Special audit document automatic generation and management method and system

Abstract

The present application relates to the technical field of auditing data processing, and more particularly to a special audit manuscript automatic generation and management method and system. The logical dependency constraints between nodes are determined based on the input information of the audit task, and the manuscript documents are generated by filling in through correlation verification and historical data reasoning. The content fingerprints of the document nodes are calculated, the influence path is determined according to the logical dependency constraints when revising, and the associated data is updated, and the version change link is managed. The automatic generation and intelligent maintenance of the audit manuscript are realized, and the accuracy of data filling and the traceability of revision management are improved.

Description

Technical Field

[0001] This invention relates to the field of audit data processing technology, and in particular to a method and system for automatically generating and managing special audit working papers. Background Technology

[0002] In the field of specialized audits, the preparation and management of audit working papers is a core aspect of the audit process. Currently, auditors typically collect audit evidence manually or with the aid of general office software, perform audit procedures, and record audit findings, ultimately forming structured audit working paper documents, based on auditing standards and project requirements. This process heavily relies on the professional judgment and experience of auditors; from identifying audit priorities and designing audit procedures to collecting and analyzing data, everything is done manually. Version control of audit working papers is mostly conducted through file naming rules or simple version control systems, and revision records and the impact of changes are usually traced through manual notes or memory.

[0003] Existing practices have significant limitations. The generation of audit working papers involves a large amount of repetitive, structured data collection and completion work, and there are often predetermined logical dependencies between different audit procedures or data nodes. Under a manual operation model, it is difficult to systematically verify the semantic consistency of data across nodes, which can easily lead to data inconsistencies within the working papers or the absence of key evidence. For example, the substantive procedure conclusions for a certain account are logically related to the conclusions of related internal control tests; manual verification is prone to omissions, affecting audit quality and efficiency.

[0004] When audit working papers require revision, existing methods struggle to effectively manage the impact of changes. Audit working papers are interconnected; a change in one piece of data can trigger a chain reaction of adjustments to related data. Conventional document version control only records changes to the overall document, failing to accurately track and describe how changes to specific data nodes affect other nodes along logical dependency paths. This leads to difficulties in version tracing, unclear audit trails, increased complexity in quality control and review, and hinders the accumulation and reuse of audit experience. Summary of the Invention

[0005] The embodiments of the present invention provide a method and system for automatically generating and managing special audit working papers, which can solve the problems in the prior art.

[0006] A first aspect of this invention provides a method for automatically generating and managing special audit working papers, comprising:

[0007] Obtain audit task input information, which includes audit object identifier and audit type identifier, and determine the logical dependency constraints between nodes in the audit type identifier;

[0008] Based on semantic consistency verification rules, cross-node association verification is performed on the dataset to be audited, and missing data identifiers are generated. According to the node position corresponding to the missing data identifier, historical audit records that match the business attribute category of the current audit object identifier are retrieved. Data content of the corresponding node position in the historical audit records is extracted and the missing position is inferred and filled in combination with the logical dependency constraints to obtain the draft document.

[0009] Content fingerprint calculation is performed on the data content of each node in the draft document to generate a node fingerprint set. When a revision operation is received for the draft document, the impact propagation path of the revision operation is determined based on the logical dependency constraint.

[0010] The consistency of the associated node position sequence is verified along the influence propagation path. When a logical dependency is detected between the associated node data and the revised content, the associated node data is updated according to the logical dependency constraint and the node fingerprint is recalculated.

[0011] The node fingerprint sets before and after the revision, the impact propagation path, and the audit task input information are linked and stored to form a version change management link.

[0012] Determining the logical dependency constraints between nodes in the audit type identifier includes:

[0013] Extract the node structure configuration information corresponding to the audit type identifier from the working paper template library. The node structure configuration information defines the identifier, hierarchical relationship and data attribute type of each node in the working paper document.

[0014] Based on the hierarchical relationship, identify node pairs with data reference relationships, determine whether the data content of the source node can be used as the data generation input of the target node according to the compatibility rules between the data attribute types, and establish a dependency relationship marker between the source node and the target node when the determination result is yes.

[0015] For node pairs with established dependency relationships, perform dependency direction analysis to determine the data flow direction between source nodes and target nodes. When the data generation of a target node depends on multiple source nodes, mark the multiple source nodes as a set of preceding dependent nodes. When the data content of a source node is called by the data generation process of multiple target nodes, mark the multiple target nodes as a set of subsequent dependent nodes.

[0016] Traverse all nodes in the draft document, extract the set of preceding dependent nodes and the set of subsequent dependent nodes for each node, and associate and store the set of preceding dependent nodes, the set of subsequent dependent nodes and the node identifier to form the logical dependency constraint between nodes.

[0017] Based on semantic consistency verification rules, cross-node association verification is performed on the dataset to be audited, and missing data identifiers are generated. Based on the node position corresponding to the missing data identifier, historical audit records matching the business attribute category of the current audit object identifier are retrieved, including:

[0018] Based on the set of preceding dependent nodes of each node in the logical dependency constraint, determine the list of required data fields for each node;

[0019] Iterate through the data content of each node in the dataset to be audited. When the actual data field identifier of each node is missing the data field identifier in the list of required data fields, record the node identifier and the missing data field identifier, and generate the missing data identifier.

[0020] Extract the node identifier and data field identifier from the missing data identifier, query the set of preceding dependent nodes corresponding to the node identifier from the logical dependency constraint, determine whether the data content of each preceding dependent node in the set of preceding dependent nodes is complete, and when the determination result is yes, mark the missing data identifier as an independent missing type.

[0021] For missing data identifiers marked as independent missing, extract the business attribute category features corresponding to the current audit object identifier; retrieve historical audit records from the historical audit working paper knowledge base that match the business attribute category features of the current audit object identifier.

[0022] Extracting the data content of the corresponding node positions from the historical audit records and using the logical dependency constraints to infer and fill in the missing positions, the resulting working paper document includes:

[0023] Based on the node identifier and data field identifier in the missing data identifier, extract historical data content from the historical audit records where both the node identifier and data field identifier match, and use the historical data content as the initial filling data.

[0024] Extract the set of preceding dependent nodes corresponding to the node with the missing data identifier from the logical dependency constraint, and obtain the current data content of each preceding dependent node in the set of preceding dependent nodes in the dataset to be audited;

[0025] The initial data is used to perform semantic compatibility verification with the current data content. The semantic compatibility verification determines whether the data format and numerical range of the initial data are logically conflicting with the current data content.

[0026] When the semantic compatibility verification result is that there is no logical conflict, the initial padding data is used as the final padding data and filled into the node position corresponding to the missing data identifier.

[0027] When the semantic compatibility verification result indicates a logical conflict, the data calculation relationship rules between the node corresponding to the missing data identifier and the set of preceding dependent nodes are extracted from the logical dependency constraints. Logical reasoning calculations are performed based on the data calculation relationship rules and the current data content to generate reasoning fill data. The reasoning fill data is then filled into the node position corresponding to the missing data identifier.

[0028] Content fingerprints are calculated for the data content of each node in the draft document to generate a node fingerprint set. When a revision operation is received for the draft document, the impact propagation path of the revision operation is determined based on the logical dependency constraints, including:

[0029] Traverse all nodes in the draft document, extract the data content of each node and perform hash operation to generate the node fingerprint value corresponding to each node, associate and store the node identifier of each node with the corresponding node fingerprint value to form a node fingerprint set.

[0030] When a revision operation is received for the draft document, the target node identifier and the revised data content in the revision operation are parsed, and a hash operation is performed on the revised data content to generate a revised fingerprint value;

[0031] Extract the original node fingerprint value corresponding to the target node identifier from the node fingerprint set, compare the revised fingerprint value with the original node fingerprint value, and when the comparison result is inconsistent, confirm that the revision operation has caused the data content of the target node to change;

[0032] Extract the set of subsequent dependent nodes corresponding to the target node identifier from the logical dependency constraint. For each subsequent dependent node in the set of subsequent dependent nodes, recursively extract the set of subsequent dependent nodes corresponding to each subsequent dependent node. Construct a multi-level dependency propagation link starting from the target node identifier. Arrange all node identifiers contained in the multi-level dependency propagation link according to the dependency hierarchy to generate the influence propagation path.

[0033] Consistency verification is performed on the sequence of associated node positions along the influence propagation path. When a logical dependency is detected between the associated node data and the revised content, the associated node data is updated according to the logical dependency constraint, and the node fingerprint is recalculated, including:

[0034] According to the hierarchy of node identifiers in the propagation path, extract the identifiers of each associated node and the corresponding associated node data content in sequence.

