An artificial intelligence-based dynamic data traceability question and answer method and system

Abstract

A dynamic data traceability question and answer method and system based on artificial intelligence, the method first acquires a question context package containing multi-dimensional information, then determines the syntax graph root node and extracts relevant parameters in combination with the index dictionary, then expands the syntax graph to generate a candidate blood relationship subgraph, then selects attribution according to semantics and institutional constraints, determines the path, and then constructs and checks the traceability map structure, and finally generates an explanatory answer and a traceability map, and forms an audit snapshot containing key information for review; the invention uses index identification, time range and organization area as three keys, combines the root node weighted score and threshold filtering of institutional priority, realizes the unique positioning of the root node, and through the syntax graph solidification range and institutional constraints, embeds the rules such as no loop, ensures that the combined link must be included, and the selected branch is screened into the boundary and semantics, completes the root node determination and bounded expansion in the same context, effectively reduces the cross-range ambiguity, and guarantees the consistency of the structure and constraints.

Description

Technical Field

[0001] This invention relates to a source tracing question-answering method, belonging to the field of data intelligence and explainable question answering, and particularly to a dynamic data source tracing question-answering method and system based on artificial intelligence. Background Technology

[0002] In data governance and indicator diagnosis scenarios, enterprises need to quickly locate the root cause of problems and form explainable conclusions around dimensions such as indicators, time, and organization. However, the definition of indicators will be dynamically adjusted with the system, and the data sources are multiple and scattered. A single query or static rule is difficult to meet the needs of review and audit.

[0003] In existing technologies, the following methods are mainly used to achieve this: First, through indicator dictionaries and metric repositories, which centrally manage indicator aliases, formulas, dimensions, units, caliber levels, and sources, and add fields for validity period and scope of application; second, data lineage and graph tools, which build directed dependency relationships based on ETL metadata and task DAGs for upstream tracing and impact analysis; third, rule-driven semantic parsing and entity mapping, which are used to extract and standardize indicators, time, and organizational fragments in the problem; fourth, question-and-answer systems based on knowledge graphs or retrieval enhancement, which generate business-oriented descriptions according to retrieval matching strategies; and fifth, compliance and auditing modules, which conduct field-level verification and output reports based on organizational permissions, source trust level, and version sequence.

[0004] While the aforementioned existing technical solutions can achieve multi-layered collaboration and provide modular capabilities for indicator diagnosis, including retrieval, tracing, explanation, and compliance, they still have the following shortcomings: they lack a problem context-driven mechanism that integrates indicator identification, time range, organizational region, semantic clues, and institutional parameters; they cannot dynamically expand lineage links within a unified boundary; they are difficult to conduct comparative attribution on alternative branches; they cannot solidify results according to the minimum completeness standard of evidence anchors; and the explanation, tracing, and auditing processes lack isomorphic mapping, making it difficult to form a closed-loop support; and the consistency of data tracing questions and answers is poor. Summary of the Invention

[0005] The purpose of this invention is to overcome the aforementioned defects and problems in the prior art and to provide a dynamic data tracing question-and-answer method and system based on artificial intelligence with better consistency in data tracing question-and-answer.

[0006] To achieve the above objectives, the technical solution of the present invention is: a dynamic data tracing and question-answering method based on artificial intelligence, comprising:

[0007] The diagnostic problem is obtained and the problem context package is parsed; the problem context includes indicator identifiers, time range, organizational region, semantic clues, and institutional parameters;

[0008] Based on the problem context package, the root node of the conjunctive-selective indicator source syntax graph is determined in the indicator dictionary, and the caliber metadata and institutional constraint parameters are extracted.

[0009] Based on the root node of the conjunction-selection indicator sourcing syntax graph and the problem context package, expand the conjunction-selection indicator sourcing syntax graph, retain the conjunction steps and include the selection branches that satisfy the problem context package, mark the evidence anchor points and generate the final candidate lineage subgraph.

[0010] Attribution selection is performed based on candidate lineage subgraphs and problem context packages. Each selected branch is used as an explanatory hypothesis. Matching is performed based on semantic clues and institutional constraint parameters. Sets that meet semantic matching and have complete evidence anchors are added to the committable set. Sets that do not meet the requirements are used to generate rejection reason labels. Path consistency is calculated within the committable set to determine the main path and comparison path, and attribution selection results are obtained.

[0011] The source map structure is generated based on the attribution selection results, and the main path of the source map structure is encoded as a set of directed links, and the folded comparison path is a comparison unit. The rejection reason labels are bound to the relevant nodes and directed links, and the evidence anchors and caliber meta-information are mapped to the node attributes of the source map structure. The consistency of the source map structure is checked.

[0012] Explanatory answers are generated based on the source map structure, and the main path criteria, institutional constraints and evidence anchors are linked together to output the explanatory answers and the source map structure.

[0013] An audit snapshot is generated based on the attribution selection results, recording the issue context package, the source map structure, and the attribution selection results to form a verifiable object.

[0014] The process involves acquiring and parsing the diagnostic problem to obtain a problem context package. This problem context includes indicator identifiers, time ranges, organizational regions, semantic clues, and institutional parameters, specifically:

[0015] Taking the diagnostic question as input, semantic cues are identified and text fragments directly related to the diagnostic question are extracted as the units to be parsed.

[0016] Based on the unit to be parsed, the indicator identifier, the standardized expression time range, and the normalized mapping organization region are determined to form the initial result of indicator identifier, time range, and organization region;

[0017] Based on semantic clues and initial results, institutional parameters are extracted and constraints are identified. The consistency between institutional parameters and indicator identifiers, time ranges, and organizational regions is checked to obtain the problem context package.

[0018] The method of determining the root node of the conjunctive-selective indicator source graph in the indicator dictionary based on the problem context package, and extracting caliber metadata and institutional constraint parameters, specifically includes:

[0019] Using the problem context package as input, the indicator dictionary is searched, the indicator identifier is used for unique positioning, and the validity period and scope of application are filtered by time range and organizational region to obtain a set of candidate indicator entries;

[0020] The institutional parameters of the candidate indicator item set are matched, and the corresponding fields of the institutional parameters are compared with the institutional constraint parameters in the candidate indicator item set to determine the subset of candidate indicator items that meet the institutional constraints.

[0021] The candidate indicator items subset is mapped to the conjunctive-selective indicator traceability syntax graph, and the root node of the conjunctive-selective indicator traceability syntax graph is located based on the matching of indicator identifier and scope of application. If there are multiple matches, the system constraint parameters are used as the priority for judgment, and then the time range coverage is used to determine the root node.

[0022] Using the root node of the conjunctive-selective indicator source grammar graph as the anchor point, extract the caliber metadata bound to the root node from the indicator dictionary, and load the formula, dimension, unit, caliber level and source into the caliber metadata by field;

[0023] The institutional constraint parameters bound to the root node in the indicator dictionary are extracted, and consistency checks are performed using the scope of application field, validity period field, and organization permission field. The parameters are then compared item by item with the institutional parameters in the issue context package to obtain the institutional constraint parameters.

[0024] The generation of the final candidate lineage subgraph specifically includes:

[0025] Using the root node of the conjunction-selection indicator source syntax graph and the problem context package as input, the conjunction-selection indicator source syntax graph is expanded layer by layer, and the expansion boundary is limited to the time range and organization region in the problem context package to establish the upstream dependency set starting from the root node.

[0026] Identify the conjunctive elements in the upstream dependency set, include the upstream nodes and directed edges corresponding to the conjunctive elements into the initial candidate lineage subgraph, and maintain the dependency order consistent with the order of the conjunctive elements.

[0027] The selection branches in the upstream dependency set are filtered by institutional parameters. The institutional parameters in the problem context package are compared with the institutional constraint parameters bound to the selection branches one by one to obtain the set of selection branches that comply with the regulations.

[0028] Semantic matching is performed on the set of branches for compliance with regulations. The semantic clues in the problem context package are matched with the caliber metadata bound to the branches to determine the branches that meet the semantic matching, and the branches that meet the semantic matching are included in the initial candidate lineage subgraph.

[0029] The nodes included in the initial candidate lineage subgraph are checked for consistency in time range and consistency in tissue region. Time range coverage and tissue region inclusion are used for checking, and nodes and directed edges that do not meet the consistency requirements are removed.

[0030] Mark evidence anchors at the node level of the initial candidate lineage subgraph. Based on the minimum completeness of evidence anchors, bind at least one evidence anchor to each conjunctive link and each included selection branch, and map the evidence anchors to the corresponding nodes.

