Multi-party data-based challenge point analysis method and device, and related equipment

Abstract

This invention discloses a method, apparatus, and related equipment for analyzing objections based on multi-party data. The method constructs a multi-dimensional evaluation decision tree comprising a data type judgment layer, an evaluation dimension judgment layer, a template trigger condition judgment layer, and an objection template leaf node layer; it reads structured context information to extract the position tags of the analysis subject, and based on these position tags, splits the input dataset into three attribution subsets: self-side, opposing side, and public, generating a data classification list; it catalogs the self-side data to generate a self-side data catalog; it traverses the multi-dimensional evaluation decision tree to generate predicted objections from opposing perspectives and calculates confidence scores; it automatically determines whether to execute a reverse analysis branch based on the existence of opposing side data; and it assembles the results of each stage into a structured analysis document. This invention can capture the interaction effect between data attribution and evaluation strategies, achieving a systematic and automatic evolution of the analysis scheme, and improving the analysis coverage and relevance.

Description

Technical Field

[0001] The embodiments of the present invention relate to the field of artificial intelligence technology, and in particular to a method, apparatus and related equipment for analyzing points of contention based on multi-party data. Background Technology

[0002] In data analysis scenarios involving multiple stakeholders, such as business competition analysis, legal evidence review, and policy impact assessment, the quality of data analysis conclusions depends on decisions at two levels: the first is the data attribution and organization level, i.e., from which stakeholder's perspective the data is examined, how to split mixed datasets into different attribution subsets according to source and purpose, and how to identify data types and extract core attributes; the second is the analysis and evaluation level, i.e., which evaluation dimensions are used, what questions are generated for each data point, at what granularity the evaluation results are output, and whether to perform reverse analysis on the other party's data. As the demand for data analysis continues to grow, users are demanding higher levels of automation and higher quality analysis conclusions.

[0003] Existing data analysis methods typically treat data attribution optimization and evaluation strategy optimization as independent problems, making it difficult to capture the interaction between the two layers. The overall analytical effect of an optimal data attribution splitting scheme paired with a suboptimal combination of evaluation dimensions may be inferior to that of a suboptimal attribution splitting scheme paired with an optimal combination of evaluation dimensions. If a sophisticated multidimensional evaluation framework is applied to a mixed dataset without attribution splitting, the output of challenges cannot distinguish between weaknesses in one's own data and weaknesses in the opponent's data, resulting in a lack of specificity in the analytical conclusions. Conversely, accurate data attribution splitting paired with only a single evaluation dimension cannot cover challenges that the opponent may raise from multiple angles, resulting in a systemic deficiency in analytical coverage. Summary of the Invention

[0004] The present invention provides a method, apparatus and related equipment for analyzing objections based on multi-party data, which aims to solve the technical problem that it is difficult to accurately determine the objections of stakeholders in data analysis involving multiple stakeholders.

[0005] In a first aspect, embodiments of the present invention provide a method for analyzing points of contention based on multi-party data, comprising: A multidimensional evaluation decision tree is constructed, which includes a data type judgment layer, an evaluation dimension judgment layer, a template trigger condition judgment layer, and a questioning template leaf node layer. The data type judgment layer performs a first-level branch based on the data type label, the evaluation dimension judgment layer performs a second-level branch based on the applicable conditions of the evaluation dimension, the template trigger condition judgment layer performs a third-level branch based on the trigger conditions of the questioning template, and the questioning template leaf node layer stores the questioning template configuration. Read structured context information and extract the stance labels of the current analysis subject; Based on the aforementioned position labels, the input dataset is split into a subset belonging to one's own side, a subset belonging to the other side, and a common subset belonging to the other side, resulting in a data classification list; Perform a cataloging operation on the data in the subset to which your own data belongs, identify data type tags and extract core attributes and supporting conclusion identifiers to obtain your own data catalog; The feature information of each data item in the self-owned data catalog is used as input. Starting from the root node of the multidimensional evaluation decision tree, the tree is traversed layer by layer. After data type judgment, evaluation dimension applicability judgment and template trigger condition judgment, the activated questioning template leaf node is reached, predicting questioning points are generated, and confidence scores are calculated for each predicting questioning point. Detect whether there are data records with the other party's belonging label in the data classification list, and determine whether to execute the reverse analysis branch based on the detection result; The results from each stage are summarized and assembled into a structured analysis document.

