triggers operation sensitivity.

[0156] In this embodiment, the comparison result matrix is ​​arranged with the review node as the row and the comparison analysis type as the column. The three types of comparison analysis results are organized into a two-dimensional table, which intuitively shows the sensitive issues of each node in different comparison dimensions. This forms the basis for setting sensitivity labels and sensitivity values ​​in the future. Taking the partial comparison result matrix of promotion assessment as an example, as shown in Table 1:

[0157] Table 1

[0158] Review Node First comparison result (permissions - trajectory) Second comparison results (business data - results) Third control result (trajectory-outcome) Preliminary qualification review stage Match (encoding 1) Contradiction (Code 2, 2.5 years of service ≠ qualified) Match (encoding 1) Material review stage Match (encoding 1) Match (encoding 1) Missing (Code 3, no trajectory record) Permission verification node Contradiction (Code 2, No permission to operate) Match (encoding 1) Match (encoding 1)

[0159] In this embodiment, the control element is the content of each cell in the control result matrix, which is the smallest unit for assigning values ​​to the sensitive label and the initial sensitivity value.

[0160] In this embodiment, the sensitive label is assigned to the comparison element based on the result type (contradictory / missing) and the standard range matrix. The label can be single or in combination to reflect the attributes of the sensitive issue. For example, if the result corresponding to the material review node and the third comparison element is missing, the sensitive label is operational sensitive. Specifically, the sensitive label is automatically generated by calling the standard range matrix based on the comparison type and result type of the comparison element.

[0161] In this embodiment, the initial sensitivity value = the benchmark value of the corresponding result in the standard range matrix + the deviation rate correction value. The benchmark value is a pre-set value extracted from the benchmark value dictionary. For example, in the preliminary qualification review - second comparison - contradiction, the benchmark value is 0.8. If the deviation is 0.16, the initial sensitivity value is 0.96.

[0162] The beneficial effects of the above technical solution are as follows: by retrieving multi-dimensional sensitive baselines and aligning analysis threads, a comprehensive analysis mechanism covering numerical values, scenarios, and operations is constructed. Then, through three types of comparative analysis, a comparative result matrix is ​​integrated. Finally, sensitive labels and initial sensitive values ​​are set by combining a standard range matrix. This avoids the limitation of traditional sensitive analysis that only focuses on single numerical sensitivity. Furthermore, the structured matrix and standardized thresholds ensure the comprehensiveness and accuracy of sensitive issue identification, effectively improving the efficiency and accuracy of sensitive issue location in organizational business data analysis.

[0163] This invention provides a comprehensive AI-based analysis method for organizational business data, which adjusts the sensitivity analysis results based on the task feature set to obtain a sensitivity vector, including:

[0164] Input the core elements of the task feature set of the same review node and the corresponding sensitivity analysis results into the association model, and construct the feature-sensitivity association matrix by calculating the co-occurrence probability of feature items and sensitive labels;

[0165] Adjustment coefficients are set based on the weight values ​​of strongly correlated features in the feature-sensitivity correlation matrix;

[0166] If the feature term is positively correlated with the sensitive label, the initial sensitivity value is adjusted based on the first formula of the adjustment coefficient;

[0167] If the feature term is negatively correlated with the sensitive label, the initial sensitivity value is adjusted based on the second formula of the adjustment coefficient;

[0168] Determine the dependencies between review nodes in the analysis thread, calculate the influence of highly sensitive nodes on the corresponding review nodes, and accumulate it into each adjusted sensitivity value under the corresponding review node. The influence is obtained by multiplying the historical collaboration frequency between review nodes with the standard deviation of all initial sensitivity values ​​under the corresponding review node.

[0169] The sensitivity vector is obtained by arranging all the accumulated sensitivity values ​​and reference elements of each review node in the order of the analysis thread. The sensitivity vector contains the sensitivity value set of each review node in the corresponding analysis thread, and the sensitivity value set contains all the accumulated sensitivity values, sensitivity tags and reference elements under the corresponding review node.