[0031] The initial candidate lineage subgraph is subjected to topological acyclicity checks and caliber-level consistency checks. Nodes and directed edges that do not meet the checks are removed until the checks are met, thus obtaining the final candidate lineage subgraph.

[0032] The process of obtaining the attribution selection result specifically includes:

[0033] Using the semantic clues and institutional constraint parameters in the candidate lineage subgraph and the problem context package as input, an explanatory hypothesis set is established for each selected branch in the candidate lineage subgraph and a list of to-be-determined branches is generated.

[0034] The list of judgments is subjected to institutional constraint judgments. The institutional constraint parameters are compared with the institutional constraint parameters bound to the selected branches one by one to filter out the set of institutional compliant branches.

[0035] Semantic matching is performed on the set of regulatory compliance branches, and semantic clues are matched with the metadata of the scope of the selected branches to obtain a set of semantically matched branches;

[0036] The completeness of the evidence anchor points is verified on the semantic matching branch set. The minimum completeness of the evidence anchor points is used as the standard to verify whether the evidence anchor points of each conjunctive link and the included selection branches are complete, and the complete evidence branch set is determined.

[0037] The set of branches with complete evidence is added to the set of branches that can be committed to. For branches that do not meet the institutional constraints, semantic matching, or incomplete evidence anchors, rejection reason labels are generated. The rejection reason labels include semantic mismatch, non-compliance with the system, incomplete evidence anchors, time out of bounds, outdated or unreliable source, and unverifiable conversion.

[0038] Path consistency is calculated within the committable set, and time range coverage alignment, unit consistency, caliber hierarchy consistency, data source version time sequence consistency, and topological acyclicity are used as check items to determine the main path and comparison path.

[0039] The main path and the comparison path, along with the set of commitments, are summarized to obtain the attribution selection results.

[0040] The generated source tracing map structure specifically includes:

[0041] Using the attribution selection results and candidate lineage subgraphs as input, initialize the node set and directed edge set of the source map structure, and limit the construction scope of the source map structure according to the main path and comparison path in the attribution selection results;

[0042] The main path is encoded, and a set of directed links from upstream to downstream is constructed based on the dependencies in the candidate lineage subgraph. The nodes and directed edges involved in the main path are added to the source map structure while maintaining the dependency order.

[0043] The comparison path is folded, compressing the set of directed links corresponding to each comparison path into a single comparison unit, while maintaining the association between the comparison unit and its start and end nodes.

[0044] The rejection reason labels are bound to the relevant nodes and directed edges. Based on the rejection reason labels in the attribution selection results, the rejection reason labels are associated with the comparison units and the nodes and directed edges corresponding to the uncommitted selection branches, respectively.

[0045] The evidence anchors are mapped to node attributes, and the evidence anchors in the candidate lineage subgraphs are written into the corresponding node attributes in the tracing map structure according to the correspondence, while maintaining the reference relationship between the evidence anchors and the directed link set.

[0046] Map the caliber metadata to node attributes, write the formula, dimension, unit, caliber level and source into the node attributes involved in the main path and comparison unit, and establish the association between the caliber metadata and the corresponding evidence anchor points;

[0047] The consistency of the traceability map structure is checked, including the directionality and connectivity of the directed link set and the integrity of the reference of the comparison unit. Directed edges that fail the check are removed, and the association relationship with node attributes and comparison units is updated synchronously.

[0048] The output explanatory answers and source map structure specifically include:

[0049] Using the source map structure as input, the directed link set and comparison unit of the main path are parsed to establish the main path node order and corresponding reference order;

[0050] The main path node sequence is adopted to organize the caliber metadata according to the main path node sequence, clarify the indicator identifier, time range and organization area, and make the caliber metadata correspond to the main path nodes;

[0051] The institutional constraint parameters are linked to the main path nodes, and described by fields such as scope of application, validity period, organizational authority and data source trust level, so that the institutional constraint parameters after each main path node are directly related to that node.

[0052] The evidence anchors are bound to the main path nodes for reference. The minimum completeness of the evidence anchors is used as the standard to ensure that each main path node references at least one evidence anchor. The data source identifier, version or timestamp and the extracted fingerprint are inserted into the node description according to the corresponding relationship.

[0053] The comparison units are explained, and the rejection reason tags corresponding to the comparison units are paired with the start and end nodes of the comparison units. The main path in the set of commitments is used as a reference to select the comparison units to be presented and the basis for selection is given one by one.

[0054] The consistency between the explanatory answers and the source map structure is checked. The time range coverage judgment and the organizational area inclusion judgment are used to check whether the reference of the caliber meta-information and institutional constraint parameters are consistent with the source map structure. Inconsistent references are removed and the order of the main path nodes is rearranged.

[0055] The main path criteria, institutional constraints, and evidence anchors are linked together to generate explanatory answers, and the explanatory answers and source map structure are output.

[0056] The generation of the audit snapshot specifically includes:

[0057] Write the main path directed link set and comparison unit folding mapping of the source tracing map structure into the audit snapshot in the order of dependency, record the formula, dimension, unit, caliber level and source in the node attributes, and bind the evidence anchor point and rejection reason label to the relevant node and directed edge according to the corresponding relationship;

[0058] Record the main path, comparison path, and branches of the committable set in the attribution selection results to the audit snapshot, and associate the rejection reason labels in the attribution selection results with the corresponding nodes and directed edges in the source map structure;

[0059] A one-time consistency check is performed on the audit snapshot; the check includes: checking whether the time range coverage determination and the organizational area inclusion determination are consistent, checking whether the directionality and connectivity of the main path are consistent with the source map structure, and checking whether the completeness indication of the evidence anchor point on the main path node is consistent with the attribution selection result; if all are consistent, the check is passed, and the audit snapshot is solidified as a re-verifiable object.

[0060] An artificial intelligence-based dynamic data tracing and question-answering system, which is applied to the above-mentioned method, includes:

[0061] The diagnostic problem parsing module is used to obtain diagnostic problems and parse them to obtain problem context packages; the problem context includes indicator identifiers, time ranges, organizational regions, semantic clues, and institutional parameters;

[0062] The root node determination module is used to determine the root node of the conjunctive-selective indicator source syntax graph in the indicator dictionary based on the problem context package, and to extract the caliber metadata and institutional constraint parameters.

[0063] The candidate lineage subgraph generation module is used to expand the conjunctive-selection criterion sourcing syntax graph based on the root node of the conjunctive-selection criterion sourcing syntax graph and the problem context package, retain the conjunctive links and include the selection branches that satisfy the problem context package, mark the evidence anchor points and generate the final candidate lineage subgraph.

[0064] The attribution selection result acquisition module is used to perform attribution selection based on candidate lineage subgraphs and problem context packages. It uses each selection branch as an explanatory hypothesis, matches semantic clues and institutional constraint parameters, adds the set that meets semantic matching and has complete evidence anchors to the committable set, generates rejection reason labels for the set that does not meet the requirements, calculates path consistency within the committable set, determines the main path and comparison path, and obtains the attribution selection result.

[0065] The source map structure generation module is used to generate a source map structure based on the attribution selection results, and encode the main path of the source map structure as a set of directed links and the folded comparison path as a comparison unit; it binds the rejection reason labels with relevant nodes and directed links, maps the evidence anchor points and caliber meta-information to the node attributes of the source map structure, and performs consistency verification on the source map structure.

[0066] The explanatory answer generation module is used to generate explanatory answers based on the source map structure, and connect the main path clarification, institutional constraints and evidence anchors to output the explanatory answer and the source map structure.

[0067] The audit snapshot generation module is used to generate audit snapshots based on the attribution selection results, record the issue context package, the source map structure, and the attribution selection results, and form a verifiable object.

[0068] A dynamic data tracing and question-answering device based on artificial intelligence, the device including a processor and a memory;

[0069] The memory is used to store computer program code and to transmit the computer program code to the processor;

[0070] The processor is used to execute the above-described AI-based dynamic data tracing and question-answering method according to the instructions in the computer program code.

[0071] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0072] 1. This invention discloses a dynamic data tracing question-answering method and system based on artificial intelligence. The method first obtains a question context package containing multi-dimensional information, then determines the root node of the syntax graph and extracts relevant parameters by combining an indicator dictionary. Next, it expands the syntax graph to generate a candidate lineage subgraph. Subsequently, it performs attribution selection based on semantics and institutional constraints to determine the path, then constructs and verifies the tracing map structure, and finally generates an explanatory answer and a tracing map. At the same time, it forms an audit snapshot containing key information for review. In application, this design uses indicator identifier, time range, and organizational region as three-key indexes, combined with weighted scoring of root nodes based on institutional priority and threshold filtering, to achieve unique positioning of root nodes. Furthermore, it solidifies the scope and institutional constraints through the syntax graph, embedding rules such as acyclicity to ensure that conjunctive links are included and selective branches are screened according to boundaries and semantics. The determination of root nodes and bounded expansion are completed within the same context, effectively reducing cross-scope ambiguity and ensuring the consistency of structure and constraints.

