A conflict resolution method based on large models and knowledge graphs and multiple data standard definitions

Abstract

This invention relates to a method for resolving conflicts in multiple data standard definitions based on large models and knowledge graphs, belonging to the fields of public data governance and data standardization. The invention includes: extraction and preprocessing of multiple data standard definitions; semantic representation and large model parsing: using a large model to perform semantic analysis on the definition of each data item, generating semantic embedding vectors and key features; knowledge graph construction and entity alignment: constructing a knowledge graph containing all data items, and using an entity alignment algorithm to connect semantically similar data item nodes from different standards to form candidate alignment relationships; conflict detection and type identification: identifying definition conflict points and classifying and labeling conflict types based on the aligned data item definitions, comparative attributes, and values; and a globally optimized conflict resolution algorithm for conflict resolution and unified definition generation. This invention solves the problems of data silos and semantic conflicts caused by inconsistent data standard definitions in existing technologies.

Description

Technical Field

[0001] This invention relates to a method for resolving conflicts in multiple data standard definitions based on large models and knowledge graphs, belonging to the technical fields of public data governance and data standardization. Background Technology

[0002] In public data governance, various departments and information systems often formulate their own data standards, lacking unified data definitions and rules. This leads to data fragmentation horizontally and inconsistent standards vertically, with different departments describing the same data in different ways and using different data formats. These inconsistencies in data standards cause cross-departmental and cross-system data semantic conflicts, hindering data sharing, exchange, and integrated utilization. To address this issue, the industry typically hopes to establish a complete, unified, and standardized data standard system to avoid semantic conflicts through unified data definitions. However, in practice, coordinating multiple data standards and eliminating conflicts usually requires significant manual intervention. Traditional methods mainly rely on manually formulating mapping rules or writing conversion scripts to align data definitions from different systems. This is not only labor-intensive but also prone to overlooking conflicts and difficult to maintain in a timely manner. While some related technical solutions and research have been developed to address the inconsistency problem of multiple data standard definitions, many shortcomings remain.

[0003] Manual mapping versus static rule-based methods: Traditional approaches often rely on domain experts to manually create standard mapping tables or write fixed conversion rules, manually mapping corresponding items from different standards. This method was used in early standardization practices; for example, some studies proposed technical strategies to resolve data encoding standard conflicts by manually pre-defining the correspondence between various standard codes. Similarly, in the field of Electronic Data Interchange (EDI), systems have emerged that centrally maintain mappings between standards for data conversion. The advantages of this type of method are its simple implementation principle, the ability to formulate targeted rules for known conflicts, and low initial investment. However, its disadvantages are also quite obvious: First, manual mapping is time-consuming, labor-intensive, and prone to errors, making it difficult to cover all conflict situations; second, once rules are formulated, they lack flexibility and need frequent updates to address new standards or changing definitions; third, manual methods often struggle to detect and define rules for semantic differences (such as subtle differences in the connotation and denotation of concepts). Therefore, static mapping rule-based methods can barely function when the number of standards is small and does not change much, but they are inadequate in an environment of constantly evolving multi-source heterogeneous data standards, urgently requiring more intelligent solutions.

[0004] Knowledge graph-based conflict detection methods: Recently, technologies have begun to utilize knowledge graphs to handle standard definition conflicts. For example, patent CN113449118A from South China University of Technology proposes first structuring the unstructured content of standard documents into a standard knowledge graph, and then using techniques such as text similarity matching and multi-entity list matching to discover conflicting entities among multiple standard knowledge graphs, thereby achieving conflict detection between standard documents. The advantage of this method is that it introduces a knowledge graph to uniformly represent the knowledge of different standards, and can automatically compare the content of different standards to a certain extent, discovering obvious contradictions. Compared with purely manual methods, knowledge graph methods improve the coverage and efficiency of conflict detection. In addition, scholars in fields such as smart homes have also studied implicit conflict detection methods based on knowledge graphs to discover conflicts in device linkage rules, further demonstrating the potential of knowledge graphs in complex conflict identification. However, such methods currently mainly focus on conflict detection, relying on similarity matching and other means, and may still miss cases where the definition wording is different but the essence is conflicting. Moreover, more importantly, there is still no clear solution for how to automatically resolve conflicts after they are detected. Existing patent CN113449118A focuses on conflict detection, but detected conflicts often still require manual analysis and resolution, and it does not provide a systematic automated resolution mechanism. Therefore, while existing knowledge graph-based methods improve the ability to detect conflicts, they are insufficient to completely resolve them and need to be combined with stronger semantic understanding and decision-making mechanisms.

[0005] Standard alignment methods combining knowledge graphs and pre-trained models: In specific domains, there have been attempts to combine knowledge graphs with pre-trained models for standardization. For example, Chinese patent CN113191156A proposes a standardization system and method for medical examination items based on medical knowledge graphs and pre-trained models. It utilizes a medical domain knowledge graph and a pre-trained language model to map the names of examination items used by different hospitals or institutions to unified standard names, thereby improving the accuracy of terminology alignment. The advantage of this method is that it introduces pre-trained models (such as BERT) to understand the semantics of medical terms, achieving better intelligent matching results compared to pure rule or dictionary methods, and solving the standardization problem of different expressions for the same examination item in the medical field. However, such solutions mainly focus on unifying standard names of terms, belonging to name mapping in a specific domain, and do not deeply address the conflicts in complex definition logic. Furthermore, it primarily uses the pre-trained model as an auxiliary matching tool and has not yet demonstrated the ability to actively resolve conflicts. In broader scenarios of cross-domain data standard conflicts, terminology matching alone is insufficient to resolve the problem: different standards may specify different calculation formulas, units of measurement, or compliance thresholds for the same concept. These conflicts go beyond simple name inconsistencies and are not addressed by existing solutions such as medical standardization. This invention proposes a comprehensive solution to these complex conflicts: it not only focuses on semantic matching but also provides means to resolve logical conflicts.

[0006] In summary, existing technologies either rely on manual / hard-coded methods, resulting in low levels of intelligence and high maintenance costs; or they can only detect conflicts or resolve localized name standardization, failing to automatically and comprehensively resolve definitional contradictions arising from the coexistence of multiple data standards. These shortcomings make it difficult to integrate different data standards, creating hidden dangers in data sharing and exchange, and seriously affecting system interaction and data quality. Therefore, it is necessary to propose a new inventive solution that combines the latest artificial intelligence large-scale model technology and knowledge graph reasoning capabilities to specifically improve upon the aforementioned deficiencies and achieve automatic resolution and unified integration of multi-source data standard definition conflicts. Summary of the Invention

[0007] The technical problem solved by this invention is that it provides a method for resolving conflicts in multiple data standard definitions based on large models and knowledge graphs, in order to address the data silos and semantic conflicts caused by inconsistent data standard definitions in existing technologies. This invention comprehensively understands data definitions in different standards and accurately detects conflicts by introducing large model semantic recognition technology and knowledge graph construction methods. Combined with conflict network modeling and intelligent optimization algorithms, it achieves automated conflict resolution. Furthermore, with standard updates and continuous feedback monitoring mechanisms, it continuously improves the data standard system, thereby enhancing the consistency, accuracy, and efficiency of data fusion and sharing in public data governance.

[0008] The technical solution of this invention is: a conflict resolution method based on multiple data standard definitions using large models and knowledge graphs, the method comprising:

[0009] S1. Extraction and preprocessing of multiple data standard definitions: Extract data item definitions from various data standard documents, clean and standardize their representation;

[0010] S2. Define semantic representation and large model parsing: Use the large model to perform semantic analysis on the definition of each data item, and generate semantic embedding vectors and key features;

[0011] S3. Knowledge Graph Construction and Entity Alignment: Construct a knowledge graph containing all data items, and use an entity alignment algorithm to connect data item nodes with similar semantics in different standards to form candidate alignment relationships;

[0012] S4. Conflict Detection and Type Identification: Define and compare attributes and values ​​for aligned data items, identify conflict points and classify and label conflict types;

[0013] S5. Globally optimized conflict resolution algorithm for conflict resolution and unified definition generation: Design a globally optimized conflict resolution algorithm to resolve conflicts; dynamically adjust matching relationships or attribute values ​​to generate a unified solution and output the fused data standard definition.

[0014] Further, S1 includes:

[0015] Processing multiple data standard documents and performing unified parsing, which contain several data items and their definitions; processing multiple data standard documents and performing unified parsing, including:

[0016] Using text parsing and format conversion technologies, standard documents from different sources are transformed into structured data representations; custom parsing scripts are written to extract "data item names", "definition text", and related attribute fields; if the document format is inconsistent or the definition is buried in the natural language description, the large model API is called to assist in parsing complex text, including enabling the large model to identify key fields in the definition sentence;

[0017] S1.1 First, input the various data standard documents and load them into the parsing tool;

[0018] For structured formats, read the fields directly;

[0019] For unstructured text, use regular expression rules or the NLP capabilities of large models to locate the "data item name: definition..." pattern, and process and extract the name and definition content;

[0020] If the definition contains attribute information, the attribute name and attribute value are also parsed out.

[0021] S1.2 Next, perform data cleaning and standardization: unify the terminology used from different sources and fill in any missing information; during this process, call the large model to simply rewrite or complete the definition text; output a standardized list of data item definitions or an initial knowledge base: containing a structured record of each data item.

[0022] Further, S2 includes:

[0023] Use large-scale model LLM or large-scale pre-trained models to define natural language processing for each data item;

[0024] On the one hand, the data item definition is converted into a high-dimensional semantic vector representation through the embedding interface of the large model;

[0025] On the other hand, the parsing capabilities of large models are utilized to extract key semantic features from the data item definitions; including:

[0026] Use a pre-trained sentence vector model to generate embedding vectors for the definition text or directly call the large model API to request the output of the defined structured information, which includes semantic summary and a list of synonyms;

[0027] The specific process includes:

[0028] First, the "name" and "definition description" are concatenated into a complete descriptive sentence, which is then used as the text for model processing.