[0170] In this embodiment, the association model is an algorithm model trained based on historical data, used to calculate the degree of association between task features and sensitive labels. For example, the association model is trained based on historical organizational analysis data.

[0171] In this embodiment, the contribution probability is the probability that the task feature and the sensitive label appear simultaneously, reflecting the degree of correlation between the two. The probability of occurrence is calculated by dividing the number of times the label appears when the feature appears by the total number of times the feature appears.

[0172] In this embodiment, the feature-sensitivity association matrix uses task features as rows and sensitive labels as columns, with each cell representing a probability of occurrence, to demonstrate the strength of the association between features and sensitivity.

[0173] In this embodiment, a strong correlation feature is a feature in the feature-sensitivity correlation matrix that has a co-occurrence probability ≥ a preset threshold with a certain sensitive label, indicating a close correlation. The preset threshold is 0.6.

[0174] In this embodiment, the weight values ​​are extracted from the task feature set-initial weight lookup table. For example, the initial weight of the numerical feature of tenure is 0.9, and the adjustment coefficient = initial weight of feature × co-occurrence probability.

[0175] In this embodiment, the first formula is: Adjusted sensitivity value = Initial sensitivity value × (1 + Adjustment coefficient). If the value is greater than 1, it is retained as 1.

[0176] Second formula: Adjusted sensitivity value = Initial sensitivity value × (1 - Adjustment coefficient).

[0177] In this embodiment, the dependency relationship is the influence of the result or status of one audit node on the sensitivity analysis result of another node. The dependency relationship is determined based on the node dependency relationship table. For example, the high sensitivity of the initial qualification review will make the material review analysis more stringent.

[0178] In this embodiment, a high-sensitivity node is a node in the analysis thread whose adjusted sensitivity value is greater than a preset high-sensitivity threshold, which is set to 0.8.

[0179] In this embodiment, the historical collaboration frequency is the number of times that highly sensitive nodes and target nodes simultaneously have sensitive issues in the historical assessment. The higher the frequency, the greater the collaborative impact. For example, the number of times that the initial qualification review and the material review were simultaneously sensitive was 15 times in the history.

[0180] In this embodiment, the standard deviation of the initial sensitivity value is the standard deviation of all initial sensitivity values ​​under the target node, and the influence degree = historical coordination frequency × standard deviation of the initial sensitivity value of the target node.

[0181] In this embodiment, the accumulation to the adjustment sensitivity value is to add the influence of the highly sensitive node to the adjustment sensitivity value of the target node. For example, if the adjustment sensitivity value of the target node is 0.8 and the influence is 0.3, the accumulation is 1.1. If it exceeds 1, it is taken as 1.

[0182] In this embodiment, the sensitivity vector is an ordered data structure containing sensitive information of each node, arranged in the order of the analysis thread. The sensitivity value set is a collection of all accumulated sensitivity values, corresponding sensitivity tags, and comparison elements under a single review node, which fully records the sensitive details of the node. For example, the sensitivity value set of the qualification review node includes: accumulated sensitivity value 1.0 and accumulated sensitivity value 0.3.

[0183] The beneficial effects of the above technical solution are as follows: By constructing a feature-sensitivity correlation matrix and adjusting the sensitivity value by combining positive / negative correlation, and then incorporating the dependency influence between review nodes, the resulting sensitivity vector accurately reflects the sensitivity of a single node and also covers the synergistic effect between nodes, thus solving the shortcomings of traditional sensitivity analysis; supported by the quantitative calculation of historical data and business rules, the sensitivity value is made more in line with the actual organizational work, providing high-precision core data for the subsequent generation of structured judgment reports and the tracing of sensitive issues, effectively improving the comprehensiveness and accuracy of organizational work data judgment.