[0073] 2. The present invention provides a dynamic data tracing question-answering method and system based on artificial intelligence. It takes the minimum completeness standard of evidence anchor points and the mechanism for generating committable sets as its core. First, it determines the institutional compliance and semantic matching degree of each selected branch. Then, it determines the main path and comparison path through multi-dimensional scoring and ranking of path self-consistency. Simultaneously, it generates and binds rejection reasons for non-committed branches. In application, this design forms a verifiable boundary for choice in interpretation, enabling a unified expression of branch-level attribution and evidence constraints. This avoids making untraceable path decisions based solely on retrieval similarity or a single rule, meeting the scenario's requirements for institutional consistency and verifiable evidence.

[0074] 3. The present invention provides a dynamic data tracing question-answering method and system based on artificial intelligence. It encodes a set of directed links using the main path, folds the comparison path into comparison units, maps metadata and evidence anchors to node attributes, and performs consistency checks on directionality, connectivity, time range coverage, and organizational area inclusion. Simultaneously, it solidifies the question context package, tracing map, and attribution selection results into an audit snapshot, establishing isomorphic reference relationships at the node and link levels. In application, this design achieves consistent output and a verifiable closed loop for explanation, tracing, and auditing. It directly supports a unified driving force with the question context package as the sole entry point in indicator diagnosis scenarios, achieving a complete technical effect from retrieval and expansion to comparative selection and auditable implementation. Attached Figure Description

[0075] Figure 1 This is a flowchart of the method of the present invention.

[0076] Figure 2 This is a flowchart of the index dictionary retrieval and positioning process in Embodiment 1 of the present invention.

[0077] Figure 3This is a flowchart of the bounded expansion of the syntax graph in Embodiment 1 of the present invention.

[0078] Figure 4 This is a comparative attribution flowchart in Embodiment 1 of the present invention.

[0079] Figure 5 This is a flowchart of the structure for generating a traceability map in Embodiment 1 of the present invention.

[0080] Figure 6 This is a system structure diagram of the present invention.

[0081] Figure 7 This is a structural diagram of the device of the present invention.

[0082] In the diagram: 1. Diagnostic problem analysis module; 2. Root node determination module; 3. Candidate lineage subgraph generation module; 4. Attribution selection result acquisition module; 5. Source map structure generation module; 6. Explanatory answer generation module; 7. Audit snapshot generation module; 8. Processor; 9. Memory; 9. Computer program code. Detailed Implementation

[0083] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0084] Example 1:

[0085] See Figure 1 A dynamic data tracing and question-answering method based on artificial intelligence includes:

[0086] S1. Obtain the diagnostic problem and parse the problem context package; the problem context includes indicator identifiers, time range, organizational region, semantic clues, and institutional parameters;

[0087] In this embodiment, the diagnostic question is taken as input, and the diagnostic question is structured and parsed. Sentence segmentation and term localization are used to split the diagnostic question into bounded paragraphs, and entity recognition and relation recognition are performed on each paragraph to extract text fragments directly related to the diagnostic target, forming the unit to be parsed.

[0088] During the formation of the unit to be parsed, non-informative fragments are removed, retaining only fragments containing indicator names, time descriptions, and organizational descriptions to ensure the effectiveness of subsequent matching steps. Semantic clue identification is performed on the unit to be parsed. Based on lexical merging and a context window, semantic clues related to indicators, time, and organization are extracted as semantic clues. These semantic clues include keyword groups indicating indicator names, time phrases indicating time boundaries, and organizational phrases indicating organizational levels, ensuring that semantic clues cover the key clues required for subsequent rule matching.

[0089] Based on the unit to be parsed and semantic clues, the indicator identifier is determined by rule matching and judgment threshold.

[0090] Taking keyword groups related to the indicator name as input, the system matches preset indicator aliases with standardized naming rules, calculates a matching score for each candidate name, and compares it with a judgment threshold. The comparison is performed, and the candidate names that meet the threshold are selected as the indicator identifiers.

[0091] The time range is standardized by parsing time phrases containing start and end boundaries into uniform interval expressions. When processing absolute dates and relative times, the boundaries are kept clear and merged using the same time base to form comparable time ranges with start and end boundaries. Organizational regions are normalized by mapping hierarchical titles and regional titles in organizational phrases according to a preset lookup table, outputting hierarchically defined organizational regions that can be used for subsequent determinations.

[0092] After the initial results of indicator labels, time ranges, and organizational regions are formed, a one-time internal consistency check is performed on the three to ensure that the names of the indicator labels do not conflict with the hierarchical levels of the organizational regions, and that the time limits of the time ranges cover the time requirements for diagnosing the problems. Items that do not meet the consistency requirements are corrected or removed so that the indicator labels, time ranges, and organizational regions form usable initial results.

[0093] Based on the initial results of semantic cues, indicator identifiers, time ranges, and organizational regions, the system parameters are extracted and constraint items are identified. The binding expressions in the semantic cues are used as guides to locate clause fragments related to the system and extract them as a set of system parameters by field. During the extraction process, constraints involving the scope of application are compared with the organizational region, constraints involving time validity are compared with the time range, and constraints involving organizational authority are compared with the usage scenarios of the indicator identifiers, forming a set of determinable constraint items. A consistency check is performed on the system parameters with the indicator identifiers, time ranges, and organizational regions, using a judgment threshold. As a threshold for judgment, each constraint is judged to be compatible with the indicator identifier, time range and organizational region. If there is an incompatible constraint, the corresponding initial result is corrected or the constraint is marked as inapplicable to ensure that the binding expression of the system parameters is consistent with the indicator identifier, time range and organizational region.

[0094] The indicator identifier, time range, organizational region, and institutional parameters are assembled to form a problem context package. During the assembly process, the indicator identifier is used as the core identifier field, the time range as the time range field, the organizational region as the organizational region field, and the institutional parameters as the institutional parameter field, while maintaining reference relationships with semantic clues. This ensures that the problem context package contains both structured identifier and range information, and retains semantic clues for subsequent judgment.

[0095] S2. Based on the problem context package, determine the root node of the conjunctive-selective indicator source syntax graph in the indicator dictionary, and extract the caliber metadata and institutional constraint parameters;

[0096] See Figure 2 In this embodiment, the problem context packet is used as input to search the indicator dictionary. The indicator identifier is used to uniquely locate the corresponding indicator entry. The location results are filtered by time range and organizational region. Start and end boundaries and hierarchical mapping are used for judgment to form a set of candidate indicator entries. The completeness of the candidate indicator entry set is verified, and candidate entries with missing key fields are removed to maintain the usability of the candidate indicator entry set.

[0097] The candidate indicator item set is matched with the institutional parameters. For each candidate item in the candidate indicator item set, the institutional constraint parameters bound to it are compared at the field level to form a subset of candidate indicator items that meet the institutional constraints. During the comparison process, the consistency of the applicable scope field with the root node of the conjunctive-selective indicator traceability syntax graph is checked; the consistency of the validity period field with the time range is checked; and the consistency of the organizational authority field with the applicable scenario of the indicator identifier is checked. A threshold is used to determine whether each comparison item passes or fails, and candidate items that fail are removed, ensuring that the subset of candidate indicator items retains only those that comply with the institutional regulations. Duplicate items are eliminated from the subset of candidate indicator items to maintain the uniqueness of each item in terms of indicator identifier and applicable scope.

[0098] A subset of candidate indicator entries is mapped to a conjunctive-selective indicator source syntax graph. Unique location is achieved using indicator identifiers and applicable scopes, thus determining the root node of the conjunctive-selective indicator source syntax graph. When multiple matches exist, institutional constraints are prioritized, followed by time range coverage. Entries with insufficient time range coverage are eliminated, and entries with identical coverage are ranked according to organizational region inclusion relationships, ensuring consistency and uniqueness in the determination of the root node of the conjunctive-selective indicator source syntax graph. After determining the root node of the conjunctive-selective indicator source syntax graph, upstream nodes are minimally expanded and verified using the root node as an anchor point. Only upstream nodes with the same indicator identifier and time range as the root node of the conjunctive-selective indicator source syntax graph are retained, providing stable anchor points for subsequent extraction of metadata and institutional constraint parameters.