[0035] For the associated node identifier at the current level, obtain the actual data of the predecessor nodes of each predecessor node in the corresponding predecessor dependency node set in the draft document;

[0036] Extract the data constraint rules between the associated node identifier and the set of preceding dependent nodes from the logical dependency constraints, and calculate the expected associated node data value based on the data constraint rules and the actual data of the preceding nodes;

[0037] The associated node data content is compared with the expected associated node data value. When the comparison result is inconsistent, it is determined that the associated node data content has a logical dependency relationship with the revised content and needs to be updated. The updated associated node data is recalculated and generated according to the data constraint rules and the actual data of the preceding node, and the original associated node data content corresponding to the node position of the associated node identifier is replaced.

[0038] Perform a hash operation on the updated associated node data to generate an updated node fingerprint value, and update the node fingerprint value corresponding to the associated node identifier in the node fingerprint set to the updated node fingerprint value.

[0039] Continue processing the associated node identifiers at the next level in the influence propagation path until the consistency verification and data update of all associated node identifiers in the influence propagation path are completed.

[0040] The version change management chain is formed by establishing and storing the node fingerprint sets before and after the revision, the impact propagation path, and the audit task input information in a related manner, including:

[0041] The difference between the node fingerprint set before revision and the node fingerprint set after revision is compared, the node identifiers whose node fingerprint values ​​have changed are extracted, and the node identifiers whose node fingerprint values ​​have changed, along with the corresponding node fingerprint values ​​before and after revision, are combined to form a set of changed node records.

[0042] Extract the operation timestamp from the revision operation, and encapsulate the operation timestamp, the target node identifier, the revised data content, the set of change node records, and the impact propagation path in a structured manner to generate a version change record;

[0043] Extract the audit object identifier and audit task identifier from the audit task input information, and establish an association index relationship between the audit object identifier, the audit task identifier and the version change record;

[0044] The version change records and the associated index relationships are stored in the version change management database. The associated index relationships support the retrieval of all corresponding version change records based on the audit object identifier and the audit task identifier, thus forming the version change management chain.

[0045] A second aspect of this invention provides a system for automatically generating and managing special audit working papers, comprising:

[0046] An audit task unit is used to acquire audit task input information, which includes audit object identifier and audit type identifier, and to determine the logical dependency constraints between nodes in the audit type identifier.

[0047] The data filling unit is used to perform cross-node association verification on the dataset to be audited based on semantic consistency verification rules and generate missing data identifiers. According to the node position corresponding to the missing data identifier, it retrieves historical audit records that match the business attribute category of the current audit object identifier, extracts the data content of the corresponding node position in the historical audit records and combines the logical dependency constraints to infer and fill the missing positions to obtain the draft document.

[0048] The fingerprint generation unit is used to perform content fingerprint calculation on the data content of each node in the draft document, generate a set of node fingerprints, and when a revision operation is received for the draft document, determine the impact propagation path of the revision operation based on the logical dependency constraint.

[0049] The influence propagation unit is used to perform consistency verification on the sequence of associated node positions along the influence propagation path. When a logical dependency relationship is detected between the associated node data and the revised content, the associated node data is updated according to the logical dependency constraint and the node fingerprint is recalculated.

[0050] The version management unit is used to establish an association between the node fingerprint sets before and after the revision, the impact propagation path, and the audit task input information, forming a version change management link.

[0051] A third aspect of the present invention provides an electronic device, comprising:

[0052] processor;

[0053] Memory used to store processor-executable instructions;

[0054] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0055] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0056] The beneficial effects of this application are as follows:

[0057] This method can automatically generate structured audit working papers, significantly improving the efficiency and standardization of working paper preparation. By parsing the logical dependency constraints in the audit task input information, the system can automatically construct an audit process framework, ensuring that the working paper generation process strictly follows the preset audit logic and data association rules, avoiding omissions and errors caused by manually sorting out dependencies.

[0058] The system possesses intelligent data completion and verification capabilities. Based on semantic consistency rules, it performs cross-node correlation verification, accurately identifying missing items in the audit dataset. By retrieving matching historical audit records and performing inference and filling based on logical dependency constraints, it can generate coherent and logically consistent preliminary audit drafts even with incomplete data, effectively reducing the hindering impact of missing data on the audit process.

[0059] This method enables intelligent impact analysis and version control for draft revisions. Any revision operation triggers the system to automatically analyze its impact along logical dependency paths and perform consistency verification and automatic updates on related nodes. Combined with content fingerprinting technology, the system can accurately track the specific content changes of each revision and associate and store the change information, impact paths, and original tasks, forming a complete and traceable version management chain, greatly enhancing the transparency and verifiability of the audit process.

[0060] Ultimately, through an integrated process of automated generation, intelligent filling, dependency maintenance, and refined version management, this method ensures high quality, high consistency, and strong traceability of audit working papers. This not only significantly reduces manual operations and repetitive work for auditors but also strengthens the rigor and reliability of audit work from a mechanism perspective, providing solid and credible data and process support for audit conclusions. Attached Figure Description

[0061] Figure 1 A flowchart illustrating the process for automatically generating and managing special audit working papers;

[0062] Figure 2 This is a schematic diagram of the cross-node correlation verification and missing data identification process. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0065] Figure 1 This is a flowchart illustrating the automatic generation and management method for special audit working papers according to an embodiment of the present invention. Figure 1 As shown, the methods for automatically generating and managing special audit working papers include:

[0066] Obtain audit task input information, which includes audit object identifier and audit type identifier, and determine the logical dependency constraints between nodes in the audit type identifier;

[0067] Based on semantic consistency verification rules, cross-node association verification is performed on the dataset to be audited, and missing data identifiers are generated. According to the node position corresponding to the missing data identifier, historical audit records that match the business attribute category of the current audit object identifier are retrieved. Data content of the corresponding node position in the historical audit records is extracted and the missing position is inferred and filled in combination with the logical dependency constraints to obtain the draft document.

[0068] Content fingerprint calculation is performed on the data content of each node in the draft document to generate a node fingerprint set. When a revision operation is received for the draft document, the impact propagation path of the revision operation is determined based on the logical dependency constraint.

[0069] The consistency of the associated node position sequence is verified along the influence propagation path. When a logical dependency is detected between the associated node data and the revised content, the associated node data is updated according to the logical dependency constraint and the node fingerprint is recalculated.

[0070] The node fingerprint sets before and after the revision, the impact propagation path, and the audit task input information are linked and stored to form a version change management link.

[0071] In one optional implementation, determining the logical dependency constraints between nodes in the audit type identifier includes:

[0072] Extract the node structure configuration information corresponding to the audit type identifier from the working paper template library. The node structure configuration information defines the identifier, hierarchical relationship and data attribute type of each node in the working paper document.

[0073] Based on the hierarchical relationship, identify node pairs with data reference relationships, determine whether the data content of the source node can be used as the data generation input of the target node according to the compatibility rules between the data attribute types, and establish a dependency relationship marker between the source node and the target node when the determination result is yes.

[0074] For node pairs with established dependency relationships, perform dependency direction analysis to determine the data flow direction between source nodes and target nodes. When the data generation of a target node depends on multiple source nodes, mark the multiple source nodes as a set of preceding dependent nodes. When the data content of a source node is called by the data generation process of multiple target nodes, mark the multiple target nodes as a set of subsequent dependent nodes.

[0075] Traverse all nodes in the draft document, extract the set of preceding dependent nodes and the set of subsequent dependent nodes for each node, and associate and store the set of preceding dependent nodes, the set of subsequent dependent nodes and the node identifier to form the logical dependency constraint between nodes.

[0076] The working paper template library pre-stores standardized node structure configuration information corresponding to different audit types. For special audit business scenarios, working paper documents typically contain various types of nodes, such as basic information nodes, data collection nodes, analysis and calculation nodes, and conclusion judgment nodes. Each node contains a unique node identifier in the node structure configuration information. This identifier uses a hierarchical encoding method. For example, "A.1.2" represents the second child node of the first child node of the second level under the first-level node A. The encoding reflects the node's hierarchical position in the entire document structure. The node structure configuration information also defines the data attribute type of each node. Common data attribute types include text, numeric, date, amount, ratio, and list types. Different data attribute types determine what form of data content the node can accept.