[0184] This invention provides a method for comprehensive AI-based analysis and judgment of organizational business data, which significantly processes the review nodes in the corresponding analysis thread according to the aforementioned sensitivity vector, including:

[0185] The sensitivity value set of each review node contained in the sensitivity vector is analyzed dimensionally to extract the comprehensive sensitivity level, the combination of sensitive types, and the identifier of the associated node.

[0186] Based on the level-combination-identifier-policy mapping table, matching identifiers are obtained and assigned according to the extraction results. During the assignment process, each sub-identifier in the matching identifier is significantly overwritten with the specified information in the corresponding bidirectional information.

[0187] In this embodiment, dimensional analysis involves classifying and decomposing multiple types of sensitive data in the sensitive data set, extracting key information from three core dimensions: sensitivity level, sensitivity type, and related nodes. For example, dimensional analysis is performed on the sensitive value set of the qualification review node.

[0188] Extract sensitivity level information from the accumulated sensitivity values ​​of 1.0 / 0.3;

[0189] Extract sensitivity type dimension information from the tags: numerical sensitivity / operational sensitivity;

[0190] Extract the dimension information of the associated nodes from the material review node dependency record associated with the reference element.

[0191] In this embodiment, the overall sensitivity level is determined based on the average of the sum of all accumulated sensitivity values ​​in the sensitive value set of the audit node. The level is based on the node-value-level lookup table, which reflects the overall sensitivity of the node. The node-value-level lookup table contains the overall sensitivity level of the corresponding node at different average values. It is pre-set and can be directly matched. For example, if the average value is 0.8, it corresponds to a high sensitivity level.

[0192] In this embodiment, the combination of sensitive types is selected from the sensitive tags of the sensitive value set of the review node, and one or two types of sensitive types with the highest frequency or the highest corresponding sensitive value are selected to form the core sensitive type combination. For example, the sensitive tags of the qualification review node are numerical sensitivity with a value of 1 and operational sensitivity with a value of 0.3. In this case, the label numerical sensitivity with a sensitivity value ≥ 0.5 is selected as the combination of sensitive types to highlight.

[0193] In this embodiment, the unique identifiers of the one or two other review nodes that have a sensitive dependency relationship with the current review node and contribute the most to the sensitivity value of the current node are highlighted, reflecting the sensitivity transmission path between nodes. All related nodes and their corresponding influence values ​​of the current node are retrieved from the node dependency database; they are sorted in descending order of influence value, and nodes with an influence value ≥ 0.1 are selected as highlighted related nodes, and a unique ID is added to each identifier to obtain the highlighted related node representation.

[0194] In this embodiment, the level-combination-identifier-policy mapping table is pre-constructed based on organizational guidelines and historical sensitive handling cases, forming a comprehensive sensitivity level-highlighted sensitivity type combination-highlighted associated node identifier → matching identifier-processing policy correspondence table. Some examples are shown in Table 2:

[0195] Table 2 Management Dimension Mapping Table

[0196] Overall Sensitivity Level Highlighting sensitive type combinations Highlight related node identifiers Matching identifier Processing strategy Gao Min Numerical sensitivity none High sensitivity - number - none Red bold highlight + pop-up notification Zhongmin Numerical sensitivity + Contextual sensitivity Preliminary qualification review stage Zhongmin-Data Field-Capital Initial Orange highlighting + related node links Gao Min Operation Sensitive Permission verification node Gao Min-Cuo-Quan He Red flashing marker + operation log traceability

[0197] In this embodiment, the specified information is the specific information unit directly associated with the sub-representation in the bidirectional information of the audit node. It is a significantly covered object, such as: medium sensitivity: sub-identifier → accumulated sensitivity value 0.7 / 0.5 in bidirectional information. A sub-identifier-bidirectional information field correspondence table is established. The system automatically locates the corresponding field in the bidirectional information based on the sub-identifier and extracts the specific information as the specified information.