[0029] The processing is then divided into two parts:

[0030] (1) Semantic embedding vector generation: Using the vectorization function provided by the large model, the definition of each data item is sent into the large model to obtain the semantic embedding representation;

[0031] (2) Key feature extraction: Utilize the question-answering or parsing capabilities of large models to extract key attributes and semantic tags from the data item definitions;

[0032] The specific approach involves constructing a prompt query model; the model outputs a structured result based on the trained knowledge; the model supplements the implicit attributes in the plain text definition; and the semantic labels and normalized attribute values ​​defined for each data item are obtained.

[0033] Further, S3 includes:

[0034] A knowledge graph is constructed to represent multi-source data standards, and an entity alignment algorithm is used to discover identical or equivalent data items in the graph. The knowledge graph uses nodes to represent data items and edges to represent potential equivalence relationships between data items of different standards.

[0035] During the alignment process, multidimensional similarity is comprehensively considered; specifically, a fusion similarity calculation method is used, a custom similarity formula is defined, and a threshold is set to judge candidate matches; at the same time, a combination of heuristics and algorithms is adopted: first, preliminary corresponding candidates are established by name or keywords, and then the matching score is accurately calculated by cosine similarity; for complex matching problems, graph algorithms are introduced to find the globally optimal matching scheme; and candidate alignment relationships are generated.

[0036] Furthermore, the specific steps of S3 include:

[0037] S3.1, The input is the node representation of each data item; first, construct the knowledge graph:

[0038] A subgraph is created for each data standard, and all data items are placed as nodes in the graph; each node has an attribute and also carries a semantic vector. ;

[0039] S3.2 Next, calculate the cross-standard node similarity; for any two data item nodes from different standards... and Define comprehensive semantic similarity It integrates multiple factors: similarity of names and terms, semantic similarity of definition text, and similarity of attribute values; and comprehensive semantic similarity. The calculation formula is as follows:

[0040]

[0041] in, It's about name similarity. It defines the semantic similarity of text. It is attribute similarity. For the weighting coefficients, satisfying These weights are used to balance the importance of each part, and are adjusted based on experience or the training set.

[0042] S3.3 Calculate all cross-standard node pairs of Then, determine the candidate alignment relationship: when Value higher than the set threshold Mark the two data items as a "candidate match pair";

[0043] In a knowledge graph, a temporary "equivalent" edge is added between each pair of candidate matching data item nodes, and the edge weight is recorded. value;

[0044] For cases with multiple standards, the process of constructing a knowledge graph is reduced to finding similar groups in a multi-partite graph;

[0045] Output a preliminary fused knowledge graph: nodes are data items from various standards, and highly similar candidates across standards are connected by edges; each node is connected to multiple nodes from other standards.

[0046] Furthermore, in S3.2, the specific calculation details of the comprehensive semantic similarity are as follows:

[0047] (1) Name similarity Compare the similarity of data item names or terms; if two names are exactly the same, let... If the names are different but overlap after word segmentation or synonym normalization, the Jaccard coefficient is used to calculate the overlap of the word sets. The calculation formula is as follows:

[0048]

[0049] , This refers to the data item node. and The corresponding data item name, Tokens, is the "set of words" after text segmentation. It is the absolute value;

[0050] At the same time, using the synonym list extracted by S2, synonymous terms are considered as matches;

[0051] (2) Text semantic similarity The semantic similarity of data item definitions is calculated using semantic embedding vectors generated by a large model; the cosine similarity formula is used to compare the semantic embedding vectors of two data items. Perform the calculation:

[0052]

[0053] in," " represents the vector dot product, It is the Euclidean norm of the semantic embedding vector. The value range is [0,1];

[0054] (3) Attribute similarity Compare the structured attribute values ​​of two data items, including data type, length limit, unit, and value range, and calculate the degree of matching of these attributes; let... For the set of all considered attributes, the attribute similarity is defined as a weighted average of the similarities of all attributes:

[0055]

[0056] in, Represents data items In attributes The value on, It is an attribute Importance weights, satisfying , |P| is the number of attributes in the attribute set P, assuming that each attribute has equal weight or that the primary key attribute is highlighted; It is a similarity function of attribute values;

[0057] By comparing attributes one by one and weighted summation, the similarity between two data items defined in terms of structural constraints is quantified.

[0058] Further, S4 includes:

[0059] For candidate matching relationships in the knowledge graph, perform consistency checks and conflict pattern identification:

[0060] A method combining rule-based algorithms with large-scale model-assisted analysis is used to determine whether semantics are truly conflicting or merely differing in expression.

[0061] The method of combining rule-based algorithms with large-scale model-assisted analysis includes: using predefined conflict determination rules to quickly screen for conflicts, while calling large-scale models to compare and analyze the matching definition text;

[0062] By combining the two, the location and type of conflict can be accurately identified; common conflict types include: hit ambiguity conflict, attribute inconsistency conflict, and definition semantic conflict.

[0063] Furthermore, the specific steps of S4 include:

[0064] S4.1 The input is the knowledge graph from step S3; First, one-to-many matching conflict detection: scan the candidate equivalent edges of each node and check whether there is a node that connects to multiple other nodes with similarity scores;

[0065] S4.2 Next, for each pair of candidate matches Perform attribute consistency checks: compare the attribute values ​​of the two nodes to identify actual conflicts;

[0066] Define a conflict indicator function: if a certain attribute exists Make Furthermore, if the difference is not negligible, then the attribute is determined. Conflict; formally represented as:

[0067]

[0068] For each detected attribute conflict, the degree of conflict is also evaluated;

[0069] Then, define semantic conflict checking: use a large model to deeply compare the definition description;

[0070] The S4 can organize a list of conflict detection results; each conflict record includes: the data item node involved, the conflict type, the specific manifestation, and the conflict severity index.

[0071] Further, S5 includes:

[0072] Design a globally optimized conflict resolution algorithm that combines intelligent optimization methods to search for the optimal alignment and fusion scheme. The globally optimized conflict resolution algorithm takes the candidate alignment relationship obtained by S3 as the initial solution, and through random perturbation and stepwise optimization, it seeks the best solution in the solution space to maximize the overall similarity and minimize the conflict.

[0073] Simultaneously, a dynamic weight adjustment mechanism is introduced to adjust the similarity weighting coefficient in real time based on conflict situations during the optimization iteration process. Gradually increase the penalty for conflicting terms to guide the algorithm away from local optima;

[0074] In addition, specific strategies are incorporated into the conflict resolution process: value fusion is applied to attribute conflicts, selection or splitting is performed in matching for one-to-many conflicts, a large model is called to generate unified definition text, the differences between multiple standards are balanced, and high-quality fusion results are output.

[0075] Furthermore, the specific steps of S5 include:

[0076] The input consists of a conflict list and the current matching graph state; the globally optimized conflict resolution algorithm is first initialized:

[0077] Establish the objective function for the optimization problem; define the objective function. Measured by "overall matching quality" minus "conflict cost", maximizing Objective function Formalized as:

[0078]

[0079] in, Represents a selected set of matching pairs, the first item It is the total similarity score of all selected matches, the second item. It's an attribute conflict penalty; the third item. It is the number of matches in the set that violate the one-to-one rule;

[0080] Then, the iterative optimization phase begins, employing a simulated annealing strategy to traverse the solution space. Specific operations include:

[0081] (1) Define neighborhood operation: randomly select a method for the current matching set The modification is used as a candidate for a new solution;

[0082] (2) Design neighborhood operations for attribute conflicts, which generate different new solutions. ;

[0083] (2) Evaluation and acceptance criteria: Calculate the new scheme objective function ;Compare ;

[0084] if Then accept unconditionally. As the current solution;

[0085] if If the solution is poor, then according to the Metropolis criterion of simulated annealing, the inferior solution will be accepted with a certain probability in order to escape the local optimum.

[0086] The specific acceptance probability depends on the current "temperature" parameter. ;

[0087] (3) Dynamic weight adjustment: During the iteration process, the similarity weighting coefficient or the objective function penalty coefficient is adjusted periodically according to the remaining conflict situation;

[0088] The dynamic adjustment function is based on the current conflict rate: Let Indicates the attributes in the current matching set The proportion of conflicts, then update the attributes. Importance weight The updated weights are:

[0089]

[0090] in This is the adjustment coefficient; It refers to the importance weight of attribute 𝑝 in the current iteration, representing the current assessment of the importance of the attribute when it was first assigned to this iteration; This means that in the next iteration, the importance weight of attribute p will be updated to this value;

[0091] (4) Iterative convergence and optimal solution generation: As the temperature decreases and the weights are adjusted, the algorithm will gradually converge to a stable matching scheme. Output the final set of matching relationships. ;

[0092] Based on this final alignment result, a unified knowledge graph is constructed: matching nodes are merged into a single concept node or a mapping table is established to indicate their equivalence; for each set of equivalent data items, the final standard definition value is determined: if the attributes differ, they are merged according to a certain strategy; common strategies include credibility priority or retaining the union, as well as majority rule;

[0093] If a unified definition description text needs to be generated during the merging process, the large model is called again, and the definitions of the multi-source data items in the same group are input, so that the model can write a comprehensive version.

[0094] If the definition generated by the large model is not ideal, the prompts are constrained based on the attribute fusion results; finally, a standardized definition for each unified concept is obtained.