[0077] After extracting the node structure configuration information corresponding to the audit type identifier, node pairs with data referencing relationships are identified based on hierarchical relationships. Identifying hierarchical relationships considers not only the parent-child relationships of nodes in the document tree structure but also the business semantics of the nodes. For example, in a special financial audit, there are revenue nodes, cost of goods sold nodes, and gross profit margin nodes. Although these three nodes are at the same level in the hierarchical structure, from a business semantic perspective, the calculation of gross profit margin depends on the data content of revenue and cost of goods sold. Therefore, revenue and cost of goods sold nodes are potential data sources for the gross profit margin node. By constructing a business semantic association rule base, the business meaning of nodes is combined with their position in the hierarchical structure to identify all node pairs with data referencing relationships.

[0078] After identifying node pairs, further filtering is performed based on compatibility rules between data attribute types. These compatibility rules define the conversion and matching relationships between different data attribute types. For example, monetary data can be directly used as input for numerical calculation nodes, and date data can be converted into time span values ​​for period analysis, but text data typically cannot be directly used in numerical calculations. When determining whether source node data can be used as input for generating data for the target node, the data attribute type of the source node is obtained. Data attribute types of the target node Query the compatibility rule base to see if there is a rule from... arrive The transformation path. If a valid transformation rule exists in the compatibility rule base, it is determined that the source node data can be used as the data generation input for the target node. A dependency relationship marker is established between the source node and the target node. This marker records the source node identifier, the target node identifier, and the data transformation rule identifier.

[0079] After establishing dependency relationships, dependency direction analysis is performed on each pair of nodes with dependencies. Dependency direction reflects the flow of data between nodes, determining which node is the data provider and which is the data user. In audit working papers, data flow typically follows a pattern of transmission from basic data nodes to analytical conclusion nodes. For example, bank statement detail nodes provide raw data, statement summary nodes aggregate the detail data, and abnormal transaction identification nodes make judgments and analyses based on the summary data. During dependency direction analysis, the order of data generation is determined based on the business attributes of the nodes and their position in the audit process. When the data generation of a target node depends on multiple source nodes simultaneously, the identifiers of these source nodes are collected and marked as the set of preceding dependent nodes for that target node. For example, if the calculation of the debt-to-asset ratio node depends on both the total assets node and the total liabilities node, then the total assets node and the total liabilities node together constitute the set of preceding dependent nodes for the debt-to-asset ratio node.

[0080] Correspondingly, when the data content of a source node is referenced by multiple target nodes, these target nodes need to be marked as the set of subsequent dependent nodes of the source node. For example, the data of the revenue node is not only referenced by the gross profit margin calculation node, but also by multiple nodes such as the revenue growth rate calculation node and the revenue structure analysis node. These nodes that reference the revenue data together constitute the set of subsequent dependent nodes of the revenue node. Identifying the set of subsequent dependent nodes is of great significance for subsequent impact propagation path analysis. When the data of the source node changes, it is necessary to quickly locate all affected target nodes through the set of subsequent dependent nodes.

[0081] After completing the dependency analysis of all nodes, the entire document is traversed to extract the set of preceding and subsequent dependent nodes for each node. The extraction process employs a graph traversal algorithm, treating nodes as vertices and dependencies as directed edges. Starting from each node, a depth-first search or breadth-first search is used to trace all its preceding dependent nodes along the direction of the dependency relationship, or to trace all its subsequent dependent nodes along the direction of the reference relationship. To improve retrieval efficiency, an adjacency list data structure is used to store the dependencies between nodes. Each node corresponds to a linked list, which stores the list of identifiers of its preceding and subsequent dependent nodes.

[0082] When storing the extracted sets of preceding dependent nodes, subsequent dependent nodes, and node identifiers, a logical dependency constraint data structure is constructed. This data structure uses key-value pairs, with node identifiers as keys and structured objects containing information such as the sets of preceding and subsequent dependent nodes, node data attribute types, and node hierarchical positions as values. During storage, the strength attribute of the dependency relationship is also recorded to distinguish between strong and weak dependencies. A strong dependency indicates that the data generation of the target node is completely dependent on the data of the source node; the absence of source node data will prevent the target node from generating valid content. A weak dependency indicates that the source node data serves only as reference information for the generation of the target node data; the absence of source node data does not affect the basic generation logic of the target node. By organizing and storing the dependency information of all nodes in a structured manner, a complete logical dependency constraint between nodes is formed. This constraint not only describes the static association between nodes but also implicitly contains the dynamic flow path of data in the audit working papers, providing a foundation for subsequent missing data reasoning and filling, impact propagation path analysis of revision operations, and consistency verification.

[0083] In practical applications, establishing logical dependency constraints also requires consideration of detecting and handling circular dependencies. A circular dependency occurs when node A depends on node B, and node B directly or indirectly depends on node A. A topological sorting algorithm is used to detect cycles in the dependency graph. When a circular dependency is detected, the dependency relationship is adjusted according to business rules, or the circularly dependent node group is marked as a special node group requiring manual intervention, ensuring the rationality and executability of the logical dependency constraints.

[0084] In one optional implementation, the dataset to be audited is subjected to cross-node association verification based on semantic consistency verification rules to generate missing data identifiers. Based on the node position corresponding to the missing data identifier, historical audit records matching the business attribute category of the current audit object identifier are retrieved, including:

[0085] Based on the set of preceding dependent nodes of each node in the logical dependency constraint, determine the list of required data fields for each node;

[0086] Iterate through the data content of each node in the dataset to be audited. When the actual data field identifier of each node is missing the data field identifier in the list of required data fields, record the node identifier and the missing data field identifier, and generate the missing data identifier.

[0087] Extract the node identifier and data field identifier from the missing data identifier, query the set of preceding dependent nodes corresponding to the node identifier from the logical dependency constraint, determine whether the data content of each preceding dependent node in the set of preceding dependent nodes is complete, and when the determination result is yes, mark the missing data identifier as an independent missing type.

[0088] For missing data identifiers marked as independent missing, extract the business attribute category features corresponding to the current audit object identifier;

[0089] Retrieve historical audit records from the historical audit working paper knowledge base that match the business attribute category characteristics of the current audit object identifier.

[0090] like Figure 2 As shown, the method includes:

[0091] Based on the established logical dependency constraints, a mapping table of required fields for each node is constructed. Specifically, for the audit flowchart structure corresponding to the audit type identifier, each node in this structure can have one or more preceding dependent nodes. The output data of the preceding dependent nodes serves as the input basis for the current node, and this dependency relationship determines the data fields that the current node must possess. Taking financial audit as an example, if there is an "Asset Appraisal" node in the audit process, and its preceding dependent nodes are "Original Value Verification" and "Depreciation Calculation," then the required data fields for the "Asset Appraisal" node should include fields such as original asset value, accumulated depreciation, and appraisal base date. By parsing the dependencies between nodes recorded in the logical dependency constraints, a list of required data fields can be generated for each node. The field identifiers in this list adopt a structured encoding method; for example, the "Original Asset Value" field can be encoded as "ASSET_ORIGI7AL_VALUE," ensuring accurate matching during subsequent verification.

[0092] After constructing the list of required data fields, the dataset to be audited is traversed node by node. The dataset is typically organized in a structured data format, with each node corresponding to a data object containing several key-value pairs of data fields. During traversal, the set of data field identifiers actually contained in the current node is extracted, and this set of identifiers is then used in a set operation with the list of required data fields for that node. If a field identifier in the list of required data fields is not present in the set of identifiers in the list of identifiers, the node is considered to have missing data. At this point, the unique identifier of the node and the specific missing field identifier are recorded, forming a missing data identifier record. For example, if the required fields for the "Asset Appraisal" node include "ASSET_ORIGI7AL_VALUE", "ACCUMULATED_DEPRECIATIO7", and "EVALUATIO7_DATE", but the actual data only contains "ASSET_ORIGI7AL_VALUE" and "EVALUATIO7_DATE", then the missing data identifier is generated as "Node ID: ASSET_EVAL, Missing Field: ACCUMULATED_DEPRECIATIO7".

[0093] After generating missing data identifiers, it is necessary to further analyze the type characteristics of the missing data, extract the node identifiers recorded in the missing data identifiers, and query the set of predecessor dependent nodes for that node in the logical dependency constraint structure. For each node in this set of predecessor dependent nodes, check the integrity of its data content. The integrity judgment is based on two dimensions: first, whether the predecessor dependent node itself contains all required fields; and second, whether the data value of the predecessor dependent node is valid. Validity verification includes checking whether numeric fields are null values, whether character fields are empty strings, and whether date fields conform to format specifications, etc. When the data of all nodes in the set of predecessor dependent nodes is complete and valid, it indicates that the data missing of the current node is not caused by the upstream data missing, but is an independent data missing of the node itself. At this time, the missing data identifier is marked as "independent missing".

