Science and technology literature entity association extraction and graph generation method based on natural language processing

Abstract

This invention belongs to the field of natural language processing and knowledge graph technology, and discloses a method for entity association extraction and knowledge graph generation from scientific and technological documents based on natural language processing. The method includes: preprocessing the scientific and technological documents; extracting scientific and technological entities using a multimodal natural language processing model; and completing entity disambiguation by combining contextual semantics and a domain dictionary; identifying the association relationships between entities based on syntactic analysis and semantic role labeling; and calculating the association strength by introducing co-occurrence frequency, semantic similarity, and domain weight; constructing an initial knowledge graph; optimizing the graph structure through redundancy removal, conflict detection, and dynamic update mechanisms; and generating a scientific and technological document knowledge graph with high accuracy, completeness, and timeliness. This invention solves the problems of low accuracy in traditional scientific and technological document entity extraction, one-sided association identification, and static knowledge graphs, achieving efficient transformation from document text to structured knowledge, and providing strong support for scientific and technological intelligence analysis and scientific research innovation mining.

Description

Technical Field

[0001] This invention relates to the field of natural language processing technology, specifically to a method for entity association extraction and graph generation in scientific and technological documents based on natural language processing. Background Technology

[0002] With the rapid development of science and technology, the number of scientific and technological documents is growing exponentially. These massive amounts of documents contain a wealth of research entities (such as technical terms, research institutions, and research results) and the relationships between them, making them an important data source for scientific and technological intelligence analysis and scientific innovation. However, traditional methods of processing scientific and technological documents mainly rely on manual reading and organization, which suffers from low efficiency, high cost, and strong subjectivity, making it difficult to meet the needs of knowledge mining from large-scale documents.

[0003] Existing entity extraction and graph generation methods suffer from the following shortcomings: Insufficient entity extraction accuracy: General natural language processing models have poor adaptability to the scientific and technological literature domain, exhibiting low accuracy in recognizing technical terms and complex entities, and are prone to entity type confusion; Limited entity disambiguation effect: Existing methods often rely on single features (such as string matching) for disambiguation, ignoring the contextual semantics and domain knowledge of scientific and technological literature, leading to misjudgment of entities with the same name and omission of entities with different names; One-sided association recognition: Traditional association extraction is based solely on entity co-occurrence frequency, failing to consider semantic logical relationships and domain characteristics, making it difficult to accurately identify complex associations such as causal relationships, application relationships, and comparative relationships; Static graph problem: The generated knowledge graphs are mostly static structures, unable to adapt to incremental updates of scientific and technological literature, and lack redundancy removal and conflict detection mechanisms, resulting in a decline in graph quality over time; Lack of association strength quantification: The reliability and importance of entity associations are not quantified, making it difficult to support accurate analysis and decision-making based on the graph.

[0004] Furthermore, some existing methods rely on large model self-evaluation to obtain confidence levels, which can lead to high scores for inaccurate content. They also have extremely high requirements for the coverage of external knowledge bases; when the knowledge base cannot cover the entity relationships in the generated content, illusion detection becomes impossible. Additionally, relying on internet data to correct generated content may result in errors being corrected with errors, failing to meet the stringent accuracy requirements of the scientific and technological intelligence field. Therefore, there is an urgent need for a method that can efficiently and accurately extract entities and relationships from scientific and technological documents and generate high-quality, dynamically updated knowledge graphs to address the shortcomings of existing technologies. Summary of the Invention

[0005] In view of the problems existing in the current methods for entity association extraction and graph generation of scientific and technological documents based on natural language processing, this invention is proposed.

[0006] Therefore, the problem to be solved by this invention is to provide a method for extracting entity associations and generating knowledge graphs from scientific and technological documents based on natural language processing. Through multimodal NLP technology, domain adaptation models and dynamic optimization mechanisms, this method can accurately extract entities and their associations from scientific and technological documents and generate highly reliable and scalable knowledge graphs.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] In a first aspect, embodiments of the present invention provide a method for extracting entity associations and generating graphs from scientific and technological documents based on natural language processing, which includes preprocessing a target set of scientific and technological documents to obtain standardized text data;

[0009] A pre-trained natural language processing model is used to extract scientific and technological entities from standardized text data to obtain an initial set of scientific and technological entities.

[0010] Based on contextual semantic similarity and domain knowledge graph, entity disambiguation is performed on the initial set of scientific and technological entities to obtain a standardized set of scientific and technological entities.

[0011] By using syntactic analysis and semantic role labeling techniques, the relationships between entities in a standardized set of scientific and technological entities are identified, forming a candidate set of entity associations.