[0095] This invention provides a method for resolving conflicts in the definition of multiple data standards based on a large model (pre-trained artificial intelligence model) and knowledge graphs. It aims to address the technical problems of inconsistent definitions and difficulty in automatically identifying and resolving conflicts when multiple data standards coexist. Specific protected technical points include:

[0096] For standard documents or data dictionaries from different sources, this invention first uses natural language processing technology to parse unstructured descriptions such as terminology definitions, extract key elements and relationships, and construct corresponding knowledge graph representations. This step transforms the definitions of various data standards into machine-understandable structured semantic networks, laying the foundation for subsequent conflict detection.

[0097] Large-scale pre-trained models (such as domain-specific pre-trained language models) are introduced to deeply understand and vectorize the semantics of various standard definitions. Leveraging the powerful language understanding capabilities of these large models, semantic alignment is performed on the same concept expressed in different ways, determining whether they refer to the same or similar entities. Compared to traditional keyword matching methods, large models can identify implicit synonym relationships and semantic conflicts, providing a basis for knowledge graph node alignment and conflict judgment.

[0098] This invention combines knowledge graph and large-scale model analysis results to automatically detect conflicting entities and relationships between multiple data standard knowledge graphs. By comparing the differences in the definitions of the same entity's attributes, constraints, and value ranges across different standards, it identifies incompatible or contradictory parts. For example, inconsistencies in units, different classification granularities, overlapping or contradictory definition ranges will all be detected. Unlike existing technologies that rely solely on simple text similarity or preset rules for detection, this invention combines the structured associations of knowledge graphs with the semantic reasoning of large-scale models to uncover more complex and hidden conflicts.

[0099] Upon detecting definition conflicts, this invention proposes an automated resolution strategy. Leveraging the reasoning capabilities of a large model, solutions are generated to address the conflicts, such as suggesting the adoption of a unified standard definition or introducing higher-level concepts into the knowledge graph to link concepts from different standards, thus forming a compatible new definition. The device of this invention includes a conflict resolution module that can fuse conflicting multi-source definitions based on suggestions generated by a pre-trained model and preset resolution rules. For example, it can automatically convert units and adjust thresholds for differences in numerical calibers, and it can merge fine-grained concepts into higher-level concepts through knowledge graph relationships for differences in concept hierarchy. After resolution is complete, the knowledge graph is updated accordingly to save the fused, consistent standard definition.

[0100] The beneficial effects of this invention are:

[0101] 1. This invention significantly improves the accuracy and recall of cross-standard semantic conflict identification through a large model's deep understanding of semantics, and can discover implicit conflicts that are difficult to detect by manual rules.

[0102] 2. This invention utilizes knowledge graphs to perform unified modeling of data standards and conflict network analysis, making the conflict localization and resolution process more intuitive and comprehensive, and making it easier to discover globally consistent solutions compared to traditional decentralized processing;

[0103] 3. This invention introduces a simulated annealing algorithm for global optimization search, avoiding suboptimal alignment that may result from greedy strategies, minimizing inconsistencies between multiple standard definitions, and ensuring more accurate and reliable fused data. The simulated annealing algorithm achieves a high degree of automation in handling data standard conflicts, greatly reducing the workload of manual comparison and modification, and shortening the cycle of cross-departmental data integration. When common data sources follow consistent standards, the shareability and fusionability of data are significantly improved, and data quality issues (such as inconsistent definitions and duplicate storage) are significantly reduced.

[0104] 4. Through a continuous feedback learning mechanism, the system can adapt to constantly changing development needs, which helps to maintain the robustness and advancement of the data standard system in the long term; the present invention will help to significantly improve the efficiency and quality of data integration and sharing in public data governance.

[0105] 5. The conflict resolution method provided by this invention can serve as an important component of standard knowledge graph applications, assisting in the automatic fusion and unification of cross-standard knowledge, and improving the quality and efficiency of standard digitization. Attached Figure Description

[0106] Figure 1 This is a flowchart from the present invention;

[0107] Figure 2 This is a schematic diagram of the technical structure of the conflict resolution method based on large models and knowledge graphs for multi-data standard definitions in this invention. Detailed Implementation

[0108] Example 1: Figure 1-Figure 2 As shown, a conflict resolution method based on large models and knowledge graphs and multiple data standards is proposed, the method comprising:

[0109] S1. Extraction and preprocessing of multiple data standard definitions: Extract data item definitions from various data standard documents, clean and standardize their representation to lay a foundation for subsequent processing;

[0110] Further, S1 includes:

[0111] This task involves processing multiple data standard documents (such as national standards, provincial standards, industry standards, and internal enterprise standards) and performing unified parsing. These documents contain numerous data items and their definitions, and their formats may vary (e.g., Excel data dictionaries, Word manuals, database metadata exports, etc.), requiring standardized parsing. Each data item typically has attributes such as name, definition description, data type, value range, and unit.

[0112] Processing multiple data standard documents and performing unified parsing, including:

[0113] Using text parsing and format conversion techniques, standard documents from different sources are transformed into structured data representations; custom parsing scripts are written (e.g., using Python's Pandas to read Excel files and regular expressions to extract Word paragraphs) to extract "data item names," "definition text," and related attribute fields; if the document format is inconsistent or the definition is embedded in the natural language description, the large model API is called to assist in parsing complex text, including enabling the large model to identify key fields in the definition sentences;

[0114] S1.1 First, input the various data standard documents and load them into the parsing tool;

[0115] For structured formats, read the fields directly;

[0116] For unstructured text, use regular expression rules or the NLP capabilities of large models to locate the "data item name: definition..." pattern, and process and extract the name and definition content;

[0117] If the definition contains attribute information (such as "type: integer, length: 10"), the attribute name and attribute value will also be parsed out.

[0118] S1.2 Next, perform data cleaning and normalization: standardize terminology used from different sources (e.g., "string" and "text" are synonymous), and fill in missing information (through inference or comparison with other standards). During this process, call the large model to simply rewrite or complete the definition text, making the expression more standardized (e.g., standardizing the expression of measurement units). Output a standardized list of data item definitions or an initial knowledge base: containing a structured record for each data item, such as {standard source, name, definition description, data type, unit, value range, ...}. This data will serve as input for subsequent steps. Each data item record includes a source identifier to distinguish cases where the same name comes from different standards. After preprocessing, ensure that data from different standards can be compared and analyzed on the same platform.

[0119] The normalized data items output in this step lay the foundation for subsequent semantic analysis and knowledge graph construction. Only through standardized and cleaned data can large models accurately understand the meaning and ensure that entities in the knowledge graph maintain a consistent attribute structure.

[0120] S2. Define semantic representation and large model parsing: Use the large model to perform semantic analysis on the definition of each data item, generate semantic embedding vectors and key features, and provide support for entity alignment.

[0121] Further, S2 includes:

[0122] Process each data item definition record from step 1, paying particular attention to the "Name" and "Definition Description" fields, as well as related attributes (type, unit, etc.). These will be used for semantic analysis to determine whether data items from different standards refer to the same concept;

[0123] Use large-scale model LLM or large-scale pre-trained models to define natural language processing for each data item;

[0124] On the one hand, the data item definition is converted into a high-dimensional semantic vector representation through the embedding interface of the large model;

[0125] On the other hand, the parsing capabilities of large models are utilized to extract key semantic features from the data item definitions; including:

[0126] Use pre-trained sentence vector models (such as BERT / DeepSeek models) to generate embedding vectors for definition texts or directly call the API of large models to request the output of structured information of the definition. Structured information includes semantic summaries and a list of synonyms. Large models can understand the meaning of professional definitions and extract implicit semantic relationships that are difficult for humans to discover.

[0127] The specific process includes:

[0128] First, the "name" and "definition description" are concatenated into a complete descriptive sentence, which is then used as the model's processing text; for example: "Data item name: X. Definition: ...";

[0129] The processing is then divided into two parts:

[0130] (1) Semantic embedding vector generation: Using the vectorization function provided by the large model, the definition of each data item is sent into the large model to obtain the semantic embedding representation. For example, calling the embedding interface of models such as DeepSeek or Sentence-BERT, taking the definition text as input, and outputting an... dimensional vector ( (This refers to the embedding dimension, such as 768 or 1024). Vector It captures the semantic features defined for this data item. The formula is expressed as:

[0131]

[0132] in The text embedding function representing the large model maps the concatenated text of data item names and definitions to a semantic vector space. In this way, all definitions are projected into the same vector space, and semantically similar definitions will have closer vector distances.

[0133] (2) Key feature extraction: Utilize the question-answering or parsing capabilities of large models to extract key attributes and semantic tags from the data item definitions;

[0134] The specific approach involves constructing prompts to query the large model, such as: "Please extract the data type, unit, value range, and keywords of the definition from the following data item definition: definition text definition text". Based on its trained knowledge, the large model outputs a structured result, such as: {data type: "Int", unit: "meter", value range: "0-100", keywords: ["length", "distance"]}. For attributes implicit in the plain text definition (such as units not explicitly given but inferred from the context), the large model is used to supplement them. This yields the semantic labels (synonyms, superordinate concepts, etc.) and normalized attribute values ​​for each data item definition.

[0135] The two sub-processes described above can be used in combination: for example, keywords and synonyms extracted by the large model can be used to enrich the vector representation (keywords are appended to the text before generating vectors to incorporate additional semantics), or the nearest neighbors of the embedded vectors can be used to roughly determine which definitions are related, and then the large model can be called to perform detailed analysis on potentially conflicting definitions. The output of this step is an enhanced list of definition representations: each data item contains its semantic embedding. The extracted semantic features (standardized attributes and keywords, etc.) will be used to construct the next knowledge graph, calculating the similarity between definitions and performing entity alignment.

