Multi-source data multi-dimensional auxiliary verification method based on knowledge graph

Abstract

The invention discloses a multi-source data multi-dimensional auxiliary verification method based on a knowledge graph. The method comprises the following steps: 1) constructing the knowledge graph of a target domain; 2) oriented by a modeled target entity in the knowledge graph, directionally converging description information related to the target entity from a plurality of different sources, and synchronously recording source identifiers and timestamps; 3) extracting a core element in each piece of description information and inputting the core element into a large language model for semantic vectorization to obtain a semantic vector of each piece of description information; 4) performing consistency verification on the multi-source data from three dimensions of content similarity analysis, entity map correlation analysis and keyword co-occurrence analysis; and 5) according to a consistency verification result, identifying and prompting inconsistent parts in the description information. According to the method and the device, the condition of missing detection caused by expression difference is greatly reduced, so that the accuracy of consistency verification is improved.

Description

Technical Field

[0001] This invention belongs to the field of information processing and knowledge graph construction technology, and relates to the technology of consistency verification and conflict identification of multi-source data, and particularly to a multi-dimensional auxiliary verification method for multi-source data based on knowledge graph. Background Technology

[0002] With the diversification of information sources (including open-source data, web crawler data, etc.), integrating data from different sources to obtain a complete and accurate data picture has become crucial. However, data reports from different sources often differ in content description and details, and may even contradict each other. Without effective verification, these differences and conflicts may lead to biased data analysis conclusions, affecting the reliability of decision-making.

[0003] Currently, data consistency analysis primarily employs methods based on keyword comparison and rule-based judgment. This means that consistency among multiple pieces of information is determined by simply matching keywords in reports or using pre-defined rules. However, such methods lack an understanding of the semantic level of information, only comparing surface-level text and failing to identify substantially identical information that has been paraphrased or used synonymously. When data from multiple sources differ in word choice or expression but are semantically related, keyword-based comparison often fails to detect implicit connections or conflicts. Furthermore, existing solutions cannot comprehensively assess multiple dimensions, including entity relationships, content semantics, and contextual relationships. For example, descriptions of the same system from different data sources may involve complex entity relationships and background context; relying solely on keyword rules cannot capture these deep-seated connections, leading to undetected potential conflicts between multi-source data.

[0004] In summary, existing technologies have significant shortcomings in multi-source data consistency verification, and there is an urgent need for a more intelligent and comprehensive method to solve the above problems. Summary of the Invention

[0005] To address the problems existing in the prior art, the purpose of this invention is to provide a multi-source data multi-dimensional auxiliary verification method based on knowledge graphs.

[0006] The main contents of this invention include:

[0007] 1. Introduction of large language models for semantic matching: This is the first time that large language models have been used for data consistency verification. Content matching analysis is performed by generating semantic vector representations of the data text, replacing the traditional approach of simply relying on keyword comparison. This significantly improves the understanding and utilization of semantic information in the data. The large language model used is either the Qwen2.5 or Qwen3 large language model, etc.

[0008] 2. Integration of RDF Knowledge Graph and Multi-Dimensional Validation Logic: This approach organically combines the RDF knowledge graph structure with multi-dimensional information validation methods such as content similarity, entity relationships, and keyword co-occurrence, achieving a high-dimensional cross-validation mechanism for data. The knowledge graph provides a structured view of relationships, enabling the system to examine the consistency of multi-source data from the perspective of entity networks, representing a fundamental difference and improvement over existing technologies.

[0009] 3. Supports Neo4j Visual Conflict Tracing: Leveraging the visualization capabilities of the Neo4j graph database, this solution can graphically trace the entities and source links related to the conflict after detecting a data conflict. This visual conflict tracing function helps users understand the specific context in which the conflict arose. It is an innovative feature in data analysis tools, enhancing the system's practicality and interpretability.

[0010] The technical solution of this invention is as follows:

[0011] A multi-source data multi-dimensional auxiliary verification method based on knowledge graphs, the steps of which include:

[0012] 1) Construct a knowledge graph for the target domain;

[0013] 2) Guided by the target entity already modeled in the knowledge graph, collect descriptive information related to the target entity from multiple different sources, and simultaneously record the source identifier and timestamp;

[0014] 3) Extract the core elements from each of the described information and input them into a large language model for semantic vectorization to obtain the semantic vector of each of the described information;

[0015] 4) Consistency verification of multi-source data is performed from three dimensions: content similarity analysis, entity graph association analysis, and keyword co-occurrence analysis;

[0016] 5) Based on the consistency check results, identify and highlight any inconsistencies in the description information:

[0017] 51) When two entities from different sources are mapped to the same entity node in the knowledge graph, the attributes of the two entities are determined to be contradictory based on the content similarity analysis results and the entity graph association analysis results. If there is a contradiction, the conflict relationship is marked under the entity node.