[0006] Secondly, embodiments of the present invention provide a device for analyzing points of contention based on multi-party data, comprising: The construction module is used to construct a multi-dimensional evaluation decision tree. The multi-dimensional evaluation decision tree includes a data type judgment layer, an evaluation dimension judgment layer, a template trigger condition judgment layer, and a questioning template leaf node layer. The data type judgment layer performs a first-level branch based on the data type label, the evaluation dimension judgment layer performs a second-level branch based on the applicable conditions of the evaluation dimension, the template trigger condition judgment layer performs a third-level branch based on the trigger conditions of the questioning template, and the questioning template leaf node layer stores the questioning template configuration. The extraction module is used to read structured context information and extract the stance tags of the current analysis subject; The splitting module is used to split the input dataset into a subset belonging to one's own side, a subset belonging to the other side, and a common subset belonging to the other side, based on the position labels, to obtain a data classification list; The identification module is used to perform cataloging operations on the data in the subset to which the party belongs, identify data type tags and extract core attributes and supporting conclusion identifiers to obtain the data catalog of the party. The scoring module is used to take the feature information of each data in the self-owned data catalog as input, traverse layer by layer from the root node of the multidimensional evaluation decision tree, and after data type judgment, evaluation dimension applicability judgment and template trigger condition judgment, reach the activated questioning template leaf node, generate predicted questioning points and calculate the confidence score for each predicted questioning point. The detection module is used to detect whether there are data records with the other party's attribution tag in the data classification list, and to determine whether to execute the reverse analysis branch based on the detection results; The assembly module is used to summarize the results of each stage and assemble them into a structured analysis document.

[0007] Thirdly, embodiments of the present invention provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the questioning analysis method based on multi-party data described in the first aspect.

[0008] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to perform the questioning analysis method based on multi-party data described in the first aspect.

[0009] This invention provides a method, apparatus, and related equipment for analyzing objections based on multi-party data. The method constructs a multi-dimensional evaluation decision tree; reads structured context information and extracts the position tags of the current analysis subject; based on the position tags, the input dataset is split into a subset of one's own attribution, a subset of the opposing party's attribution, and a subset of common attribution, resulting in a data classification list; cataloging is performed on the data in the subset of one's own attribution, identifying data type tags and extracting core attributes and supporting conclusion identifiers, resulting in a data catalog of one's own party; the feature information of each data entry in the data catalog of one's own party is used as input, and the tree is traversed layer by layer from the root node of the multi-dimensional evaluation decision tree, undergoing data type judgment, evaluation dimension applicability judgment, and template trigger condition judgment, until the activated objection template leaf node is reached, generating predicted objections and calculating a confidence score for each predicted objection; the existence of data records with opposing party attribution tags in the data classification list is detected, and the execution of a reverse analysis branch is determined based on the detection results; the results produced at each stage are summarized and assembled into a structured analysis document. This method can capture the interaction effect between data attribution and evaluation strategies, achieving systematic and automatic evolution of the analysis scheme, and improving the analysis coverage and relevance. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating an embodiment of the dispute analysis method based on multi-party data provided by the present invention. Figure 2 This is a schematic block diagram of a dispute analysis device based on multi-party data provided in an embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0014] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0015] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0016] Please see Figure 1 This is a flowchart illustrating a method for analyzing points of contention based on multi-party data, provided in an embodiment of the present invention. The method includes steps S110 to S150.

[0017] Step S110: Construct a multidimensional evaluation decision tree; In this embodiment, the multidimensional evaluation decision tree includes a data type judgment layer, an evaluation dimension judgment layer, a template triggering condition judgment layer, and a questioning template leaf node layer. The first level is the data type judgment layer, serving as the root node area of ​​the decision tree. The root node receives feature information of a piece of data to be evaluated and branches at the first level based on the data type label. Data type labels include, but are not limited to, document data, log record data, image data, and testimony statement data. Each data type label corresponds to a first-level branch node. The second level is the evaluation dimension judgment layer. Under each first-level branch node, further branches are formed based on the applicable conditions of the evaluation dimension. An evaluation dimension refers to an independent analytical perspective used to evaluate data quality or credibility. Each dimension focuses on a specific quality aspect of the data. Different dimensions are logically independent but mutually corroborative in their evaluation conclusions. When a data feature does not meet the applicable conditions of a certain evaluation dimension, that second-level branch is pruned and skipped, and does not proceed to subsequent levels. Evaluation dimensions in this embodiment include, but are not limited to, source reliability dimension, content consistency dimension, and logical relevance dimension. The source reliability dimension is used to assess whether the source of the data is verifiable, whether there is a possibility of tampering, and whether the data source party has vested interests. The content consistency dimension assesses whether the data is internally consistent, whether there are contradictions between the data and other data in the same group, and whether there are inconsistencies between different versions of the data. The logical relevance dimension assesses whether the logical chain between the data and the conclusions it supports is complete, and whether there are logical jumps or causal reversals. The third level is the template trigger condition judgment layer. Under each second-level branch node, further branches are formed based on the trigger conditions of the challenge template. Each trigger condition corresponds to a third-level branch node, which checks whether the core attributes of the data to be evaluated meet the activation conditions of the challenge template. Third-level branches that do not meet the trigger conditions are pruned and skipped. The fourth level is the challenge template leaf node layer. Each third-level branch that passes the trigger condition judgment finally reaches the leaf node. The leaf node stores the specific challenge template configuration, including the unique identifier of the challenge template, the identifier of the evaluation dimension to which it belongs, the challenge expression mode, and the template specificity level. The challenge expression mode defines the standard sentence structure of the challenge and includes placeholder variables that can be filled. The template specificity level is used for subsequent confidence calculation, indicating whether the challenge generated by the template is a general challenge or a precise challenge targeting specific data features. The hierarchical path of the decision tree thus encodes a complete chain of reasoning: for a piece of data, first determine its type (first level), then determine which evaluation dimensions are applicable (second level), then determine which template triggering conditions are met (third level), and finally reach the activated questioning template (fourth level).