[0198] The beneficial effects of the above technical solution are as follows: by performing dimensional analysis on the sensitive value set of each review node in the sensitive vector, the comprehensive sensitivity level, the combination of sensitive types and the associated node identifiers are accurately extracted, and then the exclusive identifiers are matched and associated with bidirectional information in combination with the preset mapping table to carry out significant coverage. The significant processing results can help the assessment personnel quickly focus on the core sensitive issues, reduce the interference of invalid information, and provide clear and accurate sensitive information materials for the generation of subsequent structured assessment reports.

[0199] This invention provides a comprehensive AI-based analysis method for organizational data, generating a structured analysis report based on all significant processing results, including:

[0200] Dimensional analysis is performed on all significant processing results to identify the judgment dimension system. Based on the judgment dimension system, the significant processing results are divided into a dimension feature layer, an association logic layer, and a conclusion output layer. The dimension feature layer is related to the core sensitive identifier set under each judgment dimension, the association logic layer is related to the sensitive transmission and dependency relationship between review nodes, and the conclusion output layer is related to the preliminary judgment conclusion of each review node.

[0201] Extract the set of salient identifiers for each judgment dimension in the dimensional feature layer, perform semantic enhancement processing on the bidirectional information associated with each identifier, generate a sensitive identifier-associated information mapping graph, calculate the aggregation density of highly sensitive identifiers under each judgment dimension based on the mapping graph, and determine the feature compliance level of the corresponding dimension by combining the feature distribution threshold of similar judgment dimensions in history.

[0202] Based on the association logic layer, a dependency network of audit nodes is constructed. Highly sensitive propagation paths in the dependency network are mined, the sensitive impact weight of each path is calculated, and the top N9 key sensitive transmission paths with the highest impact weight are selected to generate a sensitive transmission logic summary.

[0203] The system calls upon a pre-defined rule library for expressing organizational conclusions to transform the preliminary judgments from the output layer into standardized conclusions that conform to organizational writing conventions, and marks each standardized conclusion with a prominent identifier and a source link.

[0204] The feature compliance level, sensitive transmission logic summary, standardized conclusions, and significant identifier traceability links are automatically filled into the report template to generate a structured analysis report.

[0205] In this embodiment, the judgment dimension system is a framework of three core judgment dimensions that are fixed in the organization business. Each dimension includes exclusive review nodes, sensitive types and judgment standards, all of which are pre-set.

[0206] In this embodiment, the dimension feature layer focuses on the core sensitive representation set under each dimension, that is, the sensitive features associated with the significant matching representation set of all review nodes in each dimension. For example, the dimension feature layer of the management dimension: core sensitive identifier set: preliminary qualification review, material review, and permission verification, associated sensitive features: sensitivity of tenure numerical value, sensitivity of assessment score numerical value and review scenario sensitivity, and sensitivity of non-authorized operation.

[0207] In this embodiment, the core sensitive identifier set is the core component of the dimensional feature layer, referring to the set of significant matching identifiers that appear frequently and have high sensitivity values ​​under each judgment dimension.

[0208] In this embodiment, the association logic layer focuses on the sensitive transmission and dependency relationships of the review nodes, that is, how highly sensitive nodes affect other nodes through data references and process connections, reflecting the propagation logic of sensitive issues.

[0209] In this embodiment, historical dependency records are retrieved from the node dependency database, and the sensitive transmission direction and dependency type are determined by combining them with the associated node identifiers in the current significant processing results. → represents transmission and ← represents dependency. Transmission relationship: preliminary qualification review node for promotion → material review node, dependency type is data reference dependency; dependency relationship: training qualification review node for development ← training node, dependency type is process pre-requirement dependency.

[0210] In this embodiment, the conclusion output layer focuses on the preliminary judgment conclusions of each review node. Based on sensitivity analysis and salience processing, the preliminary judgment on whether each node is compliant and whether there are sensitive issues is the basis for subsequent standardized conclusions. For example, the preliminary judgment conclusion of the management dimension permission verification node is: there are unauthorized personnel performing review operations, which does not meet the permission management requirements.