[0018] 52) When two entities from different sources are mapped to the same entity node in the knowledge graph, and the time attribute values ​​of the two entities are inconsistent in the description information from different sources, then a time conflict is marked under the entity node.

[0019] 53) When descriptive information from different sources shares the same keyword in keyword co-occurrence analysis, but the semantic direction of the keyword is contradictory in the descriptive information from different sources, then the node corresponding to the keyword in the knowledge graph is marked with a conflict relationship.

[0020] Preferably, the core elements include entity, technical keywords, time, location, and tool number.

[0021] Preferably, the core elements of the descriptive information are extracted by combining natural language processing technology and a domain dictionary.

[0022] Preferably, the content similarity analysis method is as follows: based on the semantic vectors of the two descriptive information pieces, the content similarity between the two descriptive information pieces is calculated. If the content similarity is greater than a set threshold, it is determined that the two descriptive information pieces involve the same entity. Then, the attribute values ​​of the entity in the two descriptive information pieces are extracted, and each attribute value extracted is compared item by item. If the attribute values ​​are consistent, it is determined that the two descriptive information pieces are consistent. The two descriptive information pieces are merged and mapped to the entity node corresponding to the entity in the knowledge graph as evidence of multi-source consistency of the entity. Otherwise, it is determined that the two descriptive information pieces are inconsistent, and the two descriptive information pieces are merged and mapped to the entity node corresponding to the entity in the knowledge graph and the conflict relationship is marked.

[0023] Preferably, the method for entity graph association analysis is as follows: mapping entities in the two descriptions to corresponding nodes in the knowledge graph, and then determining whether the entities in the two descriptions are the same or have a related path based on the knowledge graph; if the entities in the two descriptions are the same or have a related path, then detecting whether the attributes of the entity in the two descriptions are consistent; if the attributes are consistent, merging the two descriptions and mapping them to the entity node corresponding to the entity in the knowledge graph as evidence of multi-source consistency of the entity; if the attributes are inconsistent, storing and marking conflict relationships for the two descriptions under the entity node.

[0024] Preferably, the keyword co-occurrence analysis method is as follows: the TF-IDF method is used to calculate the keywords and their weights in the two description information, and the shared keywords in the two description information are counted; if the shared keyword has a contextual semantic similarity of approximately a set threshold and the semantic roles are consistent in the two description information, then the two description information are merged and mapped to the entity node corresponding to the shared keyword in the knowledge graph; otherwise, it is determined that the shared keyword has a semantic conflict, and the two description information are stored in the knowledge graph under the node corresponding to the shared keyword and a conflict labeling relationship is established.

[0025] Preferably, the large language model is a Qwen2.5 large language model or a Qwen3 large language model; a knowledge graph of the target domain is constructed using a resource description framework and stored in the Neo4j graph database.

[0026] Preferably, the target fields include, but are not limited to, the fields of intelligence information processing, scientific and technological intelligence analysis, and industrial equipment monitoring.

[0027] A server is characterized by comprising a memory and a processor, the memory storing a computer program configured to be executed by the processor, the computer program including instructions for performing the methods described above.

[0028] A computer-readable storage medium having a computer program stored thereon, characterized in that the computer program implements the above-described method when executed by a processor.

[0029] The advantages of this invention are as follows:

[0030] In-depth semantic understanding: By leveraging the Qwen large-scale language model to semantically vectorize data content, the system gains the ability to understand the deeper meaning of text. Compared to simple keyword matching, semantic analysis can identify data descriptions with different wording but similar substance, significantly reducing missed detections due to differences in expression, thereby improving the accuracy of consistency checks.

[0031] Multi-dimensional comprehensive analysis: This method simultaneously verifies data from three aspects: content, related entities, and keyword context, forming a complementary multi-dimensional analysis framework. This comprehensive evaluation approach can identify inconsistencies in the data from all angles, improving the comprehensiveness and reliability of conflict identification and avoiding potential biases that may arise from relying on a single dimension.

[0032] Intuitive Conflict Localization: By integrating data into a knowledge graph and visualizing it using Neo4j, this solution can intuitively display the relationships and conflict chains between different data sets. When data conflicts occur, analysts can directly see how related entities are connected, as well as the location of conflicting information points and their upstream and downstream context, on the graph interface. This intuitive presentation helps to deeply understand the causes and impacts of conflicts, facilitating subsequent human assessment and decision support.