[0018] In this embodiment, a supplementary evaluation rule subtree specific to each data type label is constructed. The internal structure of the supplementary evaluation rule subtree is the same as the second to fourth level structure of the decision tree—that is, the supplementary rule subtree also consists of evaluation dimension judgment nodes, template trigger condition judgment nodes, and challenge template leaf nodes, but the challenge templates capture the challengeable points specific to this data type. For example, the supplementary rule subtree attached to the document type branch node includes challenge template leaf nodes related to document signature verifiability, document creation timestamp integrity, and document metadata consistency; the supplementary rule subtree attached to the image type branch node includes challenge template leaf nodes related to image metadata integrity, image editing traces, and image source chain integrity; the supplementary rule subtree attached to the log type branch node includes challenge template leaf nodes related to log record continuity, log entry timestamp rationality, and log integrity verification. The supplementary rule subtree is attached in parallel with the existing general evaluation dimension branches under this data type branch, so that the general challenge template and the supplementary challenge template are accessed uniformly within the same traversal path during the decision tree traversal, without the need for additional loading and merging operations. Decision trees support dynamic expansion. When a new data type label is added to the system, only the corresponding supplementary rule subtree needs to be built and attached to the newly added first-level branch node. Through the data type-aware supplementary rule subtree attachment mechanism, general challenge templates and type-specific challenge templates can be accessed uniformly within the same traversal path without additional loading and merging operations. This improves analysis coverage while ensuring execution efficiency, and the decision tree's support for dynamic expansion enhances the system's adaptability.

[0019] In one embodiment, existing evaluation dimensions, challenge templates, and data type supplementary rules in the system are assembled into a complete multidimensional evaluation decision tree according to a defined hierarchical structure and then standardized for registration. Each decision tree corresponds to a record in the registration database, and the record includes at least: a unique identifier for the decision tree, an applicable task scenario type identifier, a complete hierarchical structure description of the decision tree (including the number of nodes at each level, branch conditions, and leaf node template configuration), a list of associated supplementary rule subtree identifiers, the decision tree creation time, the decision tree source identifier (manually created or automatically generated by the system), and the current status of the decision tree (candidate, in use, historical, obsolete). The registration database is managed in groups according to task scenario type. Multiple decision trees can be registered under the same task scenario type, forming a candidate decision tree pool for that scenario. The registration database supports version management; each new decision tree generated from decision tree optimization is registered as a new version, while the original decision tree is downgraded to a historical version but not deleted, preserving the complete evolutionary trajectory. Through version management of the registry and grouping management by task scenario type, the system has the ability to continuously improve analysis quality and execution efficiency as analysis tasks accumulate. The evaluation framework evolves continuously with version iterations, and high-quality evaluation dimensions, questioning templates, and data type supplementary rules can be systematically identified, registered, and reused.

[0020] Step S120: Read the structured context information and extract the position tags of the current analysis subject; In this embodiment, structured context information is read, and fields related to the stakeholder roles of the analysis subject are parsed to extract the analysis subject's position tags. Structured context information refers to metadata organized in key-value pairs or structured document format, containing at least one field identifying the analysis subject's role in a multi-party relationship. The priority for extracting position tags is as follows: first, check if an explicit position identifier field exists in the structured context information; if so, extract the field value as the position tag; if the field does not exist but a field value matching a known stakeholder name exists, infer the position tag through name matching; if neither of these conditions is met, a pre-configured default position tag is used. Position tags include, but are not limited to, first stakeholder, second stakeholder, and neutral party.

[0021] In one embodiment, position label extraction employs a three-tiered priority strategy. First, it checks if an explicit position identifier field, such as "role" or "stakeholder_role," exists in the structured context information. If it does, the field value is directly extracted as the position label. If no such explicit field exists, but a field value matching a known stakeholder name is detected in the context information, the position label is inferred through name matching—for example, if the context contains the field value "Analysis Subject: Company A," and "Company A" is registered as a first stakeholder in the stakeholder registry, the position label is inferred to be the first stakeholder. If neither of these methods satisfies the criteria, a pre-configured default position label is used. The default position label can be set to first stakeholder or neutral party based on system configuration. This multi-tiered priority strategy ensures that reasonable position labels can be extracted regardless of the completeness of the context information, improving the system's robustness.