[0099] The root node selection calculation adopts a priority-based approach, taking a subset of candidate indicator items as the candidate set, and calculating a root node selection score for each candidate item. The expression is as follows:

[0100] ;

[0101] in: Candidate entries are selected from a subset of candidate indicator entries; To ensure compliance with the system parameters, they are normalized into range values ​​based on consistency with each item in terms of scope of application, validity period, and organizational authority. The time range coverage is normalized to an interval value according to the coverage ratio of the start and end boundaries; To organize the regional coverage, the values ​​are normalized to interval values ​​according to the hierarchical inclusion relationship; , , All are weighting coefficients and satisfy weighting constraints.

[0102] Using the threshold for judging the compliance of institutional parameters and the threshold for judging the coverage of time range as the pre-filtering criteria, the candidate items whose scores meet the pre-filtering criteria are sorted, and the node corresponding to the candidate item with the highest score is selected as the root node of the conjunction-selection indicator traceability syntax graph.

[0103] Using the root node of the conjunctive-selective indicator source grammar graph as the anchor point, the caliber metadata bound to the root node is extracted from the indicator dictionary. Formulas, dimensions, units, caliber levels, and sources are loaded as caliber metadata by field, ensuring that each field originates from the structured definition in the indicator dictionary and that no undefined external fields are introduced. Internal consistency checks are performed on the caliber metadata, aligning units with caliber levels and mapping dimensions with indicator identifiers, forming a set of caliber metadata directly usable for grammar graph expansion. Inconsistencies identified in the above checks are corrected using field backfilling or alternative field selection to maintain the integrity of the caliber metadata.

[0104] The institutional constraint parameters bound to the root node of the conjunctive-selectivity indicator tracing syntax graph in the indicator dictionary are extracted. Consistency checks are performed using the scope of application, validity period, and organizational authority fields, and each parameter is compared with the institutional parameters to form institutional constraint parameters consistent with the problem context. Differences between the institutional constraint parameters and the institutional parameters are marked and adjusted. Validity periods exceeding the time range are shortened, and scopes of application not included in the root node of the conjunctive-selectivity indicator tracing syntax graph are replaced or removed, ensuring that the institutional constraint parameters are consistent with the problem context in terms of boundary conditions. The checked institutional constraint parameters are then fixed to ensure clear institutional boundaries for subsequent attribution selection.

[0105] In the conjunctive-selective indicator sourcing syntax graph, the root node is marked as an indicator node, the necessary steps directly connected to the root node are marked as conjunctive nodes, the alternative calibers or sources corresponding to the root node are marked as selection nodes, intermediate processing involving data transformation is marked as transformation nodes, and the data source is marked as a data source node. Using the problem context package as input and the root node, caliber metadata, and institutional constraint parameters of the conjunctive-selective indicator sourcing syntax graph as output, it ensures that conjunctive nodes are connected to the root node with directed edges, and that each selection node is bound to caliber metadata and institutional constraint parameters at the branch level. This ensures that the root node, caliber metadata, and institutional constraint parameters of the conjunctive-selective indicator sourcing syntax graph can be referenced as stable anchor points and boundary conditions in the subsequent construction of candidate lineage subgraphs and the invocation of comparative sourcing attribution.

[0106] Further explanation: The conjunctive-selective indicator tracing syntax graph network consists of five types of nodes and four types of directed edges. Nodes include indicator nodes, conjunctive nodes, selective nodes, transformation nodes, and data source nodes; directed edges include conjunctive edges, selective edges, transformation edges, and source edges. Each node binds a caliber metadata field and a regulatory constraint parameter field at the attribute layer. The caliber metadata field includes the formula, dimension, unit, caliber level, and source; the regulatory constraint parameter field includes the scope of application, validity period, and organizational permissions. At the node level, version, timestamp, and reference relationship with the evidence anchor are retained. The indicator node, as the central anchor point of the network, uniquely corresponds to the indicator identifier, and its unique location as the root node is achieved in the network index using a three-key index composed of the indicator identifier, time range, and organizational region. Conjunction nodes represent essential steps, joining upstream nodes only through conjunction edges; alternative nodes represent substitutable branches, connected to indicator nodes or conjunction nodes through alternative edges, and bound to caliber metadata and regulatory constraint parameters at the branch level; transformation nodes handle intermediate processing, binding transformation rules, input-output caliber relationships, and verifiable parameters, and connecting to their preceding and following nodes through transformation edges; data source nodes bind source type and version time sequence, and connect to the generation node through source edges. The network embeds acyclic constraints, unit consistency and caliber hierarchy consistency constraints, time range coverage and organizational region inclusion constraints within the topology layer. Constraints are expressed as combinations of node and edge attributes, independent of external computation. In terms of hierarchical organization, indicator nodes are central, with upstream components consisting of conjunction steps and alternative branches forming a multi-level structure. Each alternative branch connects to transformation nodes and data source nodes, maintaining dependency order and field alignment through directed edges. The problem context is packaged in the index layer for root node location and boundary constraints. Network output directly reads the root node, caliber metadata, and regulatory constraint parameters at the node level.

[0107] S3. Expand the conjunctive-selection criterion sourcing syntax graph based on the root node of the conjunctive-selection criterion sourcing syntax graph and the problem context package, retain the conjunctive steps and include the selection branches that satisfy the problem context package, mark the evidence anchor points and generate the final candidate lineage subgraph.

[0108] See Figure 3 In this embodiment, the root node of the conjunction-selection index source grammar graph and the problem context package are used as inputs to expand the conjunction-selection index source grammar graph layer by layer. The time range and the root node of the conjunction-selection index source grammar graph are used as the expansion boundary. On the directed edges, the expansion is carried out only along the direction that satisfies the time range and the root node of the conjunction-selection index source grammar graph. An upstream dependency set starting from the root node of the conjunction-selection index source grammar graph is established, and a traversable sequence is generated according to the node in-degree and the direction of the directed edge, which serves as the basis for subsequent identification and inclusion.

[0109] The upstream dependency set is identified by conjunctive links. The necessary links that the root node of the conjunctive-selection index source syntax graph depends on are taken as conjunctive links. Based on the traversable sequence, all upstream nodes and directed edges corresponding to the conjunctive links are included in the initial candidate lineage subgraph, and the dependency order is kept consistent with the order of the conjunctive links. During the inclusion process, the predecessor and successor relationship of each conjunctive link is bound by directed edges to ensure the stability of the directionality and connectivity of the main path in the initial candidate lineage subgraph, and to provide structured anchor points for the screening of subsequent selection branches.

[0110] The selected branches in the upstream dependency set are filtered by institutional parameters. The institutional parameters in the problem context package are compared with the institutional constraint parameters bound to each selected branch. The applicable scope field, validity period field and organizational permission field are judged by a judgment threshold to form a set of selected branches that comply with the system. After forming a set of selected branches that comply with the system, non-compliant branches are removed to ensure that the subsequent semantic matching process is only carried out within the compliance boundary, thereby reducing the inclusion of invalid paths and the burden of subsequent verification.

[0111] Semantic matching and comprehensive judgment are performed on the selected branches for compliance with regulations. Semantic clues are matched with the metadata associated with each selected branch. A joint decision-making process based on matching score and coverage is used to determine the selected branches that meet both semantic matching and the minimum completeness standard of evidence anchors. These selected branches are then included in the initial candidate lineage subgraph. To ensure clear boundaries in the inclusion process, the following comprehensive judgment function is used to calculate each selected branch:

[0112] ;

[0113] in: This is a comprehensive decision function; To select a branch; The system parameter represents compliance, and its value is in the range of [0,1]. The semantic matching degree takes values ​​in the range [0,1] and is expressed in terms of semantic clues. With caliber metadata For input; This represents the time range coverage, with values ​​in the range [0,1]. The threshold for determining time range coverage; To determine the coverage area, the value ranges from [0,1]. Set a threshold for determining whether an organization region contains such a threshold. This is a topological penalty term, with values ​​in [0,1]. This is a penalty term based on caliber level, with values ​​in [0,1]. As an indicator of the completeness of the anchor point of evidence, For indicator functions; , , , , , All are weighting coefficients; the comprehensive judgment function is compared with the comprehensive judgment threshold, and branches that meet the threshold are included, while branches that do not meet the threshold are removed.