[0211] In this embodiment, the preliminary judgment is a preliminary judgment generated for a single review node without standardized expression. It is directly related to the sensitive characteristics of that node and serves as the original material for subsequent transformation into standardized conclusions.

[0212] In this embodiment, the sensitive identifier-associated information mapping graph uses prominent identifiers as nodes and associated information as edges. Based on Neo4j, a visual graph is constructed to intuitively display the association relationship between each identifier and bidirectional information. For example, nodes are distinguished by different colors to differentiate sensitivity levels: high sensitivity is represented by red, medium sensitivity by orange, and low sensitivity by yellow.

[0213] In this embodiment, the aggregation density of highly sensitive markers is the proportion of the number of highly sensitive markers to the total number of significant markers in each judgment dimension.

[0214] In this embodiment, the feature distribution threshold is the average ± standard deviation of the aggregation density of highly sensitive identifiers in historical projects with the same evaluation dimension over the past 3 to 5 years, serving as a reference benchmark for determining the compliance of the current dimension. Specifically:

[0215] A score ≤ (mean - standard deviation) of the benchmark is considered excellent.

[0216] If the mean minus the standard deviation is less than or equal to the reference baseline, and the mean plus the standard deviation is considered acceptable;

[0217] Otherwise, it will be considered unqualified.

[0218] The above comparison results represent the feature compliance level, namely: excellent, qualified, and unqualified.

[0219] In this embodiment, the audit node dependency network uses audit nodes as nodes and dependencies as edges. A visual network is constructed using Neo4j to fully present the connection and influence relationship between nodes in the analysis thread, which is the basis for mining the propagation path.

[0220] In this embodiment, the highly sensitive propagation path starts from the highly sensitive node and propagates to other nodes through dependencies. The path must include: highly sensitive node -- intermediate node -- endpoint node. For example, path 1: preliminary qualification review → material review → permission verification.

[0221] In this embodiment, the sensitive influence weight is the sum of the influence degrees among all remaining nodes.

[0222] In this embodiment, N9 is the total number of critical sensitive transmission paths divided by 2 and then rounded down.

[0223] In this embodiment, the sensitive transmission logic summary is a text that extracts and summarizes the core logic of key sensitive transmission paths, including the path starting point, transmission dependency type, main influencing nodes, and conclusions on the propagation of sensitive issues, which conforms to the organizational writing standards.

[0224] In this embodiment, the rule base for expressing organizational conclusions is a database pre-built based on organizational documents and organizational writing standards. Part of the contents of the rule database are shown in Table 3:

[0225] Table 3

[0226] Preliminary conclusion type Standardized expression template Numerical sensitivity (years of service) The XX node has a sensitive numerical value regarding the length of service. The actual length of service is XX years, which does not meet the requirement of Article X of the "Regulations on Selection and Appointment" that the employee must have served for XX years. Sensitive operation (non-authorization-based review) The XX node has sensitive operational requirements, and the reviewer XX does not have the necessary review authority, which does not comply with Article Y of the "Organizational Work Review Authority Management Measures". The scenario is sensitive (the public notice period was insufficient). The XX node has a sensitive scenario, and the public notice period is XX days, which does not comply with Article Z of the "Selection and Public Notice Measures" which requires a full XX working days of public notice.

[0227] In this embodiment, the standardized conclusion is obtained by substituting the preliminary judgment of the conclusion output layer into the corresponding template in the organization conclusion expression rule database.

[0228] In this embodiment, a URL link is set for each standardized conclusion to be prominently marked as a traceability link. For example, after the standardized conclusion, a traceability link is marked: View sensitive identifier details. Clicking the link will jump to the prominent identifier details page of the qualification preliminary review node, which displays the matching identifier: High Sensitivity-Number-None, prominent coverage: red bold tenure 2.5 years, sensitivity value: 1.0.