[0033] Adaptable to multi-source heterogeneous data: This solution is designed for various sources such as open-source data and web crawler data, exhibiting excellent versatility and scalability. The system can automatically process data information of different formats and sources, uniformly mapping them to knowledge graphs and vector spaces to achieve cross-source consistency checks. This reduces the workload of manually organizing and comparing information from multiple sources, significantly improving data processing efficiency in today's environment of massive and rapidly updating data volumes. Attached Figure Description

[0034] Figure 1 This is a diagram showing the overall system architecture of the method of the present invention.

[0035] Figure 2 This is a flowchart of the method of the present invention.

[0036] Figure 3 This is a diagram of the multidimensional analysis module of the method of the present invention. Detailed Implementation

[0037] The present invention will now be described in further detail with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0038] like Figure 1 As shown, the overall architecture of the method of this invention includes a data source layer, a knowledge graph construction and storage layer, a multi-dimensional analysis and verification module, and a visualization result output. The data source layer covers open-source data and data crawled from the Internet; the knowledge graph layer uses RDF (triple) format to construct a knowledge graph of the data domain and stores and manages it visually using the Neo4j graph database; the multi-dimensional analysis and verification module uses semantic vector calculation, graph relationship reasoning, and text mining techniques to perform multi-angle consistency verification of the data; finally, the verification results are output through the front-end interface or Neo4j visualization tool, marking conflicting information points and providing correlation tracing.

[0039] like Figure 2 As shown, the methodology of this scheme includes five main steps: data acquisition, preprocessing and knowledge graph construction, semantic vectorization, multidimensional verification analysis, and result output. The specific process is as follows:

[0040] 1. Constructing a Knowledge Graph Structure Based on RDF: First, design a knowledge graph data model for the target domain (including but not limited to scenarios requiring multi-source data fusion and consistency verification, such as intelligence information processing, scientific and technological intelligence analysis, and industrial equipment monitoring). Entities and their relationships are defined using the Resource Description Framework (RDF) to obtain the knowledge graph for the target domain. For example, elements such as systems, technical parameters, time, and location are modeled as node entities, and the relationships between entities (such as "tool-component," "tool-technical indicator," "event-occurrence time," etc.) are modeled as edges (relationships). Based on this, the Neo4j graph database is used to store, update, and visualize the knowledge graph. Neo4j's intuitive graphical interface facilitates the display of complex relationships between entities, providing structured support for subsequent analysis.

[0041] 2. Multi-Source Data Acquisition: The information sources for this solution include open-source data and data obtained through web crawling. Open-source data includes official announcements, news reports, and professional journal articles, covering publicly available information on the status and evolution of the tool's technology. Web crawlers automatically extract relevant content from various websites and forums. Unlike traditional indiscriminate data acquisition, this invention's data acquisition is guided by target entities already modeled in the knowledge graph. That is, for specific entities (such as systems, technical parameters, components, events, etc.), descriptive information related to that entity is collected from multiple different sources, and the source identifier and timestamp are recorded simultaneously. Through this entity-oriented acquisition strategy, it is possible to ensure that the collected multi-source data maintains semantic correspondence with the knowledge graph nodes and edges, providing high-quality input for subsequent preprocessing, semantic vectorization, and multi-dimensional verification, thereby improving the accuracy and interpretability of consistency analysis and conflict identification.

[0042] 3. Data Preprocessing and Semantic Vectorization: In-depth preprocessing and analysis are performed on the collected information. Specifically, core elements are extracted from the text, including entities (e.g., tool names, versions, related personnel), technical keywords (e.g., technical indicators, component names, and other technical terms), time and location information, and identifiers such as tool numbers. This extraction process can be completed by combining natural language processing techniques and domain dictionaries to identify valuable information units from the text. Based on this, each collected text is treated as an independent processing unit, and its extracted core elements are input into the Qwen large-scale language model for semantic vectorization. Using the Qwen large-scale language model, a corresponding high-dimensional semantic vector representation is generated for each text, embedding the semantic information of the text into the vector space. This vectorized representation can capture the semantic similarity of sentences or paragraphs, providing a foundation for subsequent content comparison. Compared with traditional keyword vectors or One-Hot representations, the semantic vectors generated by the Qwen large-scale language model can better reflect the potential meaning and contextual relationships of the content, allowing subsequent analysis to go beyond just surface word matching.