[0136] This step utilizes the cleaned data from step 1 to ensure that the large model processes definition text with a consistent format and complete content, thereby improving parsing accuracy. The generated semantic vectors and features will be directly used in step 3 and are the core basis for calculating similarity links in the knowledge graph.

[0137] S3. Knowledge Graph Construction and Entity Alignment: Construct a knowledge graph containing all data items, and use an entity alignment algorithm to connect data item nodes with similar semantics in different standards to form candidate alignment relationships;

[0138] The processing step S2 outputs a definition representation of all data items, including node information for each data item (name, semantic vector, attribute features, etc.). These data items come from multiple different data standards and may contain duplicate entities referring to the same concept. They need to be represented in a knowledge graph, and cross-standard "aligned" entity pairs need to be identified.

[0139] Further, S3 includes:

[0140] A knowledge graph is constructed to represent multi-source data standards, and an entity alignment algorithm is used to discover identical or equivalent data items in the graph. The knowledge graph uses nodes to represent data items and edges to represent potential equivalence relationships between data items of different standards.

[0141] During the alignment process, multidimensional similarity is comprehensively considered; specifically, a fusion similarity calculation method is used, a custom similarity formula is defined, and a threshold is set to judge candidate matches; at the same time, a combination of heuristics and algorithms is adopted: first, preliminary corresponding candidates are established by name or keywords, and then the matching score is accurately calculated by cosine similarity; for complex matching problems, graph algorithms (such as simulated annealing will be processed in detail in step S5) are introduced to find the globally optimal matching scheme; and candidate alignment relationships are generated.

[0142] Furthermore, the specific steps of S3 include:

[0143] S3.1, The input is the node representation of each data item; first, construct the knowledge graph:

[0144] A subgraph is created for each data standard, and all data items are placed as nodes in the graph. Each node has attached attributes, such as "data type = Int, unit = meter" (extracted from step 1 / 2), and each node also carries a semantic vector. ;

[0145] S3.2 Next, calculate the cross-standard node similarity; for any two data item nodes from different standards... and Define comprehensive semantic similarity It integrates multiple factors: similarity of names and terms, semantic similarity of definition text, and similarity of attribute values; and comprehensive semantic similarity. The calculation formula is as follows:

[0146]

[0147] in, It's about name similarity. It defines the semantic similarity of text. It is attribute similarity. For the weighting coefficients, satisfying These weights are used to balance the importance of each part, and are adjusted based on experience or the training set.

[0148] S3.3 Calculate all cross-standard node pairs of Then, determine the candidate alignment relationship: when Value higher than the set threshold (Based on experience, such as 0.8) or when they rank highly in their respective closest matches, mark the two data items as a "candidate match pair";

[0149] In a knowledge graph, a temporary "equivalent" edge is added between each pair of candidate matching data item nodes, and the edge weight is recorded. value;

[0150] For cases with multiple standards, the process of building a knowledge graph is reduced to finding similar groups in a multi-partite graph; for example, for three standards A, B, and C, we can first calculate the similarity edges A–B, A–C, and B–C in pairs, and then add all high-similarity links to the graph.

[0151] Output a preliminary fusion knowledge graph: nodes are data items from various standards, and highly similar candidates across standards are connected by edges; each node is connected to multiple nodes from other standards (representing multiple candidate equivalence relations). This graph provides the foundation for the next step of conflict detection. It is important to emphasize that the alignment relations at this stage are "candidates," and there may be one-to-many, many-to-many, and other conflict scenarios, which need to be resolved through subsequent optimization.

[0152] This step utilizes the large-scale model semantic representation and attribute extraction results obtained in Step 2 to calculate cross-standard similarity and construct a knowledge graph. The innovative combination of similarity calculation formulas ensures comprehensive consideration of both text and structure, resulting in more accurate and comprehensive alignment candidates. The candidate matching results from this step are then passed to Step 4 for conflict analysis and further optimized and confirmed in Step 5.

[0153] Furthermore, in S3.2, the specific calculation details of the comprehensive semantic similarity are as follows:

[0154] (1) Name similarity Compare the similarity of data item names or terms; if two names are exactly the same, let... If the names are different but overlap after word segmentation or synonym normalization, the Jaccard coefficient is used to calculate the overlap of the word sets. The calculation formula is as follows:

[0155]

[0156] , This refers to the data item node. and The corresponding data item name, Tokens, is the "set of words" after text segmentation. It is the absolute value;

[0157] Meanwhile, by utilizing the synonym list extracted by S2, synonymous terms are considered as matches; for example, if "age" and "age" or "year of birth" are identified as synonyms by the large model, the name similarity score can be improved.

[0158] (2) Text semantic similarity The semantic similarity of data item definitions is calculated using semantic embedding vectors generated by a large model; the cosine similarity formula is used to compare the semantic embedding vectors of two data items. Perform the calculation:

[0159]

[0160] in," " represents the vector dot product, It is the Euclidean norm of the semantic embedding vector. The value ranges from [0,1]; the larger the value, the closer the two definition texts are semantically. With the help of deep semantic understanding of large models, this similarity not only captures explicit keywords, but also reflects the degree of conceptual similarity implied in the definition sentences.

[0161] (3) Attribute similarity Compare the structured attribute values ​​of two data items, including data type, length limit, unit, and value range, and calculate the degree of matching of these attributes; let... For all considered attribute sets (e.g., {data type, unit, length, allowed value range, ...}), the attribute similarity is defined as a weighted average of the similarities of each attribute (note: if some attributes are missing, only common attributes are calculated):

[0162]

[0163] in, Represents data items In attributes The value on, It is an attribute Importance weights, satisfying , |P| is the number of attributes in the attribute set P, assuming that each attribute has equal weight or that the primary key attribute is highlighted; It is a similarity function of attribute values; for example, for the "data type" attribute, When the types are the same (both are integers, etc.), and the types are different... (Or set partial similarity, such as considering integers and long integers as similar to 0.8); For the "unit" attribute, if there is a conversion relationship, it can be defined. (For example, meters and centimeters, although different in characters, are convertible and considered equivalent units), otherwise 0; for the "range" attribute, similarity can be calculated by the interval overlap rate (e.g., one standard allows 0-100, another allows 0-120, the proportion of the interval overlap to the total range is used as the similarity). For example, the "code value set" can be compared using Jaccard similarity to identify differences between sets.

[0164] By comparing attributes one by one and weighted summation, the similarity between two data items defined in terms of structural constraints is quantified.

[0165] S4. Conflict Detection and Type Identification: Define and compare attributes and values ​​for aligned data items, identify conflict points and classify conflict types to provide a basis for resolution decisions;

[0166] The preliminary aligned knowledge graph generated in step S3 includes all candidate matching data item pairs, their similarity, attribute information, etc. Conflicts may occur at this point, such as a data item from one standard having high similarity to multiple data items from another standard (one-to-many conflict), or two data items matching but contradicting each other in attributes (definition conflict). These conflict points need to be detected and categorized.

[0167] Technical means employed: S4 includes:

[0168] For candidate matching relationships in the knowledge graph, perform consistency checks and conflict pattern identification:

[0169] A method combining rule-based algorithms with large-scale model-assisted analysis is used to determine whether semantics are truly conflicting or merely differing in expression.

[0170] The method of combining rule-based algorithms with large-scale models for analysis includes: using predefined conflict determination rules to quickly screen for conflicts (such as one-to-many matching, attribute inconsistency), while calling large-scale models to compare and analyze the matching definition text;

[0171] By combining the two, the location and type of conflict can be accurately identified. Common conflict types include: ambiguity conflict (one-to-many, many-to-many matching), attribute inconsistency conflict (inconsistency in hard constraints such as data type or unit), and definition semantic conflict (contradictory descriptions or incompatible scopes).

[0172] Furthermore, the specific steps of S4 include:

[0173] S4.1 The input is the knowledge graph (nodes, edges, and similarity scores) from step S3. First, one-to-many matching conflict detection: scan the candidate equivalent edges of each node to check if there is a node that connects to multiple other nodes with similar similarity scores; for example, node X of standard A matches both Y and Z of standard B (both edges have high and similar Sim values), which is a one-to-many conflict. Record this conflict and label the relevant node set {X,Y,Z}. Similarly, many-to-many (multiple nodes with high similarity cycles) are also recorded. This type of conflict indicates that the matching relationship is uncertain and requires subsequent resolution decisions to select the best pairing.

[0174] S4.2 Next, for each pair of candidate matches Perform attribute consistency checks: compare the attribute values ​​of the two nodes to identify actual conflicts;

[0175] Define a conflict indicator function: if a certain attribute exists Make Furthermore, if the difference is not negligible, then the attribute is determined. Conflict; formally represented as:

[0176]

[0177] For example: if one data type is "integer" and the other is "string", then... (Incompatible); the units are "meter" and "centimeter," which are different but convertible. In this example, it can still be considered compatible. (This can be resolved through unit standardization); if one specifies a value range of 1-100 and another specifies 1-80, then there is a range conflict (one allows a value in the range [81,100] while the other does not). Summarize all attributes and define attribute conflict indicators:

[0178]

[0179] like The system records the attribute conflicts in the matching pair and lists the specific conflicting attribute categories. For each detected attribute conflict, the degree of conflict is also evaluated; for example, for numerical ranges, the range overlap rate or the magnitude of the difference can be calculated to quantify the severity of the conflict.