[0012] By introducing co-occurrence frequency, semantic similarity, and domain weight parameters, the association strength of each group of entity associations in the entity association candidate set is calculated.

[0013] Based on the standardized set of scientific and technological entities, entity relationships, and relationship strength, an initial knowledge graph of scientific and technological literature is constructed.

[0014] Redundancy removal, conflict detection, and dynamic updating optimization are performed on the initial scientific and technological literature knowledge graph to generate a complete scientific and technological literature knowledge graph.

[0015] As a preferred embodiment of the scientific and technological literature entity association extraction and graph generation method based on natural language processing described in this invention, the preprocessing of the target scientific and technological literature set includes text format standardization, word segmentation, stop word removal, part-of-speech tagging and syntactic dependency analysis, wherein the word segmentation is implemented using a model combining a bidirectional long short-term memory network and a conditional random field.

[0016] As a preferred embodiment of the scientific and technological literature entity association extraction and graph generation method based on natural language processing described in this invention, the pre-trained natural language processing model is used to adapt the model to the domain based on the Transformer architecture. After fine-tuning through the scientific and technological literature corpus, the extracted scientific and technological entities are each accompanied by a type label and text location information.

[0017] As a preferred embodiment of the scientific and technological document entity association extraction and graph generation method based on natural language processing described in this invention, the entity disambiguation processing of the initial scientific and technological entity set includes:

[0018] The semantic similarity between the entity to be disambiguated and the context entities is calculated using the cosine similarity formula:

[0019]

[0020] in, , Entities , The pre-trained semantic vectors;

[0021] Query candidate matching entities of the entity to be disambiguated in the domain knowledge graph, and calculate the attribute similarity between the technology entity pair to be processed and the candidate matching entities;

[0022] By integrating semantic similarity and attribute similarity, the candidate entity with the highest similarity is selected as the disambiguation result, forming a standardized set of scientific and technological entities.

[0023] As a preferred embodiment of the scientific and technological literature entity association extraction and graph generation method based on natural language processing described in this invention, the calculation of the association strength of each group of entity associations in the candidate set of entity associations includes using a weighted summation formula:

[0024]

[0025] in, Entity association The strength of the association, This represents the normalized value of the co-occurrence frequency of entity associations. For the semantic similarity of related entity pairs, Domain weights for relationships, , , These are the weighting coefficients, and .

[0026] As a preferred embodiment of the scientific and technological literature entity association extraction and graph generation method based on natural language processing described in this invention, the conflict detection and dynamic update optimization includes redundancy removal, conflict detection, and dynamic updating.

[0027] The redundancy removal includes deleting entity associations with association strength below a specified threshold and merging duplicate entity nodes and association relationships.

[0028] The conflict detection includes identifying issues such as entity attribute conflicts and contradictory relationships through a domain rule base and a logical reasoning engine, and determining conflict solutions using a voting mechanism;

[0029] The dynamic update includes establishing an incremental update interface for documents. When new scientific and technological documents are added, steps 1-6 are performed to integrate the newly extracted entities and relationships into the existing knowledge graph and update the relationship strength.

[0030] Secondly, embodiments of the present invention provide a system for extracting entity associations and generating a knowledge graph from scientific and technological documents based on natural language processing. This system includes: a text preprocessing module for standardizing scientific and technological documents and outputting standardized text data; an entity extraction module for extracting scientific and technological entities based on a pre-trained natural language processing model and outputting an initial set of scientific and technological entities; an entity disambiguation module for combining contextual semantics and a domain knowledge graph to achieve entity disambiguation and outputting a standardized set of scientific and technological entities; an association recognition module for identifying relationships between entities through syntactic analysis and semantic role labeling and outputting a candidate set of entity associations; an association strength calculation module for calculating the association strength of entity associations and outputting a set of entity associations with strength labels; a knowledge graph construction module for constructing an initial knowledge graph from scientific and technological documents based on entities and association relationships; and a knowledge graph optimization module for performing redundancy removal, conflict detection, and dynamic updates on the initial knowledge graph and outputting the final knowledge graph.

[0031] Thirdly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any step of the above-described method for entity association extraction and graph generation of scientific and technological documents based on natural language processing.

[0032] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program thereon, wherein: when the computer program is executed by a processor, it implements any step of the above-described method for entity association extraction and graph generation of scientific and technological documents based on natural language processing.