[0043] 4. Multi-dimensional auxiliary verification analysis: This invention proposes to perform consistency verification on multi-source data from three dimensions, namely content similarity analysis, entity graph association analysis, and keyword co-occurrence analysis (e.g., Figure 3 (As shown). These three analytical dimensions work together to provide a comprehensive assessment of data consistency:

[0044] (1) Content Similarity Analysis: Based on the semantic vectors generated by the Qwen large-scale language model, similarity calculation is performed on data content from different sources. For example, cosine similarity is used to measure the semantic distance between two descriptive pieces of information in the vector space. If the similarity is greater than a set threshold, it is determined that the two descriptive pieces of information involve the same target entity, that is, they are mapped to the same entity node in the knowledge graph. On this basis, this module further checks whether their specific statements are consistent: First, the key attributes involved in the two texts (such as technical parameter values, status conclusions, event occurrence time, etc.) are extracted, and then these attributes are compared item by item. If the key attribute values ​​are completely consistent, it is determined that the two data contents are consistent and can be merged and mapped to the same knowledge graph entity node as evidence of the multi-source consistency of the entity; if there are differences or contradictions in the key attribute values, it is determined that the two data contents are inconsistent, and they are respectively attached to the same entity node in the knowledge graph, while marking the conflict relationship for subsequent tracing and manual judgment. This two-stage detection method, which first determines semantic relevance and then compares key attributes, can identify information that is different in expression but substantially consistent (avoiding missed detections) and also capture potential conflicts under the same topic (avoiding misjudgments), thus providing a more refined basis for consistency verification.

[0045] (2) Entity Graph Association Analysis: Utilizing the constructed knowledge graph, the consistency between descriptive information from different sources is evaluated from the perspective of entity relationships. In this invention, each descriptive piece of information corresponds to a text obtained from the aforementioned steps of collection and preprocessing, and is mapped to a target entity node in the knowledge graph. If two descriptive pieces of information are mapped to the same entity node, or if the corresponding entity nodes have a direct or indirect association path in the graph, then these two descriptive pieces of information are considered to be related at the entity level. This module queries the graph database to find the adjacency relationships and co-occurrence frequencies between multiple data records. Since the aforementioned step 2 has adopted an entity-oriented data collection method, descriptive information from different sources is naturally aggregated under the same entity node, which provides direct support for cross-source consistency comparison and conflict identification. If two data sets share the same entity (e.g., both mention a tool or the same subsystem technology), they will be connected to the same node in the graph; even if they do not directly share entities, if there is a path connected by an intermediate node (e.g., data A involves tool X, and data B involves a sub-component Y of tool X, then A and B are associated through XY), it also indicates that they have an indirect semantic relationship. For such related texts, the system further compares their key attributes (such as parameter values, status descriptions, event times, etc.): if the attributes are consistent, the two are used as evidence of multi-source consistency and attached to the same entity node; if the attributes are inconsistent, the two texts are stored separately under the entity node, and a "conflict" marker relationship is established for subsequent tracing and analysis.

[0046] In addition, the frequency of entities mentioned jointly by different sources in the statistical graph: entities that appear frequently in the statistical graph (such as a tool or technology repeatedly mentioned in multiple source texts). When an entity appears frequently in texts from different sources, the system will perform a consistency check on all texts containing that entity: comparing the key attributes and descriptive context of the entity in each text. If the descriptions in all texts are consistent, they are uniformly marked as a reliable evidence set for that entity; if inconsistencies exist, multiple parallel branches of source descriptions are created under the entity node, and the conflict type is marked to ensure that analysts can clearly distinguish the differences between sources.

[0047] Through the methods described above, entity graph association analysis can not only reveal direct or indirect connections between texts from different sources at the entity level, but also automatically trigger consistency checks when high-frequency entities are discovered, thereby effectively identifying cross-source entity conflicts. Knowledge graphs serve as association reasoning tools, extending consistency judgments beyond the textual level to the logical level based on entity networks.

[0048] (3) Keyword Co-occurrence Analysis: Further exploring the consistency and differences between data at the text content level. This module extracts keywords and calculates weights for data from various sources. For example, the TF-IDF method is used to select the most representative technical terms and high-frequency words in each text. Subsequently, for keywords shared in texts from different sources, the system not only counts their occurrence but also performs semantic comparison based on context. Specifically, a fixed-length context window (e.g., ±n words or sentences) is preset before and after the keyword. The text in this window is input into a semantic vector generation model to obtain a local semantic vector. Then, combined with dependency parsing and semantic role labeling, the semantic function and tendency of the keyword in the context are extracted. The system calculates the cosine similarity of the local semantic vector and combines it with the modification direction of the keyword in the syntactic dependency relationship (e.g., "upgrade → improve performance" and "upgrade → maintain the original configuration") to comprehensively judge whether the context of the keyword in different sources is consistent or contradictory.