[0180] Next, semantic conflict checking is performed: even if attributes are identical, there may be subtle differences in the meaning of the definition texts of two standards. To address this, a large model is used to deeply compare the definition descriptions; for example, two matching definition texts are input into the large model, and the question "Do these two definitions describe the exact same concept? What are the differences?" is asked, allowing the large model to provide analysis. If the large model answers by pointing out certain semantic differences (e.g., "Definition A includes a specific business rule, while B does not mention it"), a semantic conflict is identified. The large model can also assist in classifying conflict types: for example, if the answer is "Definition A is more stringent than Definition B," it is marked as a "scope inconsistency" conflict. In this way, descriptive differences that are difficult for manual rules to detect are captured.

[0181] The S4 function organizes a list of conflict detection results. Each conflict record includes: the data item node involved (and its source), the conflict type (one-to-many, attribute inconsistency, semantic difference, etc.), specific manifestations (e.g., "data type inconsistency: integer vs. string" or "large model determines semantic range inconsistency"), and a conflict severity index. This list provides an operational checklist for the next step of conflict resolution. Regarding dependencies, only by first identifying the specific conflict can we select a targeted resolution strategy. For example, one-to-many conflicts require trade-offs in matching, while attribute conflicts may be resolved by adjusting values.

[0182] This step uses the alignment results generated in step 3 as input and checks them using the attribute and semantic information extracted in step 2. The accuracy of conflict detection directly affects the formulation of the resolution plan: sufficient detection ensures that each conflict point is accurately handled. The results of this step will guide step 5 in selecting appropriate conflict resolution measures.

[0183] S5. Globally optimized conflict resolution algorithm for conflict resolution and unified definition generation: Design a globally optimized conflict resolution algorithm to resolve conflicts; dynamically adjust matching relationships or attribute values ​​to generate a unified solution and output the fused data standard definition.

[0184] The conflict list identified in step 4, along with the complete aligned knowledge graph data, is used. The goal is to resolve these conflicts simultaneously on a global scale, outputting a consistent, multi-source fusion set of data standard definitions. This requires optimizing and adjusting previous candidate matching relationships, and merging or discarding differing definitions. The final result is a unified knowledge graph (or an equivalent set of data standards) that has eliminated contradictions and conflicts.

[0185] Further, S5 includes:

[0186] Design a globally optimized conflict resolution algorithm, combining intelligent optimization methods such as simulated annealing to search for the optimal alignment and fusion scheme; the globally optimized conflict resolution algorithm takes the candidate alignment relationship obtained by S3 as the initial solution, and through random perturbation and stepwise optimization, seeks the best in the solution space to maximize the overall similarity and minimize the conflict.

[0187] Simultaneously, a dynamic weight adjustment mechanism is introduced to adjust the similarity weighting coefficient in real time based on conflict situations during the optimization iteration process. Gradually increase the penalty for conflicting terms to guide the algorithm away from local optima;

[0188] Furthermore, specific strategies are incorporated into the conflict resolution process: value fusion (such as unit conversion and range expansion) is applied to attribute conflicts; selection or splitting is performed during matching for one-to-many conflicts; a unified definition text is generated by calling a large model; differences among multiple standards are balanced; and high-quality fusion results are output. The entire process demonstrates superior performance compared to traditional static rules or greedy matching, automatically balancing differences among multiple standards and outputting high-quality fusion results.

[0189] Furthermore, the specific steps of S5 include:

[0190] The input consists of a conflict list and the current matching graph state (including all candidate matches and similarity scores); the globally optimized conflict resolution algorithm is initialized first:

[0191] Establish the objective function for the optimization problem; define the objective function. Measured by "overall matching quality" minus "conflict cost", maximizing Objective function Formalized as:

[0192]

[0193] in, Represents a selected set of matching pairs (i.e., the final set of equivalence relations selected from candidate edges), the first item It is the total similarity score of all selected matches, the second item. It is an attribute conflict penalty (for each selected pair of matches, a penalty is deducted if there is an attribute conflict). , (The weighting coefficients reflect the severity of our penalty for attribute inconsistencies); the third item. It is the number of matches in the set that violate the one-to-one rule, such as a node matching multiple nodes, multiplied by a large coefficient. Severe penalties are imposed (to ensure that the final result tends towards a one-to-one match). Ideally, through such goal design, The maximum value is achieved under the optimal matching scheme, at which point there are almost no conflicts (the conflict cost term approaches zero), and the matching... The goal is to maximize the sum (preserving as many high-similarity matches as possible). The algorithm uses the current candidate matching set as the initial solution. (This may contain conflicts);

[0194] Then, the iterative optimization phase begins, employing a simulated annealing strategy to traverse the solution space. Specific operations include:

[0195] (1) Define neighborhood operation: randomly select a method for the current matching set Modifications are used as new solution candidates; for example: for one-to-many conflicts, randomly cancel one of the matching edges (remove a certain match); for unmatched node pairs, try to add them as new matches; or randomly make swap adjustments between two different matching pairs (for example, node X of standard A originally matched Y of B, Z of standard A matched W of B, the neighborhood operation may swap to X–W, Z–Y matching) [This operation is particularly effective when there are cross-matching errors, simulating the process of human swapping matches;

[0196] (2) Design neighborhood operations for attribute conflicts. For example, if two matches conflict in terms of units, try to treat them as compatible after converting the units as a state (equivalent to adjusting the attributes to make the conflict disappear). These operations generate different new solutions. ;

[0197] (2) Evaluation and acceptance criteria: Calculate the new scheme objective function ;Compare ;

[0198] if If the new solution reduces conflicts and / or improves matching scores, it will be accepted unconditionally. As the current solution;

[0199] if (If the new solution deteriorates in some aspect), then according to the Metropolis criterion of simulated annealing, the inferior solution is accepted with a certain probability in order to escape the local optimum.

[0200] The specific acceptance probability depends on the current "temperature" parameter. For example: when hour, Initial temperature of the law The temperature is relatively high, which may allow for the acceptance of some poorer solutions to explore more combinations; the temperature is gradually reduced as iterations proceed. When the algorithm converges, it only retains the better solution. Temperature updates use a cooling scheduling function, such as... Each round of decay, or the temperature is reduced after a certain number of simulated annealing cycles.

[0201] (3) Dynamic weight adjustment: During the iteration process, the similarity weighting coefficient is adjusted periodically based on the remaining conflict situation. etc.) or the penalty coefficient of the objective function ( For example, in the initial stage, in order to match semantically similar items as much as possible, The (text semantic weight) can be set relatively large, causing the algorithm to tend to connect nodes with high semantic similarity; once the main matching is established, the attribute weight is gradually increased during the refinement stage. Strengthen the requirements for consistency in data types, etc., and avoid overly lenient matching;

[0202] The dynamic adjustment function is based on the current conflict rate: Let Indicates the attributes in the current matching set The proportion of conflicts, then update the attributes. Importance weight The updated weights are:

[0203]

[0204] in This is the adjustment coefficient; It refers to the importance weight of attribute 𝑝 in the current iteration, representing the current assessment of the importance of the attribute when it was first assigned to this iteration; This means that in the next iteration, the importance weight of attribute p will be updated to this value;

[0205] This formula shows that if a certain attribute conflicts frequently ( If the conflict is large, its weight is increased, prompting subsequent iterations to pay more attention to the consistency of this attribute; conversely, if the conflict has decreased, the weight changes little or even decreases relatively. Through multiple rounds of adjustments, the system gradually transitions from "semantic matching as the main focus" to "strict consistency," and the final output match ensures both semantic rationality and complete compatibility in terms of attributes.

[0206] (4) Iterative convergence and optimal solution generation: As the temperature decreases and the weights are adjusted, the algorithm will gradually converge to a stable matching scheme. ;at this time Once the value reaches its maximum or no longer significantly improves, and all conflicting items are either eliminated (resolved by changing matching or adjusting attributes) or reduced to an acceptable range (e.g., some small differences that cannot be completely eliminated are recorded but do not affect overall consistency), the final set of matches is output. ;

[0207] Based on this final alignment result, we construct a unified knowledge graph: merging matched nodes into a single concept node (or establishing a mapping table to indicate their equivalence). For each set of equivalent data items, we determine the final standard definition value to be adopted: if attributes differ, they are merged according to a certain strategy. Common strategies include credibility priority (referring to the data source credibility mentioned in step 4: selecting the value of the authoritative standard) or union preservation (e.g., taking the union of the value ranges of each standard to encompass all), and majority rule (if a few differences exist among multiple sources, the majority consensus value prevails). For example, for unit conflicts, a unified conversion to a certain standard unit is used; for value range conflicts, a wider range may be taken to cover the requirements of all parties (or labeled as multiple sub-constraints of the concept). If a unified definition description text needs to be generated during the merging process, the large model can be invoked again: input the multi-source definitions of the same group, and let the model write a comprehensive version. For example, prompting: "Please merge the following multiple definitions into a unified definition" and attaching the original definition list, the large model will produce a complete new definition, striving to cover the key points of each standard and eliminate contradictions. If the definitions generated by the model are not ideal, constraints can be imposed on the prompts based on the attribute fusion results (e.g., explicitly specifying the value range of the new definition). This ultimately yields a standardized definition for each unified concept.

[0208] Through the optimization process described above, the algorithm weighs all matches and conflicts globally, making it more efficient and comprehensive than manual pairwise resolution. The final output of this step is a set of fused data standard definitions (or a fused knowledge graph), where each data item concept is aligned with its source and all conflicts have been resolved, resulting in a consistent definition. The specific output can be an updated knowledge graph (a graph with merged equivalent nodes and unified attributes) and a corresponding data standard document (listing unified data item names, definition descriptions, data types, etc.). Furthermore, the system can generate a conflict resolution report, recording which data items were merged from which sources and how existing differences were handled, providing users with change tracking.