[0094] For missing data markers indicated as independent missing, an inference-based imputation mechanism based on historical audit records is initiated. Business attribute category features corresponding to the current audit object identifier are extracted. Audit object identifiers typically contain multi-dimensional information, such as enterprise code, industry classification code, and size identifier. Extracting business attribute category features requires structured parsing of this multi-dimensional information. Taking enterprise audit as an example, if the audit object identifier is "E7T20230512_MFG_L", where "E7T" represents the enterprise type, "20230512" is the registration time code, "MFG" represents the manufacturing industry classification, and "L" represents the size of a large enterprise, then the extracted business attribute category features include key attribute tags such as "Industry: Manufacturing", "Size: Large", and "Years Established: 5+ years".

[0095] When retrieving matching records in the historical audit working paper knowledge base, a multi-level matching strategy is employed. The knowledge base stores historically completed audit working papers and their metadata, with each historical record associated with the business attribute category characteristics of the audited object. The matching process performs exact matching, searching for historical records with completely identical business attribute category characteristics. If the number of exact matching results is insufficient, fuzzy matching is performed with lowered matching criteria, such as requiring only consistency in the two core attributes of industry classification and company size. To improve matching quality, a similarity score is calculated for each candidate matching record. The similarity score comprehensively considers factors such as the matching degree of business attributes, the time distance of the historical record, and the quality of historical audit conclusions. Specifically, weight coefficients are assigned to each attribute dimension, with industry classification typically having a higher weight coefficient due to significant differences in audit rules across different industries. The calculation of the impact of time distance is based on the interval between the audit task initiation time and the historical record generation time; a shorter interval indicates higher timeliness. The quality of historical audit conclusions is measured by indicators such as whether the historical audit successfully passed review and whether there were any significant revisions.

[0096] After retrieving historical audit records, data content matching the node position corresponding to the missing data identifier is extracted from the matched historical record set. The correspondence between node positions is established through the standard node coding system defined in the audit type identifier. Although different audit tasks have different audit objects, if they belong to the same audit type, their audit process structure and node definitions are unified. For example, all financial audit tasks include standard nodes such as "cash and cash equivalents audit," "accounts receivable audit," and "fixed asset audit," and these nodes have the same node identifier code in different audit tasks. Based on the node identifier and missing field identifier recorded in the missing data identifier, data content with the same node identifier and field identifier is located in the historical audit records. If valid data exists at this position in the historical records, the data value is extracted as a candidate fill value.

[0097] To ensure the reasonableness of the populated data, candidate populated values ​​are validated using logical dependency constraints. Logical dependency constraints not only define the dependencies between nodes but also include rules governing data value ranges, data format requirements, and quantitative relationships between data. For example, if the missing field is "accumulated depreciation," the logical dependency constraint stipulates that this value cannot exceed the corresponding "original asset value" and must meet the calculation rules for depreciation years and methods. During validation, candidate populated values ​​are correlated and verified with other existing field data in the current node. If a candidate populated value directly satisfies all constraints, it is written as the inferred populated value to the missing location. If a candidate populated value does not fully meet the constraints but has reference value, it is adaptively adjusted. Adjustment methods include scaling proportionally, discounting based on time factors, and correction based on size differences. Taking fixed asset depreciation as an example, if the historical enterprise size is medium and the current audit object is large, the accumulated depreciation in the historical record needs to be adjusted proportionally to the asset size before being populated.

[0098] When multiple candidate values ​​for the same node position are extracted from multiple historical records, a consensus voting mechanism is used to determine the final value. The frequency of each candidate value is counted, and the value with the highest frequency is selected as the priority value. If the frequencies of the candidate values ​​are similar, the numerical distribution characteristics of the candidate values ​​are calculated. For numerical fields, the median or weighted average of the candidate values ​​can be calculated as the value to be filled, and the weight is determined based on the similarity score of the historical records. For enumerated or categorical fields, the most conservative or most common value in business logic is selected. After the entire inference and filling process is completed, the data source is marked at the corresponding node position in the working document, indicating that the data is generated based on historical inference, and the referenced historical record identifier is recorded to facilitate subsequent review by auditors.

[0099] Through the aforementioned cross-node correlation verification and historical data-based intelligent filling mechanism, structurally complete working papers can be quickly generated at the beginning of the audit task, reducing audit process blockages caused by missing data, while ensuring that the filled data matches the business characteristics of the audit object, providing a reliable data foundation for the efficient conduct of subsequent audit work.

[0100] In one optional implementation, the data content of the corresponding node position in the historical audit record is extracted and, in conjunction with the logical dependency constraints, the missing positions are inferred and filled to obtain the working document, which includes:

[0101] Based on the node identifier and data field identifier in the missing data identifier, extract historical data content from the historical audit records where both the node identifier and data field identifier match, and use the historical data content as the initial filling data.

[0102] Extract the set of preceding dependent nodes corresponding to the node with the missing data identifier from the logical dependency constraint, and obtain the current data content of each preceding dependent node in the set of preceding dependent nodes in the dataset to be audited;

[0103] The initial data is used to perform semantic compatibility verification with the current data content. The semantic compatibility verification determines whether the data format and numerical range of the initial data are logically conflicting with the current data content.

[0104] When the semantic compatibility verification result is that there is no logical conflict, the initial padding data is used as the final padding data and filled into the node position corresponding to the missing data identifier.

[0105] When the semantic compatibility verification result indicates a logical conflict, the data calculation relationship rules between the node corresponding to the missing data identifier and the set of preceding dependent nodes are extracted from the logical dependency constraints. Logical reasoning calculations are performed based on the data calculation relationship rules and the current data content to generate reasoning fill data. The reasoning fill data is then filled into the node position corresponding to the missing data identifier.

[0106] When filling in missing data in special audit working papers, it is necessary to ensure that the filled content conforms to both historical audit experience and the logical constraints of the current audit task. For the node identifiers and data field identifiers marked in the missing data identifiers, an exact match retrieval is performed from the historical audit record database. Node identifiers are used to locate specific business processes in the audit working papers, while data field identifiers correspond to specific data items within that business process. For example, in a fixed asset audit scenario, the node identifier might be "asset depreciation calculation," and the data field identifier might be "annual depreciation rate." The historical values ​​filled in for this node and field for the same audit object or similar business attribute objects are retrieved from the historical audit records. The retrieved historical data includes the numerical value itself and its associated data type, unit, and time of value acquisition, etc. This historical data is temporarily stored as initial fill data.

[0107] After obtaining the initial data, it is necessary to verify whether the data is applicable to the specific circumstances of the current audit task. Logical dependency constraints define the sequential dependencies and calculation logic between nodes in the audit working papers, extracting the set of preceding dependent nodes corresponding to the node with missing data identifiers. Preceding dependent nodes refer to business steps in the audit process that require data entry before the current node. Taking inventory impairment provision audit as an example, if the missing data identifier corresponds to the "Impairment Provision Amount" node, its preceding dependent node set includes the "Inventory Book Value" node and the "Inventory Net Realizable Value" node. Traversing the set of preceding dependent nodes, the current data content of each preceding dependent node is read from the dataset to be audited. This current data content reflects the actual business status of this audit task.

[0108] Semantic compatibility verification is a crucial step in ensuring that historical data can be reasonably applied to the current scenario. The verification process checks whether the data format of the initially populated data is consistent with the current data content. For example, date formats need to be uniformly "YYYY-MM-DD", and monetary amounts need to be uniformly retained to two decimal places. Numerical range checks verify whether the initially populated data conforms to the reasonable range implied by the preceding dependent node data. In the audit of accounts receivable bad debt provision, if the preceding dependent node "Aging Structure" shows that the proportion of accounts receivable aged over three years is... The "provision for bad debts ratio" in the initial data was only... If so, there is a clear mismatch in numerical ranges. Logical conflict detection also includes constraint verification, such as the preceding dependent node "ending balance" being... The amount was [amount] yuan, while the "impairment loss amount" in the initial populated data was [amount]. This amount clearly violates the logical constraint that impairment losses cannot exceed the book value of assets.

[0109] When semantic compatibility verification passes, it indicates that the historical data is compatible with the current audit environment in terms of format, numerical range, and logical constraints, and can be used directly. The initial data is used as the final data and written to the node position and data field specified by the missing data identifier. The data filling operation also records the data source marker, indicating that the data comes from the automatic filling of historical audit records, which facilitates auditor identification and review.