[0114] The nodes included in the initial candidate lineage subgraph are checked for consistency in time range and organizational region. The root node of the conjunctive-selectivity index source syntax graph is used as the benchmark for determining time range coverage and organizational region inclusion, respectively. The time field and organizational field of each node are checked, and nodes and directed edges that do not meet the consistency requirements are removed, while maintaining the dependency order to prevent structural breakage. During the removal process, nodes that exceed the boundaries are marked and cleaned up to ensure that the initial candidate lineage subgraph only retains nodes and directed edges within the boundaries of the time range and the root node of the conjunctive-selectivity index source syntax graph.

[0115] Evidence anchors are marked at the node level of the initial candidate kinship subgraph. A minimum completeness standard for evidence anchors is adopted, binding at least one evidence anchor to each conjunctive step and each included selection branch. Evidence anchors are mapped to corresponding nodes, and the reference relationships of evidence anchors are recorded on directed edges, enabling evidence anchors to support the evidence completeness verification of the subsequent comparative attribution system. During the mapping process, data source identifiers, version or timestamps, and extraction fingerprints are used as basic fields of the evidence anchors, ensuring that the evidence anchors can be directly referenced and verified.

[0116] The initial candidate lineage subgraph undergoes topological acyclicity checks and caliber level consistency checks. Directed cycle detection is used as the basis for topological acyclicity checks, and comparison of node caliber level with caliber metadata is used as the basis for caliber level consistency checks. Nodes and directed edges that do not meet the checks are removed using a threshold. If the checks are met, the initial candidate lineage subgraph becomes the final candidate lineage subgraph for subsequent use by a comparative attribution attribution selector. This ensures the candidate lineage subgraph remains stable in structure and caliber level and consistently references the root node and problem context package of the conjunctive-selectivity criterion grammar graph.

[0117] S4. Attribution selection is performed based on the candidate lineage subgraph and the problem context package. Each selected branch is used as an explanatory hypothesis. Matching is performed based on semantic clues and institutional constraint parameters. The set that meets the semantic matching and has complete evidence anchor points is added to the committable set. The set that does not meet the requirements is used to generate rejection reason labels. Path consistency is calculated within the committable set to determine the main path and the comparison path, and the attribution selection results are obtained.

[0118] See Figure 4 In this embodiment, the candidate lineage subgraph and the problem context package are used as input. An explanation hypothesis set is established for each selected branch in the candidate lineage subgraph, a list to be determined is generated, and the traversal order is kept consistent with the direction of the directed edges of the candidate lineage subgraph, so that the determination and path splicing have consistent structural references.

[0119] The institutional constraint judgment unit performs institutional constraint judgment on the list of candidates to be judged. It compares each institutional constraint parameter with the institutional constraint parameters bound to the selected branches, applying judgment thresholds to the scope of application, validity period, and organizational authority fields to determine compliance, thus forming a set of compliant branches. Selected branches that do not meet the judgment thresholds are removed, and the corresponding fields that fail are recorded to support the generation of subsequent rejection reason labels, ensuring that the institutional constraint judgment is consistent with the issue context in terms of boundary conditions.

[0120] The semantic matching unit performs semantic matching on the set of regulatory compliance branches, matching semantic clues with the metadata associated with the selected branches. A judgment threshold is used to determine the semantic similarity and pass the test, forming a set of semantically matched branches. Selected branches that do not meet the semantic matching requirements are removed, and the uncovered items corresponding to the semantic clues are used as the source of subsequent rejection reason tags, maintaining consistency between the semantic judgment and the field-level metadata.

[0121] The evidence anchor verification unit performs evidence anchor completeness verification on the semantic matching branch set. Using the minimum evidence anchor completeness standard, each conjunct step and each included selection branch is bound to at least one evidence anchor. Selection branches lacking evidence anchors are removed, forming a complete evidence branch set. For selection branches with incomplete evidence, the corresponding missing fields and associated nodes are used as the basis for subsequent rejection reason labels, ensuring that the evidence verification corresponds to the node attributes of the candidate lineage subgraph.

[0122] The set generation unit adds the set of branches with complete evidence to the set of committable branches. For selected branches that do not meet institutional constraints, semantic matching requirements, or have incomplete evidence anchors, rejection reason labels are generated. These rejection reason labels include semantic mismatch, institutional non-compliance, incomplete evidence anchors, time out of bounds, outdated or unreliable sources, and unverifiable conversion. The rejection reason labels are then bound to the corresponding selected branches and their associated nodes, clearly defining the boundaries between the set of committable branches and the set of rejections.

[0123] The path consistency calculation unit performs consistency calculations on the combinable paths within the committable set, and the path consisting of a conjunction and a single choice branch is denoted as... For each Path scores are calculated and compared with a decision threshold to form a sequence of possible paths. In this embodiment, the following path consistency scoring function is used. Perform the calculation:

[0124] ;

[0125] in: Score the path self-consistency; To ensure alignment across time ranges; For unit consistency; For consistency of caliber levels; For data source version temporal consistency; The topological acyclicity determination degree; For the completeness of the anchor points of evidence; To ensure coverage of the organization's regions; For semantic matching degree; This is a topology penalty term; The penalty term is distributed as an evidence anchor point; This is a penalty for exceeding the time limit. For the completeness indicator function of the evidence anchor point; , , , , , , , , , , All are weighting coefficients.

[0126] The comparison and selection unit sorts the path sequence that meets the judgment threshold for path consistency scoring, and determines the path with the highest score as the main path. The path with the second highest score that differs from the main path in its branch selection is determined as the comparison path, ensuring that the comparison path is consistent with the main path in the conjunction stage. The label generation unit binds rejection reason labels to the unselected branches and records the selection criteria on both the main path and the comparison path, ensuring that the attribution selection results maintain consistent explanatory boundaries at both the path and branch levels.

[0127] The main path and the comparison path, along with the set of commitments, are summarized to form the attribution selection result. This enables the subsequent source map structure generation to directly reference the directed link set of the main path and the folded mapping of the comparison path, and to consistently reference the rejection reason labels and evidence anchor point completeness indicators, ensuring that the set of commitments can be verified in terms of structure and constraints.

[0128] S5. Generate a source map structure based on the attribution selection results, and encode the main path of the source map structure as a set of directed links, and the folded comparison path as a comparison unit; bind the rejection reason label with the relevant nodes and directed links, and map the evidence anchor point and the caliber meta-information to the node attributes of the source map structure, and perform consistency verification on the source map structure.

[0129] See Figure 5 In this embodiment, the source map structure is initialized using the attribution selection result and the candidate lineage subgraph as input. An empty set of node set and directed edge set is established, and the construction scope of the source map structure is limited according to the main path and comparison path set in the attribution selection result. The root node in the candidate lineage subgraph and its upstream node committed to in the attribution selection result are used as initial anchor points to determine the nodes and directed edges allowed to be added within this scope, preventing out-of-bounds expansion.

[0130] The main path is encoded, and the dependencies consistent with the main path in the candidate lineage subgraph are constructed into a set of directed links in order from upstream to downstream. Each node is read and registered in the node set, and each new node synchronously writes the directed edge bound to its predecessor to the directed edge set. The directionality of each new directed edge is checked to ensure that the start and end points of the directed edge satisfy the topological order in the candidate lineage subgraph, and that no reverse links pointing upstream are formed. After the main path encoding is completed, all nodes and directed edges involved in the main path are set as backbone markers in the tracing map structure, and the backbone markers are consistently referenced in accordance with the dependency order.

[0131] The comparison paths are folded, compressing the directed link set corresponding to each comparison path in the comparison path set into a single comparison unit, which is then registered in the comparison unit set. For each comparison unit, an association is established with its start and end nodes, allowing it to be referenced as a folded alternative branch in the source map structure, preventing redundant or parallel links from forming on the backbone. Intermediate nodes involved in the folding process are not explicitly registered in the node set, but their attributes are recorded internally within the comparison unit for later review and reference; the mapping between the comparison unit set and its source path is maintained, enabling tracing back to the source map structure. The original link set.

[0132] Rejection reason labels are bound to relevant nodes and directed edges. Based on the rejection reason labels in the attribution selection results, each rejection reason label is associated with a comparison unit in the comparison unit set and the node and directed edge corresponding to the uncommitted selection branch. A field-level binding method is used to mark semantic mismatch, non-compliance with regulations, incomplete evidence anchors, time out of bounds, outdated or unreliable sources, and unverifiable transformations to specific nodes or directed edges, ensuring that rejection reason labels correspond to structural elements and can be directly retrieved in subsequent reviews.

[0133] Evidence anchors are mapped to node attributes, and the evidence anchors in the candidate lineage subgraph are written into the corresponding node attributes in the node set according to the correspondence. During mapping, a reference relationship is established between the evidence anchor and the set of directed links it depends on, so that each evidence anchor can be located to the node and link it supports. The basic fields of the evidence anchors are solidified, including data source identifier, version or timestamp and extraction fingerprint, to ensure that the evidence anchors can be quickly retrieved and compared in subsequent calls.