[0209] The conflict resolution scheme proposed in this step overcomes the problem of greedy matching easily getting trapped in local optima through simulated annealing global optimization, and can find a better overall solution in complex multi-source matching. Simultaneously, by combining large-scale model semantic understanding and knowledge graph attribute constraints, it achieves conflict handling that combines "soft semantics" and "hard rules," making it more intelligent than traditional methods based solely on fixed rules (such as simple priority or majority voting). The mechanism of dynamically adjusting similarity weights allows the algorithm to focus on different aspects at different stages, ensuring that potential matches are not missed in the early stages while maintaining rigorous consistency in later results. This adaptive process is a major innovation of this scheme. In terms of conflict resolution decision-making, it not only considers which value to choose but also introduces operations such as attribute transformation and definition fusion (e.g., automatic unit conversion, text synthesis), further improving the quality and applicability of the fusion results. The final output unified data standard definition is free of semantic and format conflicts, providing a standardized data foundation for business applications. The entire technical process is logically rigorous and the steps are continuous. Any conflict can be effectively identified and resolved technically through the above methods, ensuring the successful implementation of multi-data standard fusion.

[0210] The improvement and innovation of this invention lies in the organic combination of semantic understanding of large models and structured reasoning of knowledge graphs, realizing fully automated processing of multi-data standard definition conflicts from detection to resolution. Compared with existing technologies, this solution can: 1) automatically understand the meaning of standard definitions and identify conflicts that are difficult for humans to detect; 2) provide intelligent conflict solutions, rather than just prompting conflicts; and 3) significantly reduce human intervention and improve the consistency of cross-standard data fusion. Current existing technologies cannot effectively handle complex semantic conflicts between cross-domain and cross-format standard definitions. For example, existing solutions mostly remain at the detection level or rely on manually formulated mapping rules, making it difficult to cope with dynamically added standard differences. This invention proposes a novel solution to address these shortcomings.

[0211] Simulation Experiment: To further verify the effectiveness of the method of the present invention, a simulation experiment was designed. In this experiment, three virtual data standards were constructed, each containing approximately 50 data elements, and a certain degree of redundancy and conflict was artificially set between these standards.

[0212] Specifically, this invention randomly generates a batch of general data items (such as name, gender, age, address, telephone number, company, code, etc.) and their different definition versions, and assigns them to three standards. Some of these data items have different names in different standards but refer to the same concept (equivalent to the aforementioned synonym conflict, such as "residential address" and "address"), while other data items have the same name but different actual meanings (equivalent to the aforementioned anisotropic conflict, such as "employee number" referring to internal employee ID in standard 1, while referring to external partner ID in standard 2).

[0213] A total of 20 pairs of conflicting data items were injected, covering various types such as inconsistent formats, inconsistent value ranges, and confusing naming.

[0214] This invention uses the method of this invention and two comparative methods to perform conflict detection and resolution on the above-mentioned simulated data: Comparison Method 1 is based solely on simple mapping using string matching rules (without utilizing a semantic model); Comparison Method 2 uses semantic similarity to detect conflicts but does not perform global optimization (resolving conflicts pair by pair without considering overall constraints). Evaluation metrics include the accuracy and recall of conflict identification, as well as the effectiveness of final conflict resolution (the proportion of remaining conflicts).

[0215] Experimental results show that the method of this invention achieves the highest accuracy and recall in the conflict identification stage: among 20 pairs of simulated implanted conflict data items, this method successfully identified 18 pairs, with the 2 pairs missed being those with extremely obscure definitions; the string matching method only identified 10 pairs, missing most conflicts with different semantic expressions; the semantic detection method without optimization identified 15 pairs, but failed to detect some conflicts requiring global trade-offs. Entering the conflict resolution stage, this method correctly merges semantically equivalent definitions through global optimization and breaks down the concepts that need to be distinguished, ultimately reducing the remaining conflict indicators in the knowledge graph. Only one conflict (a few requiring manual decision-making remained unresolved), achieving a conflict resolution rate of 95%. In contrast, the string matching method, lacking semantic analysis, incorrectly treated some items that should not be merged as identical, resulting in approximately 40% of conflicts remaining unresolved or introducing new contradictions after resolution. While the semantic method without global optimization avoided obvious mismatches, the lack of consistency in individual decision-making meant that approximately 20% of conflicts remained unresolved. Overall, the method of this invention achieved the highest conflict resolution completion and accuracy in the simulated data environment. Furthermore, the execution efficiency of the method was evaluated. With the aforementioned data scale (approximately 150 data items, 20 conflict pairs), the system's semantic analysis and knowledge graph construction were completed within seconds, and the simulated annealing optimization process converged and obtained a stable solution within tens of seconds. Considering the typically offline batch processing nature of data standard adjustment, this time consumption is entirely acceptable. If the data scale is further expanded, the method of this invention can also accelerate processing through parallel computing and appropriately reducing the number of simulated annealing iterations. Therefore, the method of the present invention has a certain degree of scalability while ensuring effectiveness, and can be applied to large-scale public data standard conflict governance tasks.

[0216] Experiments have verified the effectiveness of this invention: it can efficiently detect complex conflicts and provide reasonable resolution solutions, significantly improving the consistency and fusion quality of multi-source data. The performance improvement obtained from the above simulation experiments is consistent with other research results we cite, such as a more than 50% increase in recall rate for knowledge conflict resolution, proving the advanced nature and reliability of the present invention. It should be noted that the above embodiments and simulation results are used to illustrate the method of this invention, and any matters not covered herein can be flexibly adjusted within the scope of protection of this invention. For example, the choice of large model can be replaced according to the application domain and resource conditions: general domains can use general large models such as the DeepSeek series, while specialized domains can preferentially select domain pre-trained models or knowledge-enhanced models; when the model is temporarily unavailable, traditional ontology dictionary methods can be combined to assist semantic recognition, although the effect is relatively reduced. This invention uses simulated annealing algorithm for conflict optimization, but this is not the only option. Other heuristic algorithms such as genetic algorithms and tabu search can also be used, or the conflict resolution model can be converted into integer linear programming and solved precisely by an optimization solver. When the data scale is small, precise solution can obtain the globally optimal solution; when the data volume is large, heuristic algorithms can better balance efficiency and effect. In terms of knowledge graph implementation, semantic web standards such as RDF / OWL can be used to represent standard knowledge data, enabling integration with existing standardized platforms and ontology libraries. Semantic similarity calculations can also incorporate prior knowledge provided by domain ontology or dictionaries (such as WordNet and thesaurus) to improve accuracy. Furthermore, similarity thresholds in conflict detection and weight parameters in optimization models can be adjusted and optimized according to actual needs.

[0217] Example 2: A method for resolving conflicts in multiple data standard definitions based on large models and knowledge graphs. Assume there are two data standards, A and B, defined by different departments, containing the following corresponding data items:

[0218] Standard A:

[0219] Address: Defined as a complete residential address string (including all elements such as province, city, district, and street).

[0220] Population: Defined as the number of permanent residents within the jurisdiction (including local registered residents and non-registered residents holding long-term residence permits).

[0221] Standard B:

[0222] Address: Defined as the resident's residential address information, stored in three fields: "Province / City / Detailed Address".

[0223] Total population: defined as the number of registered residents within the jurisdiction (including only residents with local household registration).

[0224] In standards A and B mentioned above, "address" and "residence address" have essentially the same meaning, both referring to a resident's place of residence. However, standard A treats "address" as a single text field, while standard B breaks it down into multiple fields, resulting in differences in formatting. Furthermore, while both "population" in standard A and "total population" in standard B refer to the population of a certain area, the former includes non-registered long-term residents, while the latter only counts registered residents, creating a conflict in their statistical scope. These two examples reflect common inconsistencies in standards used in public data governance: synonyms with different forms (different names but the same meaning, such as address / residence address) and homonyms with different meanings (same or similar names but different meanings, such as population / total population). Regarding the above data standards A and B, the method of this invention performs conflict detection and resolution step by step:

[0225] S1. Semantic Recognition. A pre-trained large model was used to analyze the definition texts of Standard A and Standard B. The model output showed that "address" and "residence address" had a high semantic similarity score (close to 1), indicating they represent the same concept; "population" and "total population" also had high semantic similarity, leading the model to believe they refer to a highly related population concept. However, through analysis of the definition details, the model noticed that the definition of "total population" only includes "registered residents," while Standard A's "population" covers non-registered permanent residents, suggesting a semantic difference. After the semantic recognition stage, we obtained the semantic vector representation of each data item, as well as the initially identified relationships: {address ≈ residence address}, {population ≈ total population}.

[0226] S2. Knowledge Graph Construction. Based on the semantic recognition results, a cross-standard knowledge graph is constructed. The system creates nodes representing the "address" of standard A and the "residence" of standard B, and establishes a same-as equivalence relationship between them; simultaneously, an abstract node "address (concept)" is introduced to connect the "address" and "residence" nodes, indicating that they belong to the same address concept. For "population" and "total population," the system also establishes a same-as relationship and connects the two with the abstract node "population (concept)," indicating that they are considered the same "population" concept. At this point, two conflicting nodes are formed in the knowledge graph: one is an "address" conflicting node (connecting the address of standard A and the residence of standard B), and the other is a "population" conflicting node (connecting the population of standard A and the total population of standard B). In addition, the graph also records an inclusion relationship: the "total population" data of standard B is actually contained in the population data of standard A (registered population is a subset of resident population), therefore, a "total population ⊂ population" relationship is added to the knowledge graph for reference.