[0229] In this embodiment, the report template is pre-set and can be directly called and used.

[0230] In this embodiment, the structured assessment report automatically fills the analysis results, such as feature compliance level, sensitive transmission logic summary, standardized conclusions and traceability links, into the report template placeholders to generate a complete report. It has the characteristics of clear structure, standardized expression, traceability and archiving, and serves as the formal basis for organizational business decisions.

[0231] The beneficial effects of the above technical solution are as follows: by analyzing significant processing results in layers according to the judgment dimensions, combining semantic enhancement, graph construction and compliance level determination to deepen feature analysis, and then by relying on network mining to discover key sensitive transmission paths, standardizing conclusion expressions and establishing traceability links, a structured judgment report is finally automatically generated, which greatly reduces the time for manual processing and improves the efficiency of report generation.

[0232] This invention provides an artificial intelligence-based comprehensive analysis and judgment device for organizational business data, such as... Figure 2 As shown, it includes:

[0233] The thread parsing module is used to collect target information for each target person and decompose the target information according to the judgment dimensions to obtain the analysis thread for each judgment dimension. Each analysis thread contains at least one audit node, and each audit node involves bidirectional information of several audit indicators. The bidirectional information includes the personnel permissions and audit process trajectory of the auditing party, and the business data and audit result log of the audited party. The judgment dimensions include management-related dimensions, construction-related dimensions, and work-related dimensions.

[0234] The feature set acquisition module is used to acquire the judgment requirements input by the receiving and using unit, and to perform judgment task analysis on the bidirectional information of each review node based on the judgment requirements to obtain the task feature set.

[0235] The sensitivity analysis module is used to perform sensitivity analysis on all audit nodes in each analysis thread, and adjust the sensitivity analysis results based on the task feature set to obtain a sensitivity vector;

[0236] The report generation module is used to perform significant processing on the audit nodes in the corresponding analysis thread according to the sensitivity vector, and generate a structured assessment report based on all significant processing results.

[0237] The beneficial effects of the above technical solution are as follows: by systematically collecting two-way information on organizational work, breaking down and analyzing threads according to the three dimensions of management, construction, and work, dynamically extracting accurate task characteristics in combination with the needs of user units, and locating key sensitive nodes through multi-dimensional sensitivity analysis and dynamic adjustment, a structured analysis report that conforms to the norms of organizational work documents is finally generated, which significantly improves the efficiency, accuracy, and standardization of organizational work data analysis and provides reliable data support for organizational work decision-making.

[0238] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for comprehensive artificial intelligence analysis of organizational business data, characterized in that, include: Step 1: Collect target information for each target person and break down the target information according to the judgment dimensions to obtain the analysis thread for each judgment dimension. Each analysis thread contains at least one audit node, and each audit node involves bidirectional information of several audit indicators. The bidirectional information includes the personnel permissions and audit process trajectory of the auditor, and the business data and audit result log of the auditee. The judgment dimensions include management-related dimensions, construction-related dimensions, and work-related dimensions. Step 2: Obtain the assessment requirements input by the receiving and using unit, and perform assessment task analysis on the bidirectional information of each review node based on the assessment requirements to obtain the task feature set; Step 3: Perform sensitivity analysis on all audit nodes in each analysis thread, and adjust the sensitivity analysis results based on the task feature set to obtain a sensitivity vector; Step 4: Perform significant processing on the audit nodes in the corresponding analysis thread according to the sensitivity vector, and generate a structured analysis report based on all significant processing results.

2. The method for comprehensive analysis and judgment of organizational business data using artificial intelligence according to claim 1, characterized in that, The target information is broken down according to the judgment dimensions to obtain the analysis thread for each judgment dimension, including: Information is extracted from the target information based on each judgment dimension in the preset dimension set to obtain bidirectional information of different target tasks under each judgment dimension; The bidirectional information is sorted according to time sequence and task type to obtain analysis threads for corresponding judgment dimensions, wherein each target task corresponds to an audit node.