[0022] Step S130: Based on the position labels, split the input dataset into a subset belonging to one's own side, a subset belonging to the other side, and a common subset belonging to the other side, to obtain a data classification list; In this embodiment, the input dataset is split into three attribution subsets—own, opposing, and common—based on position labels. Each data record in the input dataset is traversed, and attribution determination is performed on each record. If the source identifier or content of the data explicitly supports the stakeholder corresponding to the position label, the data is assigned the own attribution label (OWN). If the source identifier or content of the data explicitly supports the stakeholder opposed to the position label, the data is assigned the opposing attribution label (OPPONENT). If the source identifier or content of the data is shared by both parties, provided by a neutral party, or the attribution cannot be determined, the data is assigned the common attribution label (SHARED). "Explicitly supports" means that the name of the data's source matches the name of the corresponding stakeholder, or that the data content contains a conclusive statement favorable to that stakeholder. Attribution determination is primarily based on the source identifier; data content is analyzed only when the source identifier cannot determine the attribution. The attribution splitting results are written to a data classification list file. The data classification list is a structured file, and each record contains a data identifier, attribution label, and allocation basis. The allocation basis records the rules on which the attribution determination was based, such as "source matching" or "content analysis," for audit traceability. The data classification list file is written to the shared workspace in the form of a declaration file, serving as an input declaration for subsequent steps. Each subsequent step obtains data ownership information by reading this declaration file, rather than by directly transferring data to memory.

[0023] In one embodiment, the allocation of attribution tags employs a two-tiered judgment strategy, prioritizing source identification and supplemented by content analysis. First, the source identification of the data is checked—it identifies the data's producer or provider. If the source name matches the name of a stakeholder corresponding to the position tag, the data is assigned an "Own Attribution" tag (OWN). If the source name matches the name of a stakeholder opposing the position tag, the data is assigned an "Opposite Attribution" tag (OPPONENT). When the source identification cannot determine attribution, the data content is analyzed—it checks whether the data content contains conclusive statements favorable to a particular stakeholder. If so, the corresponding attribution tag is assigned to the data. If the source identification or content is shared by both parties, provided by a neutral party, or cannot be determined based on existing information, the data is assigned a "Shared Attribution" tag (SHARED). The allocation results of the attribution tags are recorded based on the allocation criteria (such as "source matching" or "content analysis") for audit traceability. This two-tiered judgment strategy ensures both the accuracy of attribution determination and the ability to handle incomplete source information.

[0024] Step S140: Perform a cataloging operation on the data in the subset to which the data belongs, identify data type tags and extract core attributes and supporting conclusion identifiers to obtain the data catalog of the data belonging to the data owner; In this embodiment, a cataloging operation is performed on the data within the self-owned subset, including identifying data type tags and extracting core attributes and supporting conclusion identifiers. All data entries with the OWN tag in the data classification list are traversed, and data type identification is performed on each data entry based on the following signals: whether the data record contains an explicit type label field, matching the format characteristics of the data content (such as JSON structure, natural language text, binary encoding indication, etc.), and the system type of the data source (such as a document management system, log collection system, image storage system). The identification result assigns a data type tag to each data entry. If the data type cannot be determined, a default type tag is assigned. Core attributes include, but are not limited to, the name of the data source, the data generation timestamp, and a data content summary. The data content summary is automatically generated by extracting key sentences or key field values ​​from the data content. Supporting conclusion identifiers identify the conclusive statements supported by the data content, extracting a unique identifier for the conclusion, used for generating subsequent points of contention in the logical relevance dimension. The cataloging results are written to a structured self-owned data directory file, which is written to the shared workspace in the form of a declaration file.

[0025] In one embodiment, data type identification is based on a comprehensive judgment of multiple signals. First, it checks whether the data record contains an explicit type label field, such as "type" or "data_type". If it exists, the field value is directly used as the data type label. If no explicit label exists, the data content's format features are matched—checking whether the data content has a JSON structure (key-value pair format), natural language text features, binary encoding indications, etc., to infer the data type through these format features. If neither of the first two signals can determine the type, the system type from which the data originates is checked—if the data originates from a document management system, it is identified as document data; if it originates from a log collection system, it is identified as log record data; if it originates from an image storage system, it is identified as image data. If none of the above signals can determine the data type, a default type label is assigned. Core attribute extraction includes the data source name, data generation timestamp, and data content summary. The data content summary is automatically generated by extracting key sentences or key field values ​​from the data content; for example, extracting sentences containing key information from natural language text, and extracting key field values ​​from structured data. Supporting conclusion identifiers identify the conclusive statements supported by the data content, extract a unique identifier for the conclusion, and use it to generate subsequent challenges in the logical relevance dimension—when the evaluation framework detects a logical jump between the data and the conclusion it claims to support, it generates corresponding challenges.