[0227] S3. Conflict Detection. The system performs in-depth analysis of each conflict node in the knowledge graph, extracting the specific attributes of each relevant data item. For example, for the "Address" conflict node, it finds that the address in Standard A is a single string field, while the address in Standard B consists of multiple fields. Although the data type is the same (text), the structural format is different, which is determined to be a format conflict. For the "Population" conflict node, comparing the population definition attributes of Standard A and Standard B, the units of measurement are the same (number of people), and the data type is integer, but the defined statistical ranges are different—Standard A's population includes non-registered residents, while Standard B's total population only includes registered residents, which is a range conflict. Therefore, the system marks the "Population" conflict as a range inconsistency conflict and assigns it a higher conflict weight (because semantic range differences directly affect data statistics); correspondingly, the "Address" conflict is assigned a lower weight because it only involves format conversion. At the same time, the system notices that although the names "Population" and "Total Population" are different, their meanings are similar. This is a difference in terminology rather than naming ambiguity, so no separate naming conflict is marked. After conflict detection, the system found a total of 2 conflicts: 1 address format conflict and 1 population statistics caliber conflict.

[0228] S4. Conflict Resolution. The system establishes an optimization model for the above conflicts and runs a simulated annealing algorithm to search for the optimal solution. For "address" conflicts, since the two fields are semantically identical and only differ in format, the algorithm tends to merge them by unifying the format. Specifically, the algorithm initially assumes no conversion, which leaves conflict costs; in subsequent iterations, it attempts to eliminate format conflicts by "splitting the address of standard A" or "concatenating the address of standard B". After format conversion measures are taken, the "address" conflict... The objective function changes from 1 to 0. The corresponding reduction occurs. For "population" conflicts, the algorithm faces two strategies: merging (treating both as the same concept and attempting a unified definition) or splitting (treating them as different concepts and retaining them separately). If merging is chosen, concessions must be made in the definition, such as expanding the scope of registered population or labeling data sources; otherwise, the conflict cannot be completely eliminated. Splitting means no longer forcibly aligning the two indicators, but acknowledging their different statistical methods and independent nature. The simulated annealing algorithm randomly tried both schemes during the high-temperature phase and calculated the conflict cost: it found that the merging scheme always retained a certain conflict cost, while the splitting scheme reduced the conflict cost to zero. As the temperature decreased, the algorithm gradually converged to choosing to split the "population" concept: removing the equivalence between standard A "population" and standard B "total population," treating them as different concepts. Finally, the simulated annealing algorithm outputs the following conflict resolution scheme: standardizing the address data format and differentiating the statistical methods for population data. In other words, the address fields of Standard A and Standard B will adopt a unified storage format or provide conversion mapping to ensure format consistency; while the two indicators "population" and "total population" will maintain their respective definitions and will no longer attempt to force alignment, thereby avoiding semantic conflicts.

[0229] S5. Standard Updates and Feedback. Based on the above resolution plan, Standard A and Standard B were adjusted and configured accordingly. First, regarding the address field, both parties agreed to adopt the "three-segment" storage structure of Standard B as the standard format, as splitting the stored information provides richer information and facilitates comparison. Standard A accordingly added a structure to the address field, or provided a conversion program to split a single address string into the standard format during data exchange. In this way, the address data of the two standards can be directly mapped to each other, eliminating format conflicts. Second, regarding population indicators, it was decided to rename "Population" in Standard A to "Resident Population" in the system, and to clearly define "Total Population" in Standard B as "Registered Population," with the difference in their scope clearly noted in the metadata. After this update, the population indicators of the two standards are no longer confused: in the knowledge graph, "Resident Population" and "Registered Population" exist as different nodes, and are related by the relationship "Registered Population ⊂ Resident Population"; the original conflicting "Population" nodes were removed. Finally, the data governance team fed back these changes to the relevant standard-setting departments as the basis for subsequent standard revisions. Simultaneously, the system records the decision-making process for resolving this conflict, especially the judgment in the simulated annealing algorithm regarding splitting rather than merging the "population," and stores this information in the conflict rule base. If similar conflicts involving "partial overlap" arise in the future, the system can refer to this experience for faster decision-making. Through the above implementation, the method of this invention successfully eliminated two major data definition conflicts between standards A and B. Objective function The fact that the initial non-zero value decreased to 0 proves that the solution has achieved the goal of conflict resolution.

[0230] Example 3: This invention also provides a conflict resolution system based on large models and knowledge graphs for multiple data standard definitions, the system comprising:

[0231] The data acquisition and preprocessing module is used to extract data item definitions from multiple data standard documents and perform cleaning and normalization processing.

[0232] The semantic representation and large model parsing module is used to perform semantic analysis on the definition of data items using a large model, and generate embedding vectors and key features.

[0233] The knowledge graph construction and entity alignment module is used to build a knowledge graph containing all data items, and uses an entity alignment algorithm to connect semantically similar data item nodes from different standards.

[0234] The conflict detection and resolution optimization module is used to identify conflict points in the aligned data item definitions and resolve them through optimization algorithms to generate a unified solution.

[0235] The standard update and feedback monitoring module is used to feed the optimization results back to the original standard document, establish a continuous feedback and monitoring mechanism, monitor the use of the standard, and promptly identify new conflicts.

[0236] Furthermore, the data acquisition and preprocessing module includes: a data acquisition module, a data cleaning module, and a data normalization module;

[0237] The data acquisition module supports the import of various data formats (such as JSON, XML, CSV, etc.);

[0238] The data cleaning module is used to remove redundant information, fill in missing values, and standardize data format.

[0239] The data normalization module is used to convert data item definitions from different sources into a unified representation.

[0240] The semantic representation and large model parsing module includes: an embedding generation module, a feature extraction module, and a semantic enhancement module;

[0241] The embedding generation module is used to generate semantic embedding vectors for data items using large models (such as BERT, RoBERTa);

[0242] The feature extraction module is used to extract key features from the embedded vectors for subsequent entity alignment.

[0243] The semantic enhancement module is used to enhance the semantic representation of data item definitions by combining contextual information;

[0244] The knowledge graph construction and entity alignment layer module includes: a knowledge graph construction module, an entity alignment module, and an alignment relationship generation module;

[0245] The knowledge graph construction module is used to build knowledge graphs based on data items and their relationships.

[0246] The entity alignment module is used to encode the knowledge graph using graph neural networks (such as GCN and GAT) to identify potential alignment entities;

[0247] The alignment relationship generation module is used to generate alignment relationships and form candidate alignment data item pairs.

[0248] The conflict detection and resolution optimization module includes: a conflict detection module, a directory function scoring module, an optimization algorithm module, and a unified definition generation module;

[0249] The conflict detection module is used to compare attributes and values, identify defined conflict points, and classify and label conflict types.

[0250] The directory function scoring module is used to design multi-dimensional directory functions, taking into account factors such as definition similarity, usage frequency, and contextual consistency to score conflicting nodes.

[0251] The optimization algorithm module is used to apply the simulated annealing algorithm to search for the optimal conflict resolution solution and dynamically adjust the matching relationship or attribute value.

[0252] A unified definition generation module is used to generate the integrated data standard definition and output a unified solution;

[0253] The standard update and feedback supervision module includes: a standard update module, a feedback supervision module, and a version management module;

[0254] The standard update module is used to feed the optimization results back to the original standard document and update the standard definition;

[0255] The feedback and monitoring module is used to introduce a continuous feedback and monitoring mechanism to monitor the use of standards and promptly identify new conflicts.

[0256] The version management module is used to manage different versions of the standard, record change history, and support retrospection and auditing.

[0257] The specific embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

Claims

1. A conflict resolution method based on large models and knowledge graphs for multiple data standard definitions, characterized by: The method includes: S1. Extraction and preprocessing of multiple data standard definitions: Extract data item definitions from various data standard documents, clean and standardize their representation; S2. Define semantic representation and large model parsing: Use the large model to perform semantic analysis on the definition of each data item, and generate semantic embedding vectors and key features; S3. Knowledge Graph Construction and Entity Alignment: Construct a knowledge graph containing all data items, and use an entity alignment algorithm to connect data item nodes with similar semantics in different standards to form candidate alignment relationships; S4. Conflict Detection and Type Identification: Define, compare attributes and values ​​for aligned data items, identify conflict points and classify and label conflict types; S5. Globally optimized conflict resolution algorithm for conflict resolution and unified definition generation: Design a globally optimized conflict resolution algorithm to resolve conflicts; dynamically adjust matching relationships or attribute values ​​to generate a unified solution and output the fused data standard definition; The specific steps of S5 include: The input consists of a conflict list and the current matching graph state; the globally optimized conflict resolution algorithm is first initialized: Establish the objective function for the optimization problem; define the objective function. Measured by "overall matching quality" minus "conflict cost", maximizing Objective function Formalized as: ； in, Represents a selected set of matching pairs, the first item It is the total similarity score of all selected matches, the second item. It's an attribute conflict penalty; the third item. It is the number of matches in the set that violate the one-to-one rule; Then, the iterative optimization phase begins, employing a simulated annealing strategy to traverse the solution space. Specific operations include: (1) Define neighborhood operation: randomly select a method for a set of selected matching pairs The modification is used as a candidate for a new solution; Design neighborhood operations to address attribute conflicts, which generate different new solutions. ; (2) Evaluation and acceptance criteria: Calculate the new scheme objective function ;Compare ; if Then accept unconditionally. As the current solution; if If the solution is accepted with a certain probability according to the Metropolis criterion of simulated annealing, it can escape the local optimum. The specific acceptance probability depends on the current "temperature" parameter. ; (3) Dynamic weight adjustment: During the iteration process, the similarity weighting coefficient or the objective function penalty coefficient is adjusted periodically according to the remaining conflict situation; The dynamic adjustment function is based on the current conflict rate: Let Indicates the attributes in the current matching set The proportion of conflicts, then update the attributes. Importance weight The updated weights are: ； in This is the adjustment coefficient; This refers to the attribute in the current iteration. The importance weight represents the current assessment of the importance of the attribute when it is first assigned to this iteration. This means that in the next iteration, the importance weight of attribute p will be updated to this value; (4) Iterative convergence and optimal solution generation: As the temperature decreases and the weights are adjusted, the algorithm will gradually converge to a stable final set of matching relationships. Output the final set of matching relationships. ; Based on the final alignment results, a unified knowledge graph is constructed.