3. The method for comprehensive artificial intelligence analysis of organizational data according to claim 1, characterized in that, Based on the aforementioned assessment requirements, a task feature set is obtained by analyzing the bidirectional information of each review node, including: The judgment requirements are semantically parsed to obtain several requirement keywords, and the time sequence of each requirement keyword is analyzed to obtain several global elements and local elements. The elements are sorted according to the time sequence of the first appearance of each element and the task coverage of the time sequence of each element to obtain an element vector. The element vectors are matched with the vector-standard comparison table to obtain the first judgment standard for each element. At the same time, the element vectors are input into the vector analysis model to obtain the second judgment standard for each element. The judgment accuracy of the corresponding elements is determined according to the first judgment standard and the second judgment standard, and the first element with a judgment accuracy greater than the corresponding preset accuracy is selected. At the same time, the second element with a ranking value greater than the preset value in the element vector is selected, and the first element and the second element are used as the core elements. The structured data in the bidirectional information of each audit node is normalized and mapped into numerical features, and the unstructured data is semantically analyzed and mapped into semantic features. According to the assessment target type of the assessment requirements, the numerical features and semantic features are classified and analyzed, the mutual information value between each classification feature and each core element is established, and an initial weight is set for each classification feature based on the weight ratio of the core element weight, the reviewer's authority-related features and the reviewee's result features. The initial feature set is obtained by filtering the classification features whose initial weights are greater than the preset weights.

4. The method for comprehensive artificial intelligence analysis of organizational business data according to claim 3, characterized in that, After obtaining the initial feature set, the following is also included: Each remaining feature is sequentially compared with each valid feature in the set of valid features for similar historical needs to obtain a similarity set; If the number of qualified elements in the similarity set is greater than the standard set number, the remaining features will be classified into the initial feature set to obtain the task feature set.

5. The method for comprehensive analysis and judgment of organizational business data using artificial intelligence according to claim 1, characterized in that, Perform sensitivity analysis on all audit nodes in each analysis thread, including: The multidimensional sensitivity baseline of each analysis thread is retrieved, and the corresponding analysis threads are aligned to obtain the multidimensional sensitivity analysis mechanism of each audit node. The multidimensional sensitivity analysis mechanism is related to numerical sensitivity, scene sensitivity, and operation sensitivity. According to the aforementioned multidimensional sensitivity analysis mechanism, the personnel permissions and review process trajectory in the corresponding two-way information are subjected to a first comparative analysis, the business data and review result data are subjected to a second comparative analysis, and the review process trajectory and review result data are subjected to a third comparative analysis. Based on the results of the first, second, and third control analyses, a control result matrix was constructed. Based on the comparison result matrix and the standard range matrix of the audit nodes, a sensitive label and an initial sensitive value are set for each comparison element. The sensitive label is related to one or more of the following: numerical sensitivity, scenario sensitivity, and operational sensitivity.

6. The method for comprehensive artificial intelligence analysis of organizational business data according to claim 4, characterized in that, The sensitivity vector is obtained by adjusting the sensitivity analysis results based on the task feature set, including: Input the core elements of the task feature set of the same review node and the corresponding sensitivity analysis results into the association model, and construct the feature-sensitivity association matrix by calculating the co-occurrence probability of feature items and sensitive labels; Adjustment coefficients are set based on the weight values ​​of strongly correlated features in the feature-sensitivity correlation matrix; If the feature term is positively correlated with the sensitive label, the initial sensitivity value is adjusted based on the first formula of the adjustment coefficient; If the feature term is negatively correlated with the sensitive label, the initial sensitivity value is adjusted based on the second formula of the adjustment coefficient; Determine the dependencies between review nodes in the analysis thread, calculate the influence of highly sensitive nodes on the corresponding review nodes, and accumulate it into each adjusted sensitivity value under the corresponding review node. The influence is obtained by multiplying the historical collaboration frequency between review nodes with the standard deviation of all initial sensitivity values ​​under the corresponding review node. The sensitivity vector is obtained by arranging all the accumulated sensitivity values ​​and reference elements of each review node in the order of the analysis thread. The sensitivity vector contains the sensitivity value set of each review node in the corresponding analysis thread, and the sensitivity value set contains all the accumulated sensitivity values, sensitivity tags and reference elements under the corresponding review node.