[0110] When semantic compatibility verification reveals logical conflicts, directly using historical data can lead to erroneous audit conclusions. Logical reasoning based on the actual data of the current audit task is necessary. The rules for calculating data relationships between missing data identifiers and the set of preceding dependent nodes are extracted from logical dependency constraints. These rules describe the business logic in a structured form. For example, in fixed asset depreciation audits, the calculation relationship rule can be defined as: annual depreciation equals the original value of the fixed asset minus the estimated net residual value, divided by the estimated useful life. When using the sum-of-the-years'-digits method, the calculation formula is adjusted to the original value of the fixed asset minus the estimated net residual value multiplied by the ratio of the remaining useful life to the sum of the digits of the estimated useful life.

[0111] Based on the extracted data and calculation rules, logical reasoning is performed by substituting the current data content of the preceding dependent nodes. In the equity method accounting scenario for long-term equity investments, if the data identifier corresponding to the "investment income recognition amount" node is missing, the preceding dependent nodes include "investee's net profit" and "shareholding ratio." The calculation rule is defined as investment income equal to the investee's net profit multiplied by the shareholding ratio. Assume the current data shows the investee's net profit is... Yuan, holding a percentage of shares Then the inference fill data is calculated as follows: Yuan. For complex scenarios involving multi-level calculations, such as the recognition of deferred tax assets, the calculation rules include conditional judgments. It is necessary to first determine whether deductible temporary differences exist, and then determine the amount of deferred tax assets that can be recognized based on the applicable tax rate and the expected taxable income in future periods.

[0112] After the inference-based data is generated, it is filled into the node positions corresponding to the missing data identifiers, and the data is marked as automatically generated based on logical reasoning. Simultaneously, the reasoning basis is recorded, including the identifiers of the computational relationship rules used, the data of the preceding dependent nodes involved in the calculation and their values, and the intermediate results of the calculation process, forming a complete reasoning chain. These records allow auditors to trace and verify the reasonableness of the automatic data filling, ensuring the reliability and auditability of the audit working papers. For inference-based data filling, the system sets an audit flag, prompting auditors to manually review and confirm the data, avoiding misunderstandings or omissions of special cases that may exist in automatic reasoning.

[0113] Through the above mechanism, an intelligent connection is established between historical audit experience and current audit practice, which not only improves the efficiency of audit working paper preparation but also ensures the accuracy and logical consistency of the data. Historical data can be directly reused when applicable, and dynamic reasoning based on business logic can be performed when discrepancies exist, thus realizing the effective inheritance and flexible application of audit knowledge.

[0114] In one optional implementation, content fingerprint calculation is performed on the data content of each node in the draft document to generate a node fingerprint set. When a revision operation is received for the draft document, the impact propagation path of the revision operation is determined based on the logical dependency constraints, including:

[0115] Traverse all nodes in the draft document, extract the data content of each node and perform hash operation to generate the node fingerprint value corresponding to each node, associate and store the node identifier of each node with the corresponding node fingerprint value to form a node fingerprint set.

[0116] When a revision operation is received for the draft document, the target node identifier and the revised data content in the revision operation are parsed, and a hash operation is performed on the revised data content to generate a revised fingerprint value;

[0117] Extract the original node fingerprint value corresponding to the target node identifier from the node fingerprint set, compare the revised fingerprint value with the original node fingerprint value, and when the comparison result is inconsistent, confirm that the revision operation has caused the data content of the target node to change;

[0118] Extract the set of subsequent dependent nodes corresponding to the target node identifier from the logical dependency constraint. For each subsequent dependent node in the set of subsequent dependent nodes, recursively extract the set of subsequent dependent nodes corresponding to each subsequent dependent node. Construct a multi-level dependency propagation link starting from the target node identifier. Arrange all node identifiers contained in the multi-level dependency propagation link according to the dependency hierarchy to generate the influence propagation path.

[0119] After the working paper document is generated, precise tracking and management of the data content of each node within it is required. This involves traversing all nodes in the working paper document. Each node carries a unique node identifier, which is a combination of the audit type identifier and the node sequence number. For example, in an asset-specific audit scenario, the node identifier is "ASSET_AUDIT_N003," indicating the 3rd node under the asset audit type. The data content stored in each node is extracted. This data content can be text descriptions, numerical data, or structured tabular data. A unified hash operation method is used to process different types of data content. Specifically, the data content within a node is converted into a byte stream sequence, and the SHA-256 hash algorithm is used to calculate a 64-bit hexadecimal hash value, which is the node fingerprint value. For example, if a node stores data containing "Original value of fixed assets: 12,500,000 yuan", a hash operation will generate a node fingerprint value "3f8a9c2e7d1b4f6a5e3c9d8a7b6f4e2c1a9f8e7d6c5b4a3f2e1d9c8b7a6f5e4d". Each node's identifier is associated with its corresponding node fingerprint value, organized using a key-value pair structure. The key is the node identifier, and the value is the node fingerprint value. The set of all key-value pairs constitutes the node fingerprint set, which is stored in an in-memory database, supporting high-speed query access.

[0120] When auditors need to revise data at a specific node in working papers, they submit a revision request to the system. The revision request includes the target node identifier and the revised data content. For example, the target node identifier in the revision request might be "ASSET_AUDIT_N003," and the revised data content might be "Original value of fixed assets: 12,800,000 yuan." The revision request is parsed to extract the target node identifier and the revised data content. A hash operation, identical to the node fingerprint generation process, is performed on the revised data content "Original value of fixed assets: 12,800,000 yuan," converting it into a byte stream sequence. The SHA-256 algorithm is then used to calculate the revised fingerprint value, which is also a 64-bit hexadecimal string.

[0121] The fingerprint of the target node, "ASSET_AUDIT_N003", is retrieved from the node fingerprint set using a key-value search. This retrieves the original node fingerprint value, calculated from the data stored on the node before the revision operation. The revised fingerprint value is then compared character-by-character with the original fingerprint value. Due to the nature of hash algorithms, even minor changes in data content will result in completely different hash values. In this example, the original data "Original value of fixed assets 12,500,000 yuan" differs from the revised data "Original value of fixed assets 12,800,000 yuan," leading to a discrepancy between the revised and original fingerprint values. When the comparison results show a discrepancy, it confirms that the revision operation caused a substantial change in the target node's data content, requiring further analysis of the impact of this change on other nodes.

[0122] From the logical dependency constraint data structure established during the audit task initialization phase, the set of subsequent dependent nodes corresponding to the target node identifier "ASSET_AUDIT_N003" is retrieved. Logical dependency constraints are stored in a directed graph structure, with each node identifier as a vertex and the dependencies between nodes as directed edges. For example, in an asset-specific audit scenario, node "ASSET_AUDIT_N003" stores the original value data of fixed assets, and subsequent dependent nodes include "ASSET_AUDIT_N007" for storing accumulated depreciation calculation results and "ASSET_AUDIT_N012" for storing the summary data of asset net value. By traversing all directed edges originating from "ASSET_AUDIT_N003" in the directed graph, the endpoint node identifiers of the edges are extracted to form the first-level set of subsequent dependent nodes, which includes "ASSET_AUDIT_N007" and "ASSET_AUDIT_N012".

[0123] For each node in the first-level set of subsequent dependent nodes, recursively perform the same dependency extraction operation. For "ASSET_AUDIT_N007", extract its corresponding set of subsequent dependent nodes from the logical dependency constraint, including "ASSET_AUDIT_N015" for storing depreciation expense allocation data. For "ASSET_AUDIT_N012", extract its corresponding set of subsequent dependent nodes from the logical dependency constraint, including "ASSET_AUDIT_N018" for storing balance sheet related item data. Continue performing the recursive extraction operation on "ASSET_AUDIT_N015" and "ASSET_AUDIT_N018" until a leaf node with no subsequent dependent nodes is reached. The entire recursive process constructs a multi-level dependency propagation link starting from the target node identifier "ASSET_AUDIT_N003". This link is presented in a tree structure, with the root node being "ASSET_AUDIT_N003", the first-level child nodes being "ASSET_AUDIT_N007" and "ASSET_AUDIT_N012", and the second-level child nodes being "ASSET_AUDIT_N015" and "ASSET_AUDIT_N018".