2. The method for resolving conflicts based on large models and knowledge graphs using multiple data standard definitions according to claim 1, characterized in that: S1 includes: Processing multiple data standard documents and performing unified parsing, which contain several data items and their definitions; processing multiple data standard documents and performing unified parsing, including: Using text parsing and format conversion technologies, standard documents from different sources are transformed into structured data representations; custom parsing scripts are written to extract "data item names", "definition text", and related attribute fields; if the document format is inconsistent or the definition is buried in the natural language description, the large model API is called to assist in parsing complex text, including enabling the large model to identify key fields in the definition sentence; S1.1 First, input the various data standard documents and load them into the parsing tool; For structured formats, read the fields directly; For unstructured text, use regular expressions or the NLP capabilities of large models to locate the "data item name: definition..." pattern, and then process and extract the name and definition content; If the definition contains attribute information, the attribute name and attribute value will also be parsed out. S1.2 Next, perform data cleaning and standardization: unify the terminology used from different sources and fill in any missing information; during this process, call the large model to simply rewrite or complete the definition text; output a standardized list of data item definitions or an initial knowledge base: containing a structured record of each data item.

3. The method for resolving conflicts of multiple data standard definitions based on large models and knowledge graphs according to claim 1, characterized in that: S2 includes: Use large-scale LLM or large-scale pre-trained models to define natural language processing for each data item; On the one hand, the data item definition is converted into a high-dimensional semantic vector representation through the embedding interface of the large model; On the other hand, the parsing capabilities of large models are utilized to extract key semantic features from the data item definitions; including: Use a pre-trained sentence vector model to generate embedding vectors for the definition text or directly call the large model API to request the output of the defined structured information, which includes semantic summary and a list of synonyms; The specific process includes: First, the "data item name" and "definition text" are concatenated into a complete descriptive sentence, which is then used as the model processing text. The processing is then divided into two parts: (1) Semantic embedding vector generation: Using the vectorization function provided by the large model, the definition of each data item is sent into the large model to obtain the semantic embedding representation; (2) Key feature extraction: Utilize the question-answering or parsing capabilities of large models to extract key attributes and semantic tags from the data item definitions; The specific approach involves constructing a prompt query model; the model outputs a structured result based on the trained knowledge; the model supplements the implicit attributes in the plain text definition; and the semantic labels and normalized attribute values ​​defined for each data item are obtained.

4. The method for resolving conflicts of multiple data standard definitions based on large models and knowledge graphs according to claim 1, characterized in that: S3 includes: A knowledge graph is constructed to represent multi-source data standards, and an entity alignment algorithm is used to discover identical or equivalent data items in the graph. The knowledge graph uses nodes to represent data items and edges to represent potential equivalence relationships between data items of different standards. During the alignment process, multidimensional similarity is comprehensively considered; specifically, a fusion similarity calculation method is used, a custom similarity formula is defined, and a threshold is set to judge candidate matches; at the same time, a combination of heuristics and algorithms is adopted: first, preliminary corresponding candidates are established by name or keywords, and then the matching score is accurately calculated by cosine similarity; for complex matching problems, graph algorithms are introduced to find the globally optimal matching scheme; and candidate alignment relationships are generated.

5. The method for resolving conflicts based on large models and knowledge graphs using multiple data standard definitions according to claim 1, characterized in that: The specific steps in S3 include: S3.1, The input is the node representation of each data item; first, construct the knowledge graph: A subgraph is built for each data standard, and all data items are placed as nodes in the graph; each node is associated with an attribute and also carries a semantic embedding vector. ; S3.2 Next, calculate the cross-standard node similarity; for any two data item nodes from different standards... and Define comprehensive semantic similarity It integrates multiple factors: similarity of names and terms, semantic similarity of definition text, and similarity of attribute values; and comprehensive semantic similarity. The calculation formula is as follows: ； in, It is the similarity of names and terms. It defines the semantic similarity of text. It is the similarity of attribute values. For the weighting coefficients, satisfying These weights are used to balance the importance of each part, and are adjusted based on experience or the training set. S3.3 Calculate all cross-standard node pairs of Then, determine the candidate alignment relationship: when Value higher than the set threshold Mark the two data items as a "candidate match pair"; In a knowledge graph, a temporary "equivalent" edge is added between each pair of candidate matching data item nodes, and the edge weight is recorded. value; For cases with multiple standards, the process of constructing a knowledge graph is reduced to finding similar groups in a multi-partite graph; Output a preliminary fused knowledge graph: nodes are data items from various standards, and highly similar candidates across standards are connected by edges; each node is connected to multiple nodes from other standards.

6. The method for resolving conflicts based on large models and knowledge graphs using multiple data standard definitions according to claim 5, characterized in that: In S3.2, the specific calculation details of the comprehensive semantic similarity are as follows: (1) Similarity of names and terms Compare the similarity of data item names or terms; If the two names are exactly the same, let If the names are different but overlap after word segmentation or synonym normalization, the Jaccard coefficient is used to calculate the overlap of the word sets. The calculation formula is as follows: ； , This refers to the data item node. and The corresponding data item name, Tokens, is the "set of words" after text segmentation. It is the absolute value; At the same time, using the synonym list extracted by S2, synonymous terms are considered as matches; (2) Define text semantic similarity The semantic similarity of data item definitions is calculated using semantic embedding vectors generated by a large model; the cosine similarity formula is used to compare the semantic embedding vectors of two data items. Perform the calculation: ； in," " represents the vector dot product, It is the Euclidean norm of the semantic embedding vector. The value range is [0,1]; (3) Attribute value similarity Compare the structured attribute values ​​of two data items, including data type, length limit, unit, and value range, and calculate the degree of matching of these attributes; let... For the set of all considered attributes, the attribute similarity is defined as a weighted average of the similarities of all attributes: ； in, Represents data item nodes In attributes The value on, It is an attribute Importance weights, satisfying , |P| is the number of attributes in the attribute set P, assuming that each attribute has equal weight or that the primary key attribute is highlighted; It is a similarity function of attribute values; By comparing attributes one by one and weighted summation, the similarity between two data items defined in terms of structural constraints is quantified.

7. The method for resolving conflicts of multiple data standard definitions based on large models and knowledge graphs according to claim 1, characterized in that: S4 includes: For candidate matching relationships in the knowledge graph, perform consistency checks and conflict pattern identification: A method combining rule-based algorithms with large-scale model-assisted analysis is used to determine whether semantics are truly conflicting or merely differing in expression. The method of combining rule-based algorithms with large-scale model-assisted analysis includes: using predefined conflict determination rules to quickly screen for conflicts, while calling large-scale models to compare and analyze the matching definition text; By combining the two, the location and type of conflict can be accurately identified; common conflict types include: hit ambiguity conflict, attribute inconsistency conflict, and definition semantic conflict.

8. The method for resolving conflicts of multiple data standard definitions based on large models and knowledge graphs according to claim 1, characterized in that: The specific steps in S4 include: S4.1 The input is the knowledge graph from step S3; First, one-to-many matching conflict detection: scan the candidate equivalent edges of each node and check whether there is a node that connects to multiple other nodes with similarity scores; S4.2 Next, for each pair of nodes... Perform attribute consistency checks: compare the attribute values ​​of the two nodes to identify actual conflicts; Define a conflict indicator function: if a certain attribute exists Make Furthermore, if the difference is not negligible, then the attribute is determined. Conflict; formally represented as: ； For each detected attribute conflict, the degree of conflict is also evaluated; Then, define semantic conflict checking: use a large model to deeply compare the definition description; The S4 can organize a list of conflict detection results; each conflict record includes: the data item node involved, the conflict type, the specific manifestation, and the conflict severity index.

9. The method for resolving conflicts based on large models and knowledge graphs using multiple data standard definitions according to claim 1, characterized in that: S5 includes: Design a globally optimized conflict resolution algorithm that combines intelligent optimization methods to search for the optimal alignment and fusion scheme. The globally optimized conflict resolution algorithm takes the candidate alignment relationship obtained by S3 as the initial solution, and through random perturbation and stepwise optimization, it seeks the best solution in the solution space to maximize the overall similarity and minimize the conflict. Simultaneously, a dynamic weight adjustment mechanism is introduced to adjust the similarity weighting coefficient in real time based on conflict situations during the optimization iteration process. Gradually increase the penalty for conflicting terms to guide the algorithm away from local optima; In addition, specific strategies are incorporated into the conflict resolution process: value fusion is applied to attribute conflicts, selection or splitting is performed in matching for one-to-many conflicts, a large model is called to generate unified definition text, the differences between multiple standards are balanced, and high-quality fusion results are output.

10. The method for resolving conflicts of multiple data standard definitions based on large models and knowledge graphs according to claim 1, characterized in that: The construction of a unified knowledge graph includes: merging matched nodes into a single concept node or establishing a mapping table to indicate their equivalence; for each set of equivalent data items, determining the final standard definition value; if attributes differ, merging is performed according to a strategy; common strategies include prioritizing credibility or preserving the union. If a unified definition description text needs to be generated during the merging process, the large model is called again, and the definitions of the multi-source data items in the same group are input, so that the model can write a comprehensive version. If the definition generated by the large model is not ideal, the prompts are constrained based on the attribute fusion results; finally, a standardized definition for each unified concept is obtained.

Images