[0026] Step S150: Take the feature information of each data in the self-owned data directory as input, traverse layer by layer from the root node of the multidimensional evaluation decision tree, and after data type judgment, evaluation dimension applicability judgment and template trigger condition judgment, reach the activated questioning template leaf node, generate predicted questioning points and calculate the confidence score for each predicted questioning point. In this embodiment, the feature information of each data entry in the self-owned data directory is used as input, and the decision tree is traversed layer by layer starting from the root node. For each data entry in the self-owned data directory, its data type label is read as the first-level branch selection condition, and the corresponding data type branch node is entered. After entering the branch node, all substructures under that branch enter the traversal state. Under the selected data type branch node, all second-level evaluation dimension judgment nodes are traversed, and it is checked whether the applicable conditions carried by the node match the core attribute features of the current data. If they match, the next level of the evaluation dimension branch is entered; if they do not match, a pruning operation is performed on the branch, and the subsequent level is not entered. The pruning operation ensures that each data entry is traversed only along the branch path applicable to its data features, avoiding reaching irrelevant questioning template leaf nodes. For evaluation dimension branches that pass the applicability judgment, the third-level template trigger condition judgment nodes under them are traversed. When the core attribute or data type label of the current data meets the trigger condition, the fourth-level questioning template leaf node is reached along the branch. The system reads the challenge template configuration stored in the leaf nodes, fills the placeholder variables of the template with the specific attribute values ​​of the data, and generates semantically complete and targeted predicted challenge text. Each reached leaf node generates at least one predicted challenge. When there are multiple optional fill values ​​in the template, multiple different challenge points can be generated to cover different challenge angles. The system records the complete decision tree path of the challenge point—the complete node sequence from the root node through the data type branch nodes, evaluation dimension judgment nodes, template trigger condition judgment nodes to the leaf node—as the reasoning and tracing basis for the challenge point.

[0027] In this embodiment, the confidence score is calculated based on a weighted combination of at least one of the following factors: the matching strength (M1) between the evaluation dimension and the data feature, which measures the degree to which the evaluation dimension covers the data feature. The higher the matching strength, the more valuable the generated question points are for practical analysis. The value of M1 ranges from 0.0 to 1.0, and is calculated based on the matching depth and coverage of the data at the evaluation dimension's judgment node in the decision tree path. When the data feature meets all the sub-conditions in the applicable conditions of the dimension, M1 approaches 1.0; when only some sub-conditions are met, M1 is set to maximum. The calculation includes: proportional reduction; the specificity of the challenge template (M2), which measures the accuracy of the template that generated the challenge; the fewer placeholder variables and the more specific the semantics of the template, the higher the M2 value, ranging from 0.0 to 1.0; and the strength of the correlation signal supporting the challenge in the data content (M3), which detects whether there are correlation signals in the data content that support the challenge. For example, if the data content contains vague descriptions or citations without source attribution, the challenge related to source reliability will receive a higher M3 value, ranging from 0.0 to 1.0. The confidence score C is calculated using a weighted sum: , in , , For the preset weighting coefficients, and In this implementation, the default weight configuration is as follows: =0.3、 =0.3、 =0.4, which emphasizes the actual correlation signals in the data content. The confidence score is represented as a continuous value between 0.0 and 1.0, where a higher value indicates a higher probability that the challenge has actually been raised by the opposing party or a higher threat to the credibility of one's own data. The continuous confidence scores are mapped to discrete levels according to preset thresholds: a confidence score greater than or equal to 0.7 is mapped to high confidence, greater than or equal to 0.3 and less than 0.7 is mapped to medium confidence, and less than 0.3 is mapped to low confidence. Discrete levels are used for quick filtering and prioritization, while continuous values ​​are used for precise comparison. All predicted challenge points and their confidence scores for each piece of one's own data are summarized and written into an adversarial challenge prediction document. The document is grouped by data entry, and each group is arranged in descending order of confidence.

[0028] Step S160: Detect whether there are data records with the other party's attribution tag in the data classification list, and determine whether to execute the reverse analysis branch based on the detection result; In this embodiment, the system detects whether there are data records with the "Opposite Attribution" tag in the data classification list. Each record in the data classification list is traversed. If at least one record has an "Opposite Attribution" tag value of OPPONENT, it is determined that "Opposite data exists"; if no record with the OPPONENT tag is found after traversal, it is determined that "Opposite data does not exist". If the detection result is "Opposite data exists", the execution flag is set to execute reverse analysis, and the reverse analysis step continues. If the detection result is "Opposite data does not exist", the execution flag is set to skip reverse analysis, skipping the reverse analysis step and directly jumping to the document assembly step. The execution path decision signal is written to the shared workspace in the form of a declaration file. Under the skip reverse analysis path, the reverse analysis step is not invoked, reducing computational overhead by approximately 30% compared to the complete execution path. This is mainly reflected in skipping the data reading and parsing of the "Opposite data" catalog, and skipping the decision tree traversal and template matching calculation generated by the reverse challenge.

[0029] In one embodiment, when the execution flag is set to perform reverse analysis, the reverse analysis steps are executed. The purpose of reverse analysis is to examine the weaknesses and attack points of the opponent's data from one's own perspective, forming a symmetrical analysis structure with the positive challenge, and providing two-way intelligence for subsequent attack and defense strategy formulation. All data entries in the data classification list with the attribute tag "OPPONENT" are traversed, and the same cataloging operation as for one's own data is performed on each data entry: identifying data type tags, extracting core attributes, and extracting supporting conclusion identifiers. The cataloging results are written to a structured opponent's data directory file. The same multidimensional evaluation decision tree as for the positive analysis is loaded, and the same decision tree traversal operation is performed on each data entry in the opponent's data directory: entering the corresponding data type branch from the root node, performing evaluation dimension applicability judgment and pruning, and reaching the activated challenge template leaf node along the path to generate challenge points and calculate confidence scores. The difference between reverse challenge points and positive challenge points lies in the challenge perspective—positive challenge examines one's own data from the opponent's perspective, while reverse challenge examines the opponent's data from one's own perspective. The reverse challenge points and their confidence scores are summarized and written to a reverse challenge document file.