[0124] To facilitate subsequent impact propagation verification, all node identifiers in the multi-level dependency propagation chain need to be arranged according to the dependency hierarchy. A breadth-first search algorithm is used to traverse the tree structure. The root node "ASSET_AUDIT_N003" is added to the sequence. The first-level child nodes "ASSET_AUDIT_N007" and "ASSET_AUDIT_N012" are added to the sequence according to the priority order defined in the logical dependency constraints. Then, the second-level child nodes "ASSET_AUDIT_N015" and "ASSET_AUDIT_N018" are added to the sequence. The final generated sequence is ["ASSET_AUDIT_N003",

[0125] "ASSET_AUDIT_N007",

[0126] "ASSET_AUDIT_N012",

[0127] "ASSET_AUDIT_N015",

[0128] The sequence “ASSET_AUDIT_N018” represents the impact propagation path. The order of the impact propagation path ensures that after the target node is revised, each associated node can be verified and updated sequentially according to the logical order of data dependencies, avoiding logical errors where subsequent nodes are updated before preceding nodes.

[0129] In practical applications, the storage structure of the node fingerprint set is implemented using a distributed hash table, supporting concurrent access and fast querying. Even when the number of nodes in the audit worksheet reaches thousands, the query response time for the node fingerprint set can still be controlled within milliseconds. The choice of hash algorithm directly affects the performance and collision probability of fingerprint calculation. The SHA-256 algorithm achieves a good balance between security and computational efficiency, with an extremely low collision probability, making it suitable for large-scale audit worksheet scenarios. The construction process of the influence propagation path adopts a memoized recursive strategy. For nodes that have already been accessed, the results of their subsequent dependent node sets are cached to avoid repeated calculations, significantly improving the efficiency of constructing multi-level dependency propagation links. When the logical dependency constraint structure corresponding to the audit type identifier is complex, and multiple dependency paths converge to the same node, the influence propagation path will deduplicate duplicate node identifiers to ensure that each node appears only once in the path, while retaining its earliest occurrence level position to guarantee the correctness of influence propagation verification.

[0130] In one optional implementation, consistency verification is performed on the sequence of associated node positions along the influence propagation path. When a logical dependency is detected between the associated node data and the revised content, updating the associated node data and recalculating the node fingerprint according to the logical dependency constraint includes:

[0131] According to the hierarchy of node identifiers in the propagation path, extract the identifiers of each associated node and the corresponding associated node data content in sequence.

[0132] For the associated node identifier at the current level, obtain the actual data of the predecessor nodes of each predecessor node in the corresponding predecessor dependency node set in the draft document;

[0133] Extract the data constraint rules between the associated node identifier and the set of preceding dependent nodes from the logical dependency constraints, and calculate the expected associated node data value based on the data constraint rules and the actual data of the preceding nodes;

[0134] The associated node data content is compared with the expected associated node data value. When the comparison result is inconsistent, it is determined that the associated node data content has a logical dependency relationship with the revised content and needs to be updated. The updated associated node data is recalculated and generated according to the data constraint rules and the actual data of the preceding node, and the original associated node data content corresponding to the node position of the associated node identifier is replaced.

[0135] Perform a hash operation on the updated associated node data to generate an updated node fingerprint value, and update the node fingerprint value corresponding to the associated node identifier in the node fingerprint set to the updated node fingerprint value.

[0136] Continue processing the associated node identifiers at the next level in the influence propagation path until the consistency verification and data update of all associated node identifiers in the influence propagation path are completed.

[0137] After the initial generation of the working papers, for data revision scenarios that occur during the audit, it is necessary to ensure that the revision operations do not disrupt the established logical dependencies between nodes within the working papers. When auditors revise the data at a certain node in the working papers, a systematic consistency verification of the sequence of related node positions affected by the revision operation is performed according to the established impact propagation path.

[0138] Specifically, the influence propagation path contains a series of node identifiers, organized according to dependency hierarchy. The dependency hierarchy is based on the data flow relationship defined in the logical dependency constraints. For example, when the data value of node A is jointly determined by the data of nodes B and C, nodes B and C are at a shallower hierarchy, while node A is at a deeper hierarchy. Following the order from shallowest to deepest hierarchy, the identifiers of each associated node in the influence propagation path and their corresponding associated node data content in the draft document are extracted sequentially. Assume the influence propagation path contains node identifiers. The corresponding data contents are as follows: The processing order follows the direction of increasing dependency level depth.

[0139] For the associated node identifier currently in the verification process Based on the dependency mapping table recorded in the logical dependency constraints, obtain the set of predecessor dependency nodes for this node. The predecessor dependency nodes included in this set are identified as follows: , representing a node The data value depends on this The data of each preceding node. The actual data values ​​of these preceding dependent nodes stored in the document are read from the current version of the draft document and denoted as follows: The actual data of these preceding nodes has already been updated in the previous processing steps and can also remain in its original state; however, it is necessary to obtain the real-time data from the draft document.

[0140] Logical dependency constraints targeting nodes There are explicit data constraint rules stored between the node and its preceding dependent node set. These constraint rules describe the node. How should the data values ​​be calculated or derived based on the data from the preceding nodes? For example, in an asset and liability audit scenario, if the node... Represents the flow ratio, with preceding dependent nodes. Represents the total amount of current assets, with preceding dependent nodes. If the total current liabilities are represented, then the data constraint rule is defined as follows: Based on the extracted data constraint rules and actual data of the preceding node According to the calculation logic defined by the rules, the expected associated node data values ​​are generated. .

[0141] Nodes in the draft document Currently stored related node data content The calculated expected associated node data values A comparison is performed. The comparison process considers data type characteristics: numerical equality is used for numeric data, exact string matching is used for text data, and a reasonable error threshold is set for floating-point data with precision requirements. When the comparison result is displayed... and When differences exist, it indicates the node. The data content and the revised content have a logical dependency relationship, and the current data content no longer meets the requirements of the logical dependency constraint, so it needs to be updated.

[0142] During the data update process, according to data constraint rules and actual data of the preceding node The calculation process is re-executed to generate updated associated node data. This updated associated node data fully complies with the definition of logical dependency constraints, ensuring consistency with the data of preceding dependent nodes. The updated associated node data is then written into the node list in the working document. Replace the original associated node data content at the corresponding node location. In some audit scenarios, data constraint rules involve multiple calculation steps or conditional branch judgments. For example, the calculation of total profit requires first summarizing the details of operating revenue, and then deducting operating costs and various expenses. In this case, the calculation is performed according to the complete logical chain defined by the constraint rules.

[0143] After updating the data, to ensure the integrity of the version change management chain, the nodes need to be recalculated. The updated node fingerprint value is generated by performing the same hash algorithm on the updated associated node data as the initial fingerprint calculation. The hash algorithm can be a standard algorithm such as SHA-256 or MD5, ensuring that the same data content produces the same fingerprint value, and different data content produces different fingerprint values. The node identifier is located in the node fingerprint set. For the corresponding fingerprint record, update the node fingerprint value stored in that record with the newly calculated updated node fingerprint value. At the same time, the node fingerprint value before the revision is retained as historical version information.

[0144] Complete the identification of related nodes at the current level. After consistency verification and data updates, continue processing the identifiers of associated nodes at the next level in the propagation path. Since the dependency hierarchy is organized according to the data flow, the set of preceding dependent nodes of the next level node may include nodes that have already been updated at the current level. Therefore, it is necessary to obtain the latest data status of the draft document for subsequent verification. Repeat the complete process of obtaining preceding dependent node data, calculating expected data values, comparing and verifying, updating data, and recalculating fingerprints until all nodes in the propagation path are affected. All associated node identifiers have been processed.

[0145] During the processing, for certain special nodes, the data constraint rules can be defined as either maintaining the original value or requiring manual review. In this case, if inconsistencies are detected during the consistency verification phase, the data is not automatically updated. Instead, a consistency anomaly marker is added to the node location in the working paper, prompting auditors to intervene manually. For related nodes across forms or chapters, when obtaining the actual data of the preceding nodes, it is necessary to accurately locate the corresponding node's storage location in the document according to the structured storage format of the working paper and read the correct data content.

[0146] After the entire consistency verification and data update process is completed, all related node data in the working paper affected by the revision operation has been synchronized and updated according to logical dependency constraints. The corresponding fingerprint values ​​in the node fingerprint set have also been updated to the latest state, ensuring that the internal logical consistency of the working paper is maintained and providing a reliable data foundation for the subsequent version change management chain. Through this hierarchical, rule-driven verification and update mechanism, the problem of missing or incorrect updates of related data that occurs during manual revision is effectively avoided, improving the data quality and reliability of the special audit working papers.