[0134] The metadata of the standards is mapped to node attributes, and the formulas, dimensions, units, standards level, and sources from the metadata are written into the node attributes involved in the main path and the comparison unit set. For nodes on the main path, units and standards level are aligned first, and then the formula and dimension fields are loaded. The same field loading is performed on the start and end nodes involved in the comparison unit, and the intermediate node attributes of the records within the comparison unit are linked by reference. A bidirectional association is established between the metadata of the standards and the corresponding evidence anchors, so that the source of each metadata of the standards can be supported by the evidence anchors and can be verified.

[0135] After encoding, folding, and mapping, the consistency of the source map structure is checked, focusing on the directionality and connectivity of directed link sets and the completeness of references to comparison units. Directed edges that do not meet the directionality requirements are removed, and their associations with node attributes and comparison units are updated synchronously to prevent dangling references. Nodes that do not meet the connectivity requirements are cleaned up, removing nodes that have lost valid predecessors or successors and their associated directed edges. For comparison units with incomplete references, if a start or end node association is missing, their referential status in the source map structure is paused, and the path is reverted to the original path for completion.

[0136] At the consistency check point, a consistency score for the source map structure is used to make a single judgment and sorting of the source map structure to ensure that the elimination and retention have a unified standard. The expression is as follows:

[0137] ;

[0138] in: Scoring the consistency of the source map structure; and These are the counts of directed edges and nodes, respectively. The topological order in the candidate lineage subgraph; for Indicator that it can be reached from the root node; For comparison of the unit and its starting node With the termination node Association integrity; For folding fidelity; A set of commitments; To trace the origin of map structure Branch selection in the middle The presentation instructions; and These are the consistency of units and the consistency of scope levels, respectively. and These are time range alignment and organizational region coverage, respectively. Counting for suspended directed edges; Count the nodes that lack a rejection reason label; This indicates a missing node evidence anchor point. This is a loopless indicator; , , , , , , , , , All are weighting coefficients.

[0139] A threshold is used to score the consistency of the source map structure in a single assessment. Structural elements scoring below the threshold are removed, triggering attribute updates to maintain the directionality, connectivity, and reference integrity of the source map structure. Breaks in the relationships caused by removal are repaired retrospectively, prioritizing breaks on the main path, followed by breaks related to comparison units. After verification, the source map structure is output, containing the directed link set of the main path and the folded mapping of comparison units. The structure ensures the correspondence between rejection reason labels and evidence anchors in node attributes, enabling subsequent explanatory responses to directly connect the main path's statements, institutional constraints, and evidence anchors, and to reference them consistently with the source map structure.

[0140] S6. Generate explanatory answers based on the source map structure, and connect the main path criteria, institutional constraints and evidence anchors to output the explanatory answers and source map structure;

[0141] In this embodiment, the source map structure is used as input, the directed link set and comparison units of the main path are parsed, each directed edge in the directed link set is read according to the registration order in the source map structure, the main path node order from upstream to downstream is established, each comparison unit in the comparison unit set is paired and referenced according to its start node and end node, the generation path of the explanatory answer is clarified and ensured to be consistent with the structural reference of the source map structure.

[0142] The metadata is organized according to the main path node sequence. At each node in the main path node sequence, the indicator identifier, time range, and organizational area are clearly marked. Formulas, dimensions, units, scope levels, and sources from the metadata are loaded into the corresponding nodes by field, maintaining the correspondence between the metadata and the main path nodes to form a narrative skeleton. Adjacent nodes with unit differences are aligned by unit, and adjacent nodes with scope level differences are aligned by scope level, ensuring field consistency within the same main path segment and preventing field conflicts in the narrative skeleton.

[0143] The institutional constraint parameters are concatenated with the main path nodes. Each node in the main path node sequence is followed by a description of its scope of application, validity period, organizational authority, and data source trust level, and these fields are directly associated with that node. The scope of application field is explained in relation to the organizational region; the validity period field is explained in terms of time range coverage; the organizational authority field is explained in relation to the use cases related to the indicator identifier; and the data source trust level is explained in terms of source level. This ensures that the institutional constraint parameters are concatenated at the node level and can be referenced node by node.

[0144] Evidence anchors are bound to main path nodes, and a minimum completeness standard for evidence anchors is adopted to ensure that each node in the main path node sequence references at least one evidence anchor. When referencing an anchor, the data source identifier, version or timestamp, and extracted fingerprint are inserted into the node's description according to their correspondence, making the source and version of each node clear and verifiable. For nodes with multiple evidence anchors, the node is selected based on the nearest support relationship with the directed link set in the source tracing map structure. Unselected evidence anchors are reserved as backup references in the node attributes to ensure that they can be replaced and supplemented during subsequent consistency checks.

[0145] The comparison unit set is explained, and the rejection reason label corresponding to each comparison unit is paired with its start and end nodes. Using the main path in the committable set as a reference, a judgment threshold is used to select the comparison units to be presented, and the selection criteria are given for each. For semantic mismatch, the alignment difference with the metadata is used as the basis; for non-compliance with regulations, the failure of the field in the regulatory constraint parameter is used as the basis; for incomplete evidence anchors, the missing items of the evidence anchors are used as the basis; for time out of bounds, the difference in the coverage of the time range is used as the basis; for outdated or unreliable sources, the difference in the level of the source and version fields is used as the basis; for unverifiable conversion, the verifiability field of the conversion process is used as the basis. This ensures that the comparison explanation is verifiable at both the node and branch levels.

[0146] The consistency between the explanatory answers and the source map structure is checked. Time range coverage and organizational region inclusion are used to verify the consistency of references to metadata and institutional constraint parameters with the source map structure. References failing the time range coverage check are removed, and the time ranges of the relevant nodes are restated to align with the upstream boundary of the directed link set. References failing the organizational region inclusion check are also removed, and the organizational regions of the relevant nodes are restated to align with the organizational hierarchy of the directed link set. Narrative breakpoints resulting from these removals are rearranged, and the main path nodes are written back in the dependency order registered in the source map structure to ensure that the generated path does not contain reversed or suspended segments.

[0147] The completed main path definition, institutional constraints, and evidence anchors are linked together to generate explanatory answers. Using the main path node sequence as the backbone, the indicator identifiers, time ranges, organizational regions, and definition metadata of each node are explained in parallel. Institutional constraint parameters are explained at the field level, and evidence anchors are explained at the citation level, forming a segmented narrative from upstream to downstream. Selected comparison units are cited in parallel according to their differences from the main path, and rejection reason labels are inserted in corresponding positions, enabling readers to form a clear selection logic between the main path and comparison units. Finally, the explanatory answers and the source map structure are output together, ensuring consistent citation at the node, link, and field levels. This allows subsequent verification to directly use the source map structure to re-verify the time range coverage and organizational region inclusion determinations.

[0148] S7. Generate an audit snapshot based on the attribution selection results, record the issue context package, the source map structure and the attribution selection results, and form a verifiable object;

[0149] In this embodiment, the indicator identifiers, time ranges, organizational regions, and institutional parameters of the issue context package are written into the audit snapshot by field, maintaining consistency with the field order within the issue context package. The main path directed link set and the comparison unit folding mapping of the source map structure are written into the audit snapshot in dependency order. The formulas, dimensions, units, caliber levels, and sources in the node attributes are recorded together, and the evidence anchors and rejection reason labels are bound to the relevant nodes and directed edges according to their correspondence. The main path, comparison path, and branches of the committable set in the attribution selection results are recorded in the audit snapshot, and the rejection reason labels in the attribution selection results are associated with the corresponding nodes and directed edges in the source map structure at the field level, maintaining consistency between the path-level and branch-level selection criteria and structural references.

[0150] A one-time consistency check is performed on the audit snapshot to verify consistency among the time range coverage determination, the organizational area inclusion determination, the directionality and connectivity of the main path, and the completeness indication of the evidence anchor points on the main path nodes, all of which are consistent with the attribution selection results. After the verification is passed, the audit snapshot is solidified as a reviewable object, enabling subsequent reviews to directly reference and verify the issue context package, the attribution map structure, and the attribution selection results item by item based on the audit snapshot.