7. The method for comprehensive artificial intelligence analysis of organizational business data according to claim 1, characterized in that, The review nodes in the corresponding analysis thread are significantly processed according to the aforementioned sensitivity vector, including: The sensitivity value set of each review node contained in the sensitivity vector is analyzed dimensionally to extract the comprehensive sensitivity level, the combination of sensitive types, and the identifier of the associated node. Based on the level-combination-identifier-policy mapping table, matching identifiers are obtained and assigned according to the extraction results. During the assignment process, each sub-identifier in the matching identifier is significantly overwritten with the specified information in the corresponding bidirectional information.

8. The method for comprehensive analysis and judgment of organizational business data using artificial intelligence according to claim 7, characterized in that, A structured analysis report is generated based on all significant processing results, including: Dimensional analysis is performed on all significant processing results to identify the judgment dimension system. Based on the judgment dimension system, the significant processing results are divided into a dimension feature layer, an association logic layer, and a conclusion output layer. The dimension feature layer is related to the core sensitive identifier set under each judgment dimension, the association logic layer is related to the sensitive transmission and dependency relationship between review nodes, and the conclusion output layer is related to the preliminary judgment conclusion of each review node. Extract the set of salient identifiers for each judgment dimension in the dimensional feature layer, perform semantic enhancement processing on the bidirectional information associated with each identifier, generate a sensitive identifier-associated information mapping graph, calculate the aggregation density of highly sensitive identifiers under each judgment dimension based on the mapping graph, and determine the feature compliance level of the corresponding dimension by combining the feature distribution threshold of similar judgment dimensions in history. Based on the association logic layer, a dependency network of audit nodes is constructed. Highly sensitive propagation paths in the dependency network are mined, the sensitive impact weight of each path is calculated, and the top N9 key sensitive transmission paths with the highest impact weight are selected to generate a sensitive transmission logic summary. The system calls upon a pre-defined rule library for expressing organizational conclusions to transform the preliminary judgments from the output layer into standardized conclusions that conform to organizational writing conventions, and marks each standardized conclusion with a prominent identifier and a source link. The feature compliance level, sensitive transmission logic summary, standardized conclusions, and significant identifier traceability links are automatically filled into the report template to generate a structured analysis report.

9. A device for comprehensive analysis and judgment of organizational business data using artificial intelligence, characterized in that, include: The thread parsing module is used to collect target information for each target person and decompose the target information according to the judgment dimensions to obtain the analysis thread for each judgment dimension. Each analysis thread contains at least one audit node, and each audit node involves bidirectional information of several audit indicators. The bidirectional information includes the personnel permissions and audit process trajectory of the auditing party, and the business data and audit result log of the audited party. The judgment dimensions include management-related dimensions, construction-related dimensions, and work-related dimensions. The feature set acquisition module is used to acquire the judgment requirements input by the receiving and using unit, and to perform judgment task analysis on the bidirectional information of each review node based on the judgment requirements to obtain the task feature set. The sensitivity analysis module is used to perform sensitivity analysis on all audit nodes in each analysis thread, and adjust the sensitivity analysis results based on the task feature set to obtain a sensitivity vector; The report generation module is used to perform significant processing on the audit nodes in the corresponding analysis thread according to the sensitivity vector, and generate a structured assessment report based on all significant processing results.

Images