[0049] If the contextual semantic similarity is high and the semantic roles are consistent, it is determined that the usage context of the keyword from different sources is consistent. The relevant texts can be merged under the keyword node as evidence of multi-source consistency of the keyword's meaning.

[0050] If the semantic similarity of the context is low or the keywords point to significantly different semantic roles in syntactic dependencies (for example, in one text, "promote system upgrade" points to "power enhancement," while in another it points to "maintain the original configuration"), a semantic conflict is determined to exist. In this case, the system will store the relevant texts under the keyword node in the knowledge graph and establish a "conflict" labeling relationship for subsequent visualization and manual analysis.

[0051] Through this innovative processing method of "keyword co-occurrence + context window local semantic modeling + syntactic dependency comparison", this invention can not only identify the semantic differences behind seemingly identical keywords, but also refine it to the level of "contextual meaning contradiction", thereby avoiding misjudgment caused by simple word frequency statistics and providing a third layer of evidence for multi-dimensional verification.

[0052] 5. Verification Result Output and Conflict Identification: After the multi-dimensional analysis and verification described above, the method of this invention will output a comprehensive result, identifying and highlighting inconsistencies in the multi-source data. The generation of these conflict annotations is directly based on the three types of analysis results from step 4 (semantic vector similarity calculation, entity graph relationship comparison, and keyword context semantic comparison), ensuring that the output results have clear semantic and structural basis. Specifically, these include the following types of data conflict points:

[0053] (1) Contradictions in the technical description of the same entity: When texts from different sources are mapped to the same entity node in the knowledge graph (e.g., the same tool or the same technical detail), the system will compare their key attributes (such as technical parameters and configuration status) based on the results of the content similarity analysis in step 4(1) and the entity graph association analysis in step 4(2). If contradictory attribute values ​​are found, a conflict relationship will be marked under the entity node. For example, if one source claims that a certain type of tool has been upgraded, while another source reports that the tool still maintains its original configuration, a "conflict" mark will be automatically generated under the tool entity.

[0054] (2) Update conflicts in the time dimension: During the attribute comparison process in steps 4(1) and 4(2), if different texts give inconsistent values ​​for the time attribute of the same entity event (such as upgrade time, release time), the system will mark it as a time conflict. For example, if the data released by the same organization in succession records the occurrence time of a certain technology upgrade as 2019 and 2020 respectively, a conflict record will be created under the time attribute of the event entity to remind the analyst to pay attention to the version difference.

[0055] (3) Ambiguity and conflict in the semantic direction of keywords: When texts from different sources share the same keyword in the keyword co-occurrence analysis in step 4(3), but their context windows are determined to have contradictory semantic directions after local semantic vectorization and syntactic dependency comparison, a conflict relationship is marked under the keyword node. For example, different sources all mention "scan upgrade", but one context points to "increase scanning distance" and the other points to "improve anti-interference ability". In this case, a conflict marker is generated under the keyword node and two context descriptions are retained for subsequent tracing.

[0056] Through the above steps, this invention achieves auxiliary verification of consistency for multi-source data. When the system detects conflicting data points, it can record relevant information in the knowledge graph or generate a conflict report, including the source of the conflicting data, content summary, conflict type, etc., for data analysts to review. Using Neo4j's visualization capabilities, users can also intuitively see how conflicting data is associated with the same entity or event node in the knowledge graph and trace the conflict's origin along the graph's relational chain. This combination of multi-dimensional verification and visual tracing significantly improves the efficiency and reliability of data consistency analysis.

[0057] Figure 3 The schematic diagram of the multidimensional analysis module of this invention depicts the internal structure and working principle of the multidimensional auxiliary verification module. It shows the content similarity analysis submodule, entity graph association analysis submodule, and keyword co-occurrence analysis submodule side-by-side, illustrating their respective inputs (such as data semantic vectors, knowledge graph entity relationships, extracted keywords, etc.) and how their outputs are aggregated for the final conflict determination. Figure 3 This allows for a clear understanding of how the three analytical dimensions work together to complete data consistency verification.