[0151] Example 2:

[0152] See Figure 6 A dynamic data tracing and question-answering system based on artificial intelligence, which is applied to the method described in Example 1, the system comprising:

[0153] The diagnostic problem parsing module 1 is used to obtain diagnostic problems and parse them to obtain a problem context package; the problem context includes indicator identifiers, time ranges, organizational regions, semantic clues, and institutional parameters;

[0154] Root node determination module 2 is used to determine the root node of the conjunctive-selective indicator source syntax graph in the indicator dictionary based on the problem context package, and extract the caliber metadata and institutional constraint parameters;

[0155] The candidate lineage subgraph generation module 3 is used to expand the conjunctive-selection indicator sourcing syntax graph based on the root node of the conjunctive-selection indicator sourcing syntax graph and the problem context package, retain the conjunctive links and include the selection branches that satisfy the problem context package, mark the evidence anchor points and generate the final candidate lineage subgraph.

[0156] The attribution selection result acquisition module 4 is used to perform attribution selection based on the candidate lineage subgraph and the problem context package. It takes each selected branch as the explanatory hypothesis, matches according to semantic clues and institutional constraint parameters, adds the set that meets the semantic matching and the evidence anchor point to the commit set, generates rejection reason labels for the set that does not meet the requirements, calculates path consistency within the commit set, determines the main path and the comparison path, and obtains the attribution selection result.

[0157] The source map structure generation module 5 is used to generate a source map structure based on the attribution selection results, and encode the main path of the source map structure as a set of directed links and the folded comparison path as a comparison unit; it binds the rejection reason labels with relevant nodes and directed links, maps the evidence anchor points and caliber meta-information to the node attributes of the source map structure, and performs consistency verification on the source map structure.

[0158] The explanatory answer generation module 6 is used to generate explanatory answers based on the source map structure, and connect the main path caliber, institutional constraints and evidence anchors to output the explanatory answer and the source map structure.

[0159] The audit snapshot generation module 7 is used to generate audit snapshots based on the attribution selection results, record the issue context package, the source map structure and the attribution selection results, and form a verifiable object.

[0160] Furthermore, the specific implementation steps of the diagnostic problem parsing module 1, root node determination module 2, candidate lineage subgraph generation module 3, attribution selection result acquisition module 4, source map structure generation module 5, explanatory answer generation module 6, and audit snapshot generation module 7 can be found in the corresponding description in Example 1, and will not be repeated here.

[0161] Example 3:

[0162] See Figure 7A dynamic data tracing and question-answering device based on artificial intelligence, the device including a processor 8 and a memory 9;

[0163] The memory 9 is used to store computer program code 91 and to transmit the computer program code 91 to the processor 8;

[0164] The processor 8 is used to execute the AI-based dynamic data tracing and question-answering method described in Embodiment 1 according to the instructions in the computer program code 91.

[0165] This embodiment also includes a computer-readable storage medium storing computer-executable instructions. When the computer-executable instructions are executed on a computer, the dynamic data tracing and question-answering method based on artificial intelligence described in Embodiment 1 is implemented.

[0166] Generally, the computer instructions for implementing the method of the present invention can be carried on any combination of one or more computer-readable storage media. Non-transitory computer-readable storage media can include any computer-readable medium except for the signal itself, which is temporarily propagating.

[0167] Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EKROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0168] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smarttalk, and C++—as well as conventional procedural programming languages—such as the "C" language or similar programming languages. In particular, Python, suitable for neural network computation, and platform frameworks such as TensorFlow and PyTorch can be used. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer or to an external computer (e.g., via the Internet using an Internet service provider) through any type of network, including a local area network (LAN) or a wide area network (WAN).

[0169] The aforementioned devices and non-transitory computer-readable storage media can be found in the detailed description of an AI-based dynamic data tracing question-and-answer method and its beneficial effects, which will not be repeated here.

[0170] Although embodiments of the present invention have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A dynamic data tracing and question-answering method based on artificial intelligence, characterized in that, include: Obtain the diagnostic problem and parse the resulting problem context packet; The context of the problem includes indicator identifiers, time ranges, organizational regions, semantic clues, and institutional parameters; Based on the problem context package, the root node of the conjunctive-selective indicator source syntax graph is determined in the indicator dictionary, and the caliber metadata and institutional constraint parameters are extracted. Based on the root node of the conjunction-selection indicator sourcing syntax graph and the problem context package, expand the conjunction-selection indicator sourcing syntax graph, retain the conjunction steps and include the selection branches that satisfy the problem context package, mark the evidence anchor points and generate the final candidate lineage subgraph. Attribution selection is performed based on candidate lineage subgraphs and problem context packages. Each selected branch is used as an explanatory hypothesis. Matching is performed based on semantic clues and institutional constraint parameters. Sets that meet semantic matching and have complete evidence anchors are added to the committable set. Sets that do not meet the requirements are used to generate rejection reason labels. Path consistency is calculated within the committable set to determine the main path and comparison path, and attribution selection results are obtained. The source map structure is generated based on the attribution selection results, and the main path of the source map structure is encoded as a set of directed links, and the folded comparison path is a comparison unit. The rejection reason labels are bound to the relevant nodes and directed links, and the evidence anchors and caliber meta-information are mapped to the node attributes of the source map structure. The consistency of the source map structure is checked. Explanatory answers are generated based on the source map structure, and the main path criteria, institutional constraints and evidence anchors are linked together to output the explanatory answers and the source map structure. An audit snapshot is generated based on the attribution selection results, recording the issue context package, the source map structure, and the attribution selection results to form a verifiable object.

2. The dynamic data tracing and question-answering method based on artificial intelligence according to claim 1, characterized in that: The process involves acquiring and parsing the diagnostic problem to obtain a problem context package. This problem context includes indicator identifiers, time ranges, organizational regions, semantic clues, and institutional parameters, specifically: Taking the diagnostic question as input, semantic cues are identified and text fragments directly related to the diagnostic question are extracted as the units to be parsed. Based on the unit to be parsed, the indicator identifier, the standardized expression time range, and the normalized mapping organization region are determined to form the initial result of indicator identifier, time range, and organization region; Based on semantic clues and initial results, institutional parameters are extracted and constraints are identified. The consistency between institutional parameters and indicator identifiers, time ranges, and organizational regions is checked to obtain the problem context package.

3. The dynamic data tracing and question-answering method based on artificial intelligence according to claim 1, characterized in that: The method of determining the root node of the conjunctive-selective indicator source graph in the indicator dictionary based on the problem context package, and extracting caliber metadata and institutional constraint parameters, specifically includes: Using the problem context package as input, the indicator dictionary is searched, the indicator identifier is used for unique positioning, and the validity period and scope of application are filtered by time range and organizational region to obtain a set of candidate indicator entries; The institutional parameters of the candidate indicator item set are matched, and the corresponding fields of the institutional parameters are compared with the institutional constraint parameters in the candidate indicator item set to determine the subset of candidate indicator items that meet the institutional constraints. The candidate indicator items subset is mapped to the conjunctive-selective indicator traceability syntax graph, and the root node of the conjunctive-selective indicator traceability syntax graph is located based on the matching of indicator identifier and scope of application. If there are multiple matches, the system constraint parameters are used as the priority for judgment, and then the time range coverage is used to determine the root node. Using the root node of the conjunctive-selective indicator source grammar graph as the anchor point, extract the caliber metadata bound to the root node from the indicator dictionary, and load the formula, dimension, unit, caliber level and source into the caliber metadata by field; The institutional constraint parameters bound to the root node in the indicator dictionary are extracted, and consistency checks are performed using the scope of application field, validity period field, and organization permission field. The parameters are then compared item by item with the institutional parameters in the issue context package to obtain the institutional constraint parameters.

4. The dynamic data tracing and question-answering method based on artificial intelligence according to claim 1, characterized in that: The generation of the final candidate lineage subgraph specifically includes: Using the root node of the conjunction-selection indicator source syntax graph and the problem context package as input, the conjunction-selection indicator source syntax graph is expanded layer by layer, and the expansion boundary is limited to the time range and organization region in the problem context package to establish the upstream dependency set starting from the root node. Identify the conjunctive elements in the upstream dependency set, include the upstream nodes and directed edges corresponding to the conjunctive elements into the initial candidate lineage subgraph, and maintain the dependency order consistent with the order of the conjunctive elements. The selection branches in the upstream dependency set are filtered by institutional parameters. The institutional parameters in the problem context package are compared with the institutional constraint parameters bound to the selection branches one by one to obtain the set of selection branches that comply with the regulations. Semantic matching is performed on the set of branches for compliance with regulations. The semantic clues in the problem context package are matched with the caliber metadata bound to the branches to determine the branches that meet the semantic matching, and the branches that meet the semantic matching are included in the initial candidate lineage subgraph. The nodes included in the initial candidate lineage subgraph are checked for consistency in time range and consistency in tissue region. Time range coverage and tissue region inclusion are used for checking, and nodes and directed edges that do not meet the consistency requirements are removed. Mark evidence anchors at the node level of the initial candidate lineage subgraph. Based on the minimum completeness of evidence anchors, bind at least one evidence anchor to each conjunctive link and each included selection branch, and map the evidence anchors to the corresponding nodes. The initial candidate lineage subgraph is subjected to topological acyclicity checks and caliber-level consistency checks. Nodes and directed edges that do not meet the checks are removed until the checks are met, thus obtaining the final candidate lineage subgraph.