[0030] Step S170: Summarize the results of each stage and assemble them into a structured analysis document.

[0031] In this embodiment, the output file paths for each preceding step are recorded, including the path to the self-side data directory file, the path to the adversarial challenge prediction document file, the path to the opponent's data directory file (if it exists), and the path to the reverse challenge document file (if it exists). The declaration file is in key-value pair format, where the key is the step identifier and the value is the relative path of the output file of that step in the shared workspace. The document assembly module obtains the locations of all intermediate files to be merged by parsing the declaration file, without needing to establish direct communication with the preceding steps. Content is merged according to a preset document template. The preset document template defines the chapter structure and layout order of the final analysis document: the first chapter is filled with the self-side data directory content; the second chapter is filled with the adversarial challenge prediction document content; the third chapter is filled with the opponent's data directory content if it exists, otherwise a preset placeholder text is inserted, which is a fixed-format prompt statement; the fourth chapter is filled with the reverse challenge content if it exists, otherwise a preset placeholder text is inserted. The merged document content is rendered into the final structured analysis document according to the format specifications of the preset template, written to the specified output path in the shared workspace, and the output path is registered through the declaration file. In the final structured analysis document, each predicted point of contention is accompanied by its confidence score and a complete record of its decision tree reasoning path. The points of contention are arranged in descending order of confidence score, facilitating risk prioritization and resource allocation for users. Regardless of whether the execution path is a complete analysis or a simplified analysis, the structural integrity of the document is guaranteed.

[0032] This solution achieves systematic and automatic evolution of data analysis solutions by constructing a multi-dimensional evaluation decision tree, splitting data attribution from multiple perspectives, generating adversarial challenges based on decision tree traversal, outputting probabilistic confidence scores, implementing data-driven conditional execution control, and loosely coupled stage communication based on declaration files. It can capture the interaction effect between data attribution and evaluation strategies, improve the coverage and relevance of analysis, and dynamically adjust computational overhead according to data characteristics to reduce unnecessary waste of computational resources.

[0033] This method constructs a multidimensional evaluation decision tree; reads structured context information and extracts the position labels of the current analysis subject; based on the position labels, the input dataset is split into a subset of self-attribution, a subset of opposing party's attribution, and a subset of common attribution, resulting in a data classification list; cataloging is performed on the data in the subset of self-attribution, identifying data type labels and extracting core attributes and supporting conclusion identifiers, resulting in a self-data catalog; the feature information of each data entry in the self-data catalog is used as input, and the multidimensional evaluation decision tree is traversed layer by layer from the root node, through data type judgment, evaluation dimension applicability judgment, and template trigger condition judgment, until the activated questioning template leaf node is reached, generating predicted questioning points and calculating confidence scores for each predicted questioning point; the existence of data records with opposing party attribution labels in the data classification list is detected, and the execution of a reverse analysis branch is determined based on the detection results; the results of each stage are summarized and assembled into a structured analysis document. This method can capture the interaction effect between data attribution and evaluation strategies, realize the systematic and automatic evolution of the analysis scheme, and improve the analysis coverage and relevance.

[0034] This invention also provides a device for analyzing objections based on multi-party data, which is used to execute any of the aforementioned embodiments of the method for analyzing objections based on multi-party data. Specifically, please refer to... Figure 2 , Figure 2 This is a schematic block diagram of a dispute analysis device based on multi-party data provided in an embodiment of the present invention. The dispute analysis device 100 based on multi-party data can be configured in a server.

[0035] like Figure 2 As shown, the questioning analysis device 100 based on multi-source data includes a construction module 110, an extraction module 120, a splitting module 130, an identification module 140, a scoring module 150, a detection module 160, and an assembly module 170.

[0036] The construction module 110 is used to construct a multi-dimensional evaluation decision tree. The multi-dimensional evaluation decision tree includes a data type judgment layer, an evaluation dimension judgment layer, a template trigger condition judgment layer, and a questioning template leaf node layer. The data type judgment layer performs a first-level branch based on the data type label, the evaluation dimension judgment layer performs a second-level branch based on the applicable conditions of the evaluation dimension, the template trigger condition judgment layer performs a third-level branch based on the trigger conditions of the questioning template, and the questioning template leaf node layer stores the questioning template configuration. Extraction module 120 is used to read structured context information and extract the position tags of the current analysis subject; The splitting module 130 is used to split the input dataset into a subset belonging to one's own side, a subset belonging to the other side, and a common subset belonging to the other side based on the position labels, so as to obtain a data classification list; The identification module 140 is used to perform a cataloging operation on the data in the subset to which the self belongs, identify data type tags and extract core attributes and supporting conclusion identifiers to obtain the self data catalog; The scoring module 150 is used to take the feature information of each data in the self-owned data catalog as input, traverse layer by layer from the root node of the multidimensional evaluation decision tree, and after data type judgment, evaluation dimension applicability judgment and template trigger condition judgment, reach the activated questioning template leaf node, generate predicted questioning points and calculate the confidence score for each predicted questioning point. The detection module 160 is used to detect whether there are data records with the other party's belonging label in the data classification list, and to determine whether to execute the reverse analysis branch based on the detection result; Assembly module 170 is used to summarize the results of each stage and assemble them into a structured analysis document.