[0058] Although specific embodiments of the invention have been disclosed for illustrative purposes to aid in understanding and implementing the invention, those skilled in the art will understand that various substitutions, variations, and modifications are possible without departing from the spirit and scope of the invention and the appended claims. Therefore, the invention should not be limited to the content disclosed in the preferred embodiments, and the scope of protection claimed by the invention is defined by the claims.

Claims

1. A multi-source data multi-dimensional auxiliary verification method based on knowledge graphs, comprising the following steps: 1) Construct a knowledge graph for the target domain; 2) Guided by the target entity already modeled in the knowledge graph, collect descriptive information related to the target entity from multiple different sources, and simultaneously record the source identifier and timestamp; 3) Extract the core elements from each of the described information and input them into a large language model for semantic vectorization. Obtain the semantic vector for each of the described information; 4) Consistency verification of multi-source data is performed from three dimensions: content similarity analysis, entity graph association analysis, and keyword co-occurrence analysis; 5) Based on the consistency check results, identify and highlight any inconsistencies in the description information: 51) When two entities from different sources are mapped to the same entity node in the knowledge graph, the attributes of the two entities are determined to be contradictory based on the content similarity analysis results and the entity graph association analysis results. If there is a contradiction, the conflict relationship is marked under the entity node. 52) When two entities from different sources are mapped to the same entity node in the knowledge graph, and the time attribute values ​​of the two entities are inconsistent in the description information from different sources, then a time conflict is marked under the entity node. 53) When descriptive information from different sources shares the same keyword in keyword co-occurrence analysis, but the semantic direction of the keyword is contradictory in the descriptive information from different sources, then the node corresponding to the keyword in the knowledge graph is marked with a conflict relationship.

2. The method according to claim 1, characterized in that, The core elements include entity, technical keywords, time, location, and tool number.

3. The method according to claim 2, characterized in that, The core elements of the descriptive information are extracted by combining natural language processing techniques and domain dictionaries.

4. The method according to claim 1, 2, or 3, characterized in that, The content similarity analysis method is as follows: Based on the semantic vectors of the two descriptive information pieces, the content similarity between the two descriptive information pieces is calculated. If the content similarity is greater than a set threshold, it is determined that the two descriptive information pieces involve the same entity. Then, the attribute values ​​of the entity in the two descriptive information pieces are extracted, and each attribute value extracted is compared item by item. If the attribute values ​​are consistent, it is determined that the two descriptive information pieces are consistent, and the two descriptive information pieces are merged and mapped to the entity node corresponding to the entity in the knowledge graph as evidence of the multi-source consistency of the entity. Otherwise, it is determined that the two descriptive information pieces are inconsistent, and the two descriptive information pieces are merged and mapped to the entity node corresponding to the entity in the knowledge graph and the conflict relationship is marked.

5. The method according to claim 1, 2, or 3, characterized in that, The entity graph association analysis method is as follows: mapping the entities in the two description information to the corresponding nodes in the knowledge graph, and then determining whether the entities in the two description information are the same or have an association path based on the knowledge graph; if the entities in the two description information are the same or have an association path, then detecting whether the attributes of the entity in the two description information are consistent; if the attributes are consistent, merging the two description information and mapping them to the entity node corresponding to the entity in the knowledge graph as evidence of the multi-source consistency of the entity; If the attributes are inconsistent, the two descriptions are stored and marked as conflicting under the entity node.

6. The method according to claim 1, 2, or 3, characterized in that, The keyword co-occurrence analysis method is as follows: the TF-IDF method is used to calculate the keywords and their weights in the two description information, and the shared keywords in the two description information are counted; if the semantic similarity of the shared keyword in the context of the two description information is approximately a set threshold and the semantic roles are consistent, then the two description information are merged and mapped to the entity node corresponding to the shared keyword in the knowledge graph; otherwise, it is determined that the shared keyword has a semantic conflict, and the two description information are stored in the knowledge graph under the node corresponding to the shared keyword and a conflict labeling relationship is established.

7. The method according to claim 1, characterized in that, The large-scale language model is either the Qwen2.5 large-scale language model or the Qwen3 large-scale language model; a knowledge graph of the target domain is constructed using a resource description framework and stored in the Neo4j graph database.

8. The method according to claim 1, characterized in that, The target areas include, but are not limited to, intelligence information processing, scientific and technological intelligence analysis, and industrial equipment monitoring.

9. A server, characterized in that, It includes a memory and a processor, the memory storing a computer program configured to be executed by the processor, the computer program including instructions for performing the method of any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 8.

Images