5. The dynamic data tracing and question-answering method based on artificial intelligence according to claim 1, characterized in that: The process of obtaining the attribution selection result specifically includes: Using the semantic clues and institutional constraint parameters in the candidate lineage subgraph and the problem context package as input, an explanatory hypothesis set is established for each selected branch in the candidate lineage subgraph and a list of to-be-determined branches is generated. The list of judgments is subjected to institutional constraint judgments. The institutional constraint parameters are compared with the institutional constraint parameters bound to the selected branches one by one to filter out the set of institutional compliant branches. Semantic matching is performed on the set of regulatory compliance branches, and semantic clues are matched with the metadata of the scope of the selected branches to obtain a set of semantically matched branches; The completeness of the evidence anchor points is verified on the semantic matching branch set. The minimum completeness of the evidence anchor points is used as the standard to verify whether the evidence anchor points of each conjunctive link and the included selection branches are complete, and the complete evidence branch set is determined. The set of branches with complete evidence is added to the set of branches that can be committed to. For branches that do not meet the institutional constraints, semantic matching, or incomplete evidence anchors, rejection reason labels are generated. The rejection reason labels include semantic mismatch, non-compliance with the system, incomplete evidence anchors, time out of bounds, outdated or unreliable source, and unverifiable conversion. Path consistency is calculated within the committable set, and time range coverage alignment, unit consistency, caliber hierarchy consistency, data source version time sequence consistency, and topological acyclicity are used as check items to determine the main path and comparison path. The main path and the comparison path, along with the set of commitments, are summarized to obtain the attribution selection results.

6. The dynamic data tracing and question-answering method based on artificial intelligence according to claim 1, characterized in that: The generated source tracing map structure specifically includes: Using the attribution selection results and candidate lineage subgraphs as input, initialize the node set and directed edge set of the source map structure, and limit the construction scope of the source map structure according to the main path and comparison path in the attribution selection results; The main path is encoded, and a set of directed links from upstream to downstream is constructed based on the dependencies in the candidate lineage subgraph. The nodes and directed edges involved in the main path are added to the source map structure while maintaining the dependency order. The comparison path is folded, compressing the set of directed links corresponding to each comparison path into a single comparison unit, while maintaining the association between the comparison unit and its start and end nodes. The rejection reason labels are bound to the relevant nodes and directed edges. Based on the rejection reason labels in the attribution selection results, the rejection reason labels are associated with the comparison units and the nodes and directed edges corresponding to the uncommitted selection branches, respectively. The evidence anchors are mapped to node attributes, and the evidence anchors in the candidate lineage subgraphs are written into the corresponding node attributes in the tracing map structure according to the correspondence, while maintaining the reference relationship between the evidence anchors and the directed link set. Map the caliber metadata to node attributes, write the formula, dimension, unit, caliber level and source into the node attributes involved in the main path and comparison unit, and establish the association between the caliber metadata and the corresponding evidence anchor points; The consistency of the traceability map structure is checked, including the directionality and connectivity of the directed link set and the integrity of the reference of the comparison unit. Directed edges that fail the check are removed, and the association relationship with node attributes and comparison units is updated synchronously.

7. The dynamic data tracing and question-answering method based on artificial intelligence according to claim 1, characterized in that: The output explanatory answers and source map structure specifically include: Using the source map structure as input, the directed link set and comparison unit of the main path are parsed to establish the main path node order and corresponding reference order; The main path node sequence is adopted to organize the caliber metadata according to the main path node sequence, clarify the indicator identifier, time range and organization area, and make the caliber metadata correspond to the main path nodes; The institutional constraint parameters are linked to the main path nodes, and described by fields such as scope of application, validity period, organizational authority and data source trust level, so that the institutional constraint parameters after each main path node are directly related to that node. The evidence anchors are bound to the main path nodes for reference. The minimum completeness of the evidence anchors is used as the standard to ensure that each main path node references at least one evidence anchor. The data source identifier, version or timestamp and the extracted fingerprint are inserted into the node description according to the corresponding relationship. The comparison units are explained, and the rejection reason tags corresponding to the comparison units are paired with the start and end nodes of the comparison units. The main path in the set of commitments is used as a reference to select the comparison units to be presented and the basis for selection is given one by one. The consistency between the explanatory answers and the source map structure is checked. The time range coverage judgment and the organizational area inclusion judgment are used to check whether the reference of the caliber meta-information and institutional constraint parameters are consistent with the source map structure. Inconsistent references are removed and the order of the main path nodes is rearranged. The main path criteria, institutional constraints, and evidence anchors are linked together to generate explanatory answers, and the explanatory answers and source map structure are output.

8. The dynamic data tracing and question-answering method based on artificial intelligence according to claim 1, characterized in that: The generation of the audit snapshot specifically includes: Write the main path directed link set and comparison unit folding mapping of the source tracing map structure into the audit snapshot in the order of dependency, record the formula, dimension, unit, caliber level and source in the node attributes, and bind the evidence anchor point and rejection reason label to the relevant node and directed edge according to the corresponding relationship; Record the main path, comparison path, and branches of the committable set in the attribution selection results to the audit snapshot, and associate the rejection reason labels in the attribution selection results with the corresponding nodes and directed edges in the source map structure; A one-time consistency check is performed on the audit snapshot; the check includes: checking whether the time range coverage determination and the organizational area inclusion determination are consistent, checking whether the directionality and connectivity of the main path are consistent with the source map structure, and checking whether the completeness indication of the evidence anchor point on the main path node is consistent with the attribution selection result; if all are consistent, the check is passed, and the audit snapshot is solidified as a re-verifiable object.

9. A dynamic data tracing and question-answering system based on artificial intelligence, characterized in that, The system is applied to the method according to any one of claims 1-8, the system comprising: The diagnostic problem parsing module (1) is used to obtain diagnostic problems and parse them to obtain problem context packages; the problem context includes indicator identifiers, time ranges, organizational regions, semantic clues, and institutional parameters; The root node determination module (2) is used to determine the root node of the conjunctive-selective indicator tracing syntax graph in the indicator dictionary based on the problem context package, and extract the caliber metadata and institutional constraint parameters; The candidate lineage subgraph generation module (3) is used to expand the conjunctive-selection indicator tracing syntax graph based on the root node of the conjunctive-selection indicator tracing syntax graph and the problem context package, retain the conjunctive link and include the selection branch that satisfies the problem context package, mark the evidence anchor point and generate the final candidate lineage subgraph. The attribution selection result acquisition module (4) is used to perform attribution selection based on the candidate lineage subgraph and the problem context package. It takes each selected branch as the explanatory hypothesis, matches according to semantic clues and institutional constraint parameters, adds the set that meets the semantic matching and the evidence anchor point to the commit set, generates rejection reason labels for the set that does not meet the requirements, calculates path consistency within the commit set, determines the main path and the comparison path, and obtains the attribution selection result. The source map structure generation module (5) is used to generate the source map structure based on the attribution selection result, and encode the main path of the source map structure as a set of directed links and the folded comparison path as a comparison unit; bind the rejection reason label with the relevant nodes and directed links, map the evidence anchor point and the caliber meta information to the node attributes of the source map structure, and perform consistency verification on the source map structure. The explanatory answer generation module (6) is used to generate explanatory answers based on the source map structure, and connect the main path caliber, institutional constraints and evidence anchors to output the explanatory answer and the source map structure; The audit snapshot generation module (7) is used to generate an audit snapshot based on the attribution selection results, record the issue context package, the source map structure and the attribution selection results, and form a verifiable object.

10. A dynamic data tracing and question-answering device based on artificial intelligence, characterized in that: The device includes a processor (8) and a memory (9); The memory (9) is used to store computer program code (91) and to transmit the computer program code (91) to the processor (8). The processor (8) is used to execute the dynamic data tracing and question-answering method based on artificial intelligence as described in any one of claims 1-8 according to the instructions in the computer program code (91).

Images