[0037] In one embodiment, the construction module 110 includes: The construction unit is used to construct a supplementary evaluation rule subtree for the corresponding data type based on the preset data type label, and attach the supplementary evaluation rule subtree to the branch point of the data type corresponding to the data type judgment layer. The supplementary evaluation rule subtree is attached in parallel with the existing general evaluation dimension branch of the corresponding data type branch.

[0038] In one embodiment, the extraction module 120 includes: The classification unit is used to extract the value of a position identifier field as a position label if there is a clear position identifier field in the structured context information; if there is no position identifier field but there is a field value that matches the name of a known stakeholder, the position label is determined by the field value that matches the name; if none of the above conditions are met, a pre-configured default position label is used.

[0039] In one embodiment, the splitting module 130 includes: The label definition unit is used to define a piece of data as its own attribution label if the source identifier or content of the data clearly supports the stakeholder corresponding to the position label; to define a piece of data as the other party's attribution label if the source identifier or content of the data clearly supports the stakeholder opposed to the position label; and to define a piece of data as a common attribution label if the source identifier or content of the data is shared by both parties, provided by a neutral party, or the attribution cannot be determined.

[0040] In one embodiment, the identification module 140 includes: The category identification unit is used to traverse all data entries in the data classification list that belong to itself, and perform data type identification for each data entry based on at least one of the following signals: explicit type label field, data content format feature matching, and data source system type. The writing unit is used to extract the core attributes and supporting conclusion identifiers of the data and write the cataloging results into a structured self-owned data catalog file.

[0041] In one embodiment, the bisecting module 150 includes: The reading unit is used to read the corresponding data type tag for each piece of data in the self-owned data directory as the first-level branch selection condition to enter the corresponding data type branch node; The detection unit is used to traverse all evaluation dimension judgment nodes of the second level, and detect whether the applicable conditions carried by the corresponding node match the core attribute features of the current data. If they match, it proceeds to the next level; otherwise, it performs pruning operations. The filling unit is used to continue traversing the template trigger condition judgment node of the third level for the evaluation dimension branch that has passed the applicability judgment. When the core attribute of the data meets the trigger condition, it reaches the questioning template leaf node of the fourth level, reads the questioning template configuration stored in the leaf node, and fills the specific attribute value of the data into the placeholder variable of the template to generate the predicted questioning point.

[0042] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the questioning analysis method based on multi-party data as described above.

[0043] In another embodiment of the invention, a computer-readable storage medium is provided. This computer-readable storage medium may be a non-volatile computer-readable storage medium. The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to perform the challenge analysis method based on multi-party data as described above.

[0044] Those skilled in the art will readily understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention.

[0045] In the embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Units with the same function may be grouped into one unit. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, or it may be an electrical, mechanical, or other form of connection.

[0046] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention, depending on actual needs.

[0047] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0048] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), magnetic disks, or optical disks.

[0049] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for analyzing points of contention based on multi-source data, characterized in that, include: A multidimensional evaluation decision tree is constructed, which includes a data type judgment layer, an evaluation dimension judgment layer, a template trigger condition judgment layer, and a questioning template leaf node layer. The data type judgment layer performs a first-level branch based on the data type label, the evaluation dimension judgment layer performs a second-level branch based on the applicable conditions of the evaluation dimension, the template trigger condition judgment layer performs a third-level branch based on the trigger conditions of the questioning template, and the questioning template leaf node layer stores the questioning template configuration. Read structured context information and extract the stance labels of the current analysis subject; Based on the aforementioned position labels, the input dataset is split into a subset belonging to one's own side, a subset belonging to the other side, and a common subset belonging to the other side, resulting in a data classification list; Perform a cataloging operation on the data in the subset to which your own data belongs, identify data type tags and extract core attributes and supporting conclusion identifiers to obtain your own data catalog; The feature information of each data item in the self-owned data catalog is used as input. Starting from the root node of the multidimensional evaluation decision tree, the system traverses layer by layer, passing through the data type judgment layer, the evaluation dimension judgment layer, and the template trigger condition judgment layer, until the activated questioning template leaf node is reached. Predicted questioning points are generated and a confidence score is calculated for each predicted questioning point. Detect whether there are data records with the other party's belonging label in the data classification list, and determine whether to execute the reverse analysis branch based on the detection result; The results from each stage are summarized and assembled into a structured analysis document.