[0147] In one optional implementation, the node fingerprint sets before and after the revision, the impact propagation path, and the audit task input information are linked and stored to form a version change management chain, including:

[0148] The difference between the node fingerprint set before revision and the node fingerprint set after revision is compared, the node identifiers whose node fingerprint values ​​have changed are extracted, and the node identifiers whose node fingerprint values ​​have changed, along with the corresponding node fingerprint values ​​before and after revision, are combined to form a set of changed node records.

[0149] Extract the operation timestamp from the revision operation, and encapsulate the operation timestamp, the target node identifier, the revised data content, the set of change node records, and the impact propagation path in a structured manner to generate a version change record;

[0150] Extract the audit object identifier and audit task identifier from the audit task input information, and establish an association index relationship between the audit object identifier, the audit task identifier and the version change record;

[0151] The version change records and the associated index relationships are stored in the version change management database. The associated index relationships support the retrieval of all corresponding version change records based on the audit object identifier and the audit task identifier, thus forming the version change management chain.

[0152] During the revision of audit working papers, accurately recording and tracking each change is crucial to ensuring the traceability of the audit work. The process involves obtaining the pre-revision and post-revision node fingerprint sets and employing a node-by-node comparison strategy for difference detection. For each node identifier in the set, its corresponding pre-revision and post-revision node fingerprint values ​​are compared bit-by-bit. If any character difference exists between the two fingerprint values, the node is considered to have undergone a substantial change. All node identifiers that meet the change criteria are extracted, and their corresponding pre-revision and post-revision node fingerprint values ​​are recorded. These three pieces of information are combined to form a change node record. All change node records are then aggregated to form a change node record set, which comprehensively reflects all node data changes resulting from this revision operation.

[0153] Extract the operation timestamp, accurate to the second, from the metadata of the revision operation. This timestamp identifies the precise moment the revision action occurred. Locate the target node identifier that triggered the revision, indicating the location of the node directly modified by the user. Obtain the latest data content in the target node after the revision operation is completed. This data content reflects the revision result input by the user or generated by the system. Organize the three core elements—operation timestamp, target node identifier, and revised data content—along with the change node record set generated in the previous steps and the impact propagation path. Employ a structured data encapsulation method, using the operation timestamp as the time dimension identifier, the target node identifier as the trigger source location, the revised data content as the change value carrier, the change node record set as the impact scope description, and the impact propagation path as the propagation chain record, forming a version change record containing complete change information. This version change record is encapsulated using a unified data structure to ensure that each information field has clear semantic definitions and access interfaces.

[0154] The audit task input information is parsed to extract two key fields: Audit Object Identifier and Audit Task Identifier. The Audit Object Identifier identifies the specific business entity targeted in this audit, while the Audit Task Identifier distinguishes different audit projects or batches. A ternary association structure is created, using the Audit Object Identifier and Audit Task Identifier as a composite primary key, with version change records as the association target. A mapping relationship is established from the composite primary key to version change records, ensuring that any given combination of Audit Object Identifier and Audit Task Identifier can accurately locate all corresponding version change records. A data structure with an association index is set up, supporting retrieval based on a single Audit Object Identifier as well as precise queries based on combinations of Audit Object Identifier and Audit Task Identifier. A bidirectional association pointer is established in the index structure, enabling quick location of version change records from audit task information and reverse tracing from version change records back to the corresponding audit task context.

[0155] Version change records are serialized, converting structured records into a data format suitable for database storage. Corresponding storage spaces are created or located in the version change management database, and the serialized version change records are written to the version change record table. Simultaneously, the related index relationships are written to a separate index table, using the audit object identifier and audit task identifier as index keys, and the storage location or record identifier of the version change record as the index value. Foreign key relationships are established between the index table and the version change record table to ensure the integrity constraints of the index relationships. A database query interface is configured that receives the audit object identifier and audit task identifier as query parameters, locates all corresponding version change records through the related index relationships, and returns the complete change history sequence in chronological order of operation timestamps. This organizational method forms a version change management chain, supporting time-series tracing of all revision activities for the same audit object under the same audit task.

[0156] In version change management scenarios, when it's necessary to query the revision history of a specific audit object within a particular audit task, the system can quickly retrieve all version change records corresponding to the audit object identifier and audit task identifier through the associated index relationship. These version change records are arranged in ascending order of operation timestamps, clearly showing the complete evolution of the working paper document from its initial state to its current state. For each version change record, the target node identifier identifies the specific location of each revision, the revised data content reveals the specific content of the revision, the set of change node records allows tracking the scope of the chain reaction triggered by each revision, and the impact propagation path analyzes how the dependencies between nodes affect the direction of change propagation. This complete change chain record provides a reliable data foundation for audit quality control, accountability, and compliance checks.

[0157] The version change management pipeline also supports change backtracking analysis. When an anomaly is detected in the data of a node in the working document, a reverse search can be performed in the version change management pipeline using that node's identifier to find all version change records related to that node. Analyzing the revised data content and operation timestamps in these version change records can determine the historical change trajectory of the node's data and the time point of each change. By combining the changes in node fingerprint values ​​in the change node record set, it is possible to verify whether each revision led to the expected data change and whether it triggered unexpected changes to related nodes. Through analysis of the impact propagation path, the root node causing the change can be traced, thereby pinpointing the true cause of the problem.

[0158] In multi-user collaborative auditing scenarios, the version change management chain supports conflict detection for concurrent revisions. When multiple auditors revise the same working paper, each revision operation generates an independent version change record and stores it in the version change management database. By retrieving all version change records within a specific time window through association index relationships and comparing the target node identifiers and change node record sets in these records, concurrent revisions targeting the same node can be identified. Combining the order of operation timestamps can determine the time sequence of revision operations, providing a basis for conflict resolution decisions. Through intersection analysis of impact propagation paths, it is possible to discover whether there are indirect logical dependency conflicts between different revision operations, even if the target nodes of the revisions are different, but data inconsistencies arise due to dependency propagation.

[0159] The storage structure of the version change management chain also supports incremental backup and fast recovery. Since each version change record contains a complete set of change node records and revised data content, the complete state of the original document can be reconstructed from any historical version by sequentially applying these version change records. When it is necessary to restore to a specific historical version, all version change records with operation timestamps earlier than the target recovery time are retrieved from the version change management database. These change records are then applied sequentially according to time to reconstruct the original document state at that point in time. Verification using node fingerprint values ​​ensures that the reconstructed document state is completely consistent with the historical record, guaranteeing the accuracy and reliability of the recovery operation.

[0160] A second aspect of this invention provides a system for automatically generating and managing special audit working papers, comprising:

[0161] An audit task unit is used to acquire audit task input information, which includes audit object identifier and audit type identifier, and to determine the logical dependency constraints between nodes in the audit type identifier.

[0162] The data filling unit is used to perform cross-node association verification on the dataset to be audited based on semantic consistency verification rules and generate missing data identifiers. According to the node position corresponding to the missing data identifier, it retrieves historical audit records that match the business attribute category of the current audit object identifier, extracts the data content of the corresponding node position in the historical audit records and combines the logical dependency constraints to infer and fill the missing positions to obtain the draft document.

[0163] The fingerprint generation unit is used to perform content fingerprint calculation on the data content of each node in the draft document, generate a set of node fingerprints, and when a revision operation is received for the draft document, determine the impact propagation path of the revision operation based on the logical dependency constraint.

[0164] The influence propagation unit is used to perform consistency verification on the sequence of associated node positions along the influence propagation path. When a logical dependency relationship is detected between the associated node data and the revised content, the associated node data is updated according to the logical dependency constraint and the node fingerprint is recalculated.

[0165] The version management unit is used to establish an association between the node fingerprint sets before and after the revision, the impact propagation path, and the audit task input information, forming a version change management link.

[0166] A third aspect of the present invention provides an electronic device, comprising:

[0167] processor;

[0168] Memory used to store processor-executable instructions;

[0169] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0170] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0171] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0172] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for automatically generating and managing special audit working papers, characterized in that, include: Obtain audit task input information, which includes audit object identifier and audit type identifier, and determine the logical dependency constraints between nodes in the audit type identifier; Based on semantic consistency verification rules, cross-node association verification is performed on the dataset to be audited, and missing data identifiers are generated. According to the node position corresponding to the missing data identifier, historical audit records that match the business attribute category of the current audit object identifier are retrieved. Data content of the corresponding node position in the historical audit records is extracted and the missing position is inferred and filled in combination with the logical dependency constraints to obtain the draft document. Content fingerprint calculation is performed on the data content of each node in the draft document to generate a node fingerprint set. When a revision operation is received for the draft document, the impact propagation path of the revision operation is determined based on the logical dependency constraint. The consistency of the associated node position sequence is verified along the influence propagation path. When a logical dependency is detected between the associated node data and the revised content, the associated node data is updated according to the logical dependency constraint and the node fingerprint is recalculated. The node fingerprint sets before and after the revision, the impact propagation path, and the audit task input information are linked and stored to form a version change management link.