2. The method for analyzing points of contention based on multi-party data as described in claim 1, characterized in that, The construction of the multidimensional evaluation decision tree includes: Based on the preset data type labels, a supplementary evaluation rule subtree for the corresponding data type is constructed, and the supplementary evaluation rule subtree is attached to the branch point of the data type corresponding to the data type judgment layer. The supplementary evaluation rule subtree is attached in parallel with the existing general evaluation dimension branch of the corresponding data type branch.

3. The method for analyzing points of contention based on multi-party data as described in claim 1, characterized in that, The step of reading structured context information and extracting the stance tags of the current analysis subject includes: If there is a clear stance identifier field in the structured context information, extract the value of that field as a stance label; If the position identifier field does not exist but a field value matching a known stakeholder name exists, the position label is determined by the field value matching the name. If none of the above conditions are met, the pre-configured default position label will be used.

4. The method for analyzing points of contention based on multi-party data as described in claim 1, characterized in that, The step of splitting the input dataset into a subset of one's own affiliation, a subset of the other party's affiliation, and a common affiliation subset based on the position labels includes: If the source identifier or content of the data clearly supports the stakeholder corresponding to the position label, then the data is defined as the own affiliation label; If the source identifier or content of the data clearly supports a stakeholder that is opposed to the stated position label, then the data is defined as the other party's affiliation label; If the source identifier or content of the data is shared by both parties, provided by a neutral party, or its ownership cannot be determined, then the data is defined as a public ownership label.

5. The method for analyzing points of contention based on multi-party data as described in claim 1, characterized in that, The cataloging operation performed on the data in the subset to which the party belongs includes: Iterate through all data entries in the data classification list that belong to itself, and perform data type identification for each data entry based on at least one of the following signals: explicit type label field, data content format feature matching, and data source system type. Extract the core attributes and supporting conclusion identifiers of the data, and write the cataloging results into a structured self-owned data catalog file.

6. The method for analyzing points of contention based on multi-party data as described in claim 1, characterized in that, The process involves traversing the multidimensional evaluation decision tree layer by layer, starting from the root node, passing through the data type judgment layer, the evaluation dimension judgment layer, and the template trigger condition judgment layer, until reaching the activated questioning template leaf node, generating predicted questioning points, including: For each piece of data in the self-owned data directory, the corresponding data type tag is read as the first-level branch selection condition to enter the corresponding data type branch node; Iterate through all evaluation dimension judgment nodes of the second level, and check whether the applicable conditions carried by the corresponding node match the core attribute features of the current data. If they match, proceed to the next level; otherwise, perform pruning operations. For the evaluation dimension branch that passes the applicability judgment, continue to traverse the template trigger condition judgment node of the third level. When the core attribute of the data meets the trigger condition, reach the questioning template leaf node of the fourth level, read the questioning template configuration stored in the leaf node, and fill the specific attribute value of the data into the placeholder variable of the template to generate the predicted questioning point.

7. The method for analyzing points of contention based on multi-party data as described in claim 1, characterized in that, The calculation of a confidence score for each predicted point of doubt includes: The confidence score is calculated using the following formula: ， In the formula, This indicates the strength of the match between the evaluation dimension and the data features. This indicates a question about the specificity of the template. This indicates the strength of the correlation signals in the data content that support the current point of contention. , , This represents the preset weighting coefficients, and .

8. A device for analyzing points of contention based on multi-party data, characterized in that, include: The construction module is used to construct a multi-dimensional evaluation decision tree. The multi-dimensional evaluation decision tree includes a data type judgment layer, an evaluation dimension judgment layer, a template trigger condition judgment layer, and a questioning template leaf node layer. The data type judgment layer performs a first-level branch based on the data type label, the evaluation dimension judgment layer performs a second-level branch based on the applicable conditions of the evaluation dimension, the template trigger condition judgment layer performs a third-level branch based on the trigger conditions of the questioning template, and the questioning template leaf node layer stores the questioning template configuration. The extraction module is used to read structured context information and extract the stance tags of the current analysis subject; The splitting module is used to split the input dataset into a subset belonging to one's own side, a subset belonging to the other side, and a common subset belonging to the other side, based on the position labels, to obtain a data classification list; The identification module is used to perform cataloging operations on the data in the subset to which the party belongs, identify data type tags and extract core attributes and supporting conclusion identifiers to obtain the data catalog of the party. The scoring module is used to take the feature information of each data in the self-owned data catalog as input, traverse layer by layer from the root node of the multidimensional evaluation decision tree, and after data type judgment, evaluation dimension applicability judgment and template trigger condition judgment, reach the activated questioning template leaf node, generate predicted questioning points and calculate the confidence score for each predicted questioning point. The detection module is used to detect whether there are data records with the other party's attribution tag in the data classification list, and to determine whether to execute the reverse analysis branch based on the detection results; The assembly module is used to summarize the results of each stage and assemble them into a structured analysis document.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the challenge analysis method based on multi-party data as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to perform the challenge analysis method based on multi-party data as described in any one of claims 1 to 7.

Images