2. The method according to claim 1, characterized in that, Determining the logical dependency constraints between nodes in the audit type identifier includes: Extract the node structure configuration information corresponding to the audit type identifier from the working paper template library. The node structure configuration information defines the identifier, hierarchical relationship and data attribute type of each node in the working paper document. Based on the hierarchical relationship, identify node pairs with data reference relationships, determine whether the data content of the source node can be used as the data generation input of the target node according to the compatibility rules between the data attribute types, and establish a dependency relationship marker between the source node and the target node when the determination result is yes. For node pairs with established dependency relationships, perform dependency direction analysis to determine the data flow direction between source nodes and target nodes. When the data generation of a target node depends on multiple source nodes, mark the multiple source nodes as a set of preceding dependent nodes. When the data content of a source node is called by the data generation process of multiple target nodes, mark the multiple target nodes as a set of subsequent dependent nodes. Traverse all nodes in the draft document, extract the set of preceding dependent nodes and the set of subsequent dependent nodes for each node, and associate and store the set of preceding dependent nodes, the set of subsequent dependent nodes and the node identifier to form the logical dependency constraint between nodes.

3. The method according to claim 1, characterized in that, Based on semantic consistency verification rules, cross-node association verification is performed on the dataset to be audited, and missing data identifiers are generated. Based on the node position corresponding to the missing data identifier, historical audit records matching the business attribute category of the current audit object identifier are retrieved, including: Based on the set of preceding dependent nodes of each node in the logical dependency constraint, determine the list of required data fields for each node; Iterate through the data content of each node in the dataset to be audited. When the actual data field identifier of each node is missing the data field identifier in the list of required data fields, record the node identifier and the missing data field identifier, and generate the missing data identifier. Extract the node identifier and data field identifier from the missing data identifier, query the set of preceding dependent nodes corresponding to the node identifier from the logical dependency constraint, determine whether the data content of each preceding dependent node in the set of preceding dependent nodes is complete, and when the determination result is yes, mark the missing data identifier as an independent missing type. For missing data identifiers marked as independent missing, extract the business attribute category features corresponding to the current audit object identifier; retrieve historical audit records from the historical audit working paper knowledge base that match the business attribute category features of the current audit object identifier.

4. The method according to claim 1, characterized in that, Extracting the data content of the corresponding node positions from the historical audit records and using the logical dependency constraints to infer and fill in the missing positions, the resulting working document includes: Based on the node identifier and data field identifier in the missing data identifier, extract historical data content from the historical audit records where both the node identifier and data field identifier match, and use the historical data content as the initial filling data. Extract the set of preceding dependent nodes corresponding to the node with the missing data identifier from the logical dependency constraint, and obtain the current data content of each preceding dependent node in the set of preceding dependent nodes in the dataset to be audited; The initial data is used to perform semantic compatibility verification with the current data content. The semantic compatibility verification determines whether the data format and numerical range of the initial data are logically conflicting with the current data content. When the semantic compatibility verification result is that there is no logical conflict, the initial padding data is used as the final padding data and filled into the node position corresponding to the missing data identifier. When the semantic compatibility verification result indicates a logical conflict, the data calculation relationship rules between the node corresponding to the missing data identifier and the set of preceding dependent nodes are extracted from the logical dependency constraints. Logical reasoning calculations are performed based on the data calculation relationship rules and the current data content to generate reasoning fill data. The reasoning fill data is then filled into the node position corresponding to the missing data identifier.

5. The method according to claim 1, characterized in that, Content fingerprints are calculated for the data content of each node in the draft document to generate a node fingerprint set. When a revision operation is received for the draft document, the impact propagation path of the revision operation is determined based on the logical dependency constraints, including: Traverse all nodes in the draft document, extract the data content of each node and perform hash operation to generate the node fingerprint value corresponding to each node, associate and store the node identifier of each node with the corresponding node fingerprint value to form a node fingerprint set. When a revision operation is received for the draft document, the target node identifier and the revised data content in the revision operation are parsed, and a hash operation is performed on the revised data content to generate a revised fingerprint value; Extract the original node fingerprint value corresponding to the target node identifier from the node fingerprint set, compare the revised fingerprint value with the original node fingerprint value, and when the comparison result is inconsistent, confirm that the revision operation has caused the data content of the target node to change; Extract the set of subsequent dependent nodes corresponding to the target node identifier from the logical dependency constraint. For each subsequent dependent node in the set of subsequent dependent nodes, recursively extract the set of subsequent dependent nodes corresponding to each subsequent dependent node. Construct a multi-level dependency propagation link starting from the target node identifier. Arrange all node identifiers contained in the multi-level dependency propagation link according to the dependency hierarchy to generate the influence propagation path.

6. The method according to claim 1, characterized in that, Consistency verification is performed on the sequence of associated node positions along the influence propagation path. When a logical dependency is detected between the associated node data and the revised content, the associated node data is updated according to the logical dependency constraint, and the node fingerprint is recalculated, including: According to the hierarchy of node identifiers in the propagation path, extract the identifiers of each associated node and the corresponding associated node data content in sequence. For the associated node identifier at the current level, obtain the actual data of the predecessor nodes of each predecessor node in the draft document from the corresponding predecessor dependency node set; Extract the data constraint rules between the associated node identifier and the set of preceding dependent nodes from the logical dependency constraints, and calculate the expected associated node data value based on the data constraint rules and the actual data of the preceding nodes; The associated node data content is compared with the expected associated node data value. When the comparison result is inconsistent, it is determined that the associated node data content and the revised content have a logical dependency relationship and need to be updated. The updated associated node data is recalculated and generated according to the data constraint rules and the actual data of the preceding node, and the original associated node data content corresponding to the node position of the associated node identifier is replaced. Perform a hash operation on the updated associated node data to generate an updated node fingerprint value, and update the node fingerprint value corresponding to the associated node identifier in the node fingerprint set to the updated node fingerprint value. Continue processing the associated node identifiers at the next level in the influence propagation path until the consistency verification and data update of all associated node identifiers in the influence propagation path are completed.

7. The method according to claim 5, characterized in that, The version change management chain is formed by establishing and storing the node fingerprint sets before and after the revision, the impact propagation path, and the audit task input information in a related manner, including: The difference between the node fingerprint set before revision and the node fingerprint set after revision is compared, the node identifiers whose node fingerprint values ​​have changed are extracted, and the node identifiers whose node fingerprint values ​​have changed, along with the corresponding node fingerprint values ​​before and after revision, are combined to form a set of changed node records. Extract the operation timestamp from the revision operation, and encapsulate the operation timestamp, the target node identifier, the revised data content, the set of change node records, and the impact propagation path in a structured manner to generate a version change record; Extract the audit object identifier and audit task identifier from the audit task input information, and establish an association index relationship between the audit object identifier, the audit task identifier and the version change record; The version change records and the associated index relationships are stored in the version change management database. The associated index relationships support the retrieval of all corresponding version change records based on the audit object identifier and the audit task identifier, thus forming the version change management chain.

8. A special audit working paper automatic generation and management system, used to implement the method as described in any one of claims 1-7, characterized in that, include: An audit task unit is used to acquire audit task input information, which includes audit object identifier and audit type identifier, and to determine the logical dependency constraints between nodes in the audit type identifier. The data filling unit is used to perform cross-node association verification on the dataset to be audited based on semantic consistency verification rules and generate missing data identifiers. According to the node position corresponding to the missing data identifier, it retrieves historical audit records that match the business attribute category of the current audit object identifier, extracts the data content of the corresponding node position in the historical audit records and combines the logical dependency constraints to infer and fill the missing positions to obtain the draft document. The fingerprint generation unit is used to perform content fingerprint calculation on the data content of each node in the draft document, generate a set of node fingerprints, and when a revision operation is received for the draft document, determine the impact propagation path of the revision operation based on the logical dependency constraints. The influence propagation unit is used to perform consistency verification on the sequence of associated node positions along the influence propagation path. When a logical dependency relationship is detected between the associated node data and the revised content, the associated node data is updated according to the logical dependency constraint and the node fingerprint is recalculated. The version management unit is used to establish an association between the node fingerprint sets before and after the revision, the impact propagation path, and the audit task input information, forming a version change management link.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images