[0033] The beneficial effects of this invention are as follows: This invention addresses the scientific and technological literature scenario by constructing a high-precision, high-efficiency, and high-security knowledge graph construction and optimization system. Through the integration of domain-adapted pre-trained language models and professional dictionaries, it significantly improves the accuracy of identifying scientific entities such as technical terms, patent names, and experimental equipment. It achieves efficient entity disambiguation by combining contextual semantics with the domain knowledge graph, ensuring standardized consistency of entities. It accurately captures complex semantic relationships such as causality, application, and attribution using syntactic analysis and semantic role labeling, and evaluates the strength of these relationships through weighted quantification methods, supporting accurate decision-making in scientific and technological intelligence. Simultaneously, the system has a dynamic update mechanism that can automatically remove redundancy, detect conflicts, and incrementally integrate new literature, ensuring the timeliness and vitality of the knowledge graph. Relying on localized deployment and offline operation capabilities, it significantly reduces dependence on external resources while meeting the confidentiality requirements of scientific research data. The overall process is highly automated, with structured transformation efficiency more than 50 times higher than manual methods, effectively freeing up researchers' energy and allowing them to focus on core innovation. Attached Figure Description

[0034] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0035] Figure 1 This is a schematic diagram of a scientific and technological document entity association extraction and graph generation system based on natural language processing, provided as an embodiment of the present invention. Detailed Implementation

[0036] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0037] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0038] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0039] This invention is described in detail with reference to the schematic diagrams. When detailing the embodiments of this invention, for ease of explanation, the cross-sectional views illustrating the device structure may be partially enlarged, not adhering to the usual scale. Furthermore, the schematic diagrams are merely examples and should not be construed as limiting the scope of protection of this invention. In actual fabrication, the three-dimensional spatial dimensions of length, width, and depth should be included.

[0040] Furthermore, in the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are used solely for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In addition, the terms "first," "second," or "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0041] Unless otherwise explicitly specified and limited, the terms "installation," "connection," and "joining" in this invention should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; similarly, they can refer to mechanical connections, electrical connections, or direct connections, or indirect connections through an intermediate medium, or internal connections between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0042] Example

[0043] Reference Figure 1 This is the first embodiment of the present invention, which provides a method for entity association extraction and graph generation in scientific and technological documents based on natural language processing, including:

[0044] Step S1, Scientific and Technological Literature Preprocessing: Performing operations such as format standardization, word segmentation, and stop word removal on the original scientific and technological literature to lay the foundation for subsequent entity extraction and association recognition;

[0045] Step S2, Technology Entity Extraction: Using a domain-adapted pre-trained NLP model, extract multiple types of technology entities from the literature to obtain an initial entity set;

[0046] Step S3, Entity disambiguation: Integrate contextual semantics and domain knowledge graph to resolve the issues of entity homonyms and heteronyms, and form a standardized entity set;

[0047] Step S4, Entity Association Recognition: Based on syntactic analysis and semantic role labeling, identify complex association relationships between entities and generate an association candidate set;

[0048] Step S5, Association Strength Calculation: Combine co-occurrence frequency, semantic similarity, and domain weights to quantify the reliability and importance of entity associations;

[0049] Step S6, Initial Knowledge Graph Construction: Construct an initial knowledge graph with standardized entities as nodes, association relationships as edges, and association strengths as attributes;

[0050] Step S7, Knowledge Graph Optimization and Update: Optimize the graph structure through redundancy removal, conflict detection, and incremental updates to ensure the accuracy and timeliness of the graph.

[0051] In Step S1, the original scientific and technological literature includes various formats such as PDF, Word, and text. First, perform format standardization processing to convert non-text format literature into unified UTF-8 encoded text. Subsequently, use a model that combines bidirectional long short-term memory network (BiLSTM) and conditional random field (CRF) for word segmentation. Considering the characteristics of professional terms in scientific and technological literature, optimize the word segmentation dictionary and add domain-specific vocabulary. Then, remove stop words (such as meaningless words like "of", "and", "based on", etc.) in the text, and use a词性标注工具 (such as Stanford Parser) to label the词性 (noun, verb, adjective, etc.) of each word. Finally, perform syntactic dependency analysis to obtain the grammatical structure relationship between words, providing support for subsequent association recognition.

[0052] In Step S2, use a pre-trained model based on the Transformer architecture (such as BERT, RoBERTa), and fine-tune it through scientific and technological literature corpora to construct a domain-adapted entity extraction model. This model can identify multiple types of entities in scientific and technological literature.

[0053] In Step S3, the core of entity disambiguation is to solve the problems of "same name with different meanings" and "different names with the same meaning". The specific steps are as follows:

[0054] Semantic Similarity Calculation: Obtain the semantic vectors of the entity to be disambiguated and the context entities through a pre-trained model, and use the cosine similarity formula to calculate the semantic association degree between the two. The higher the semantic similarity, the greater the probability that the entities belong to the same concept;

[0055] Domain Knowledge Matching: Query the domain knowledge graph to obtain candidate matching entities for the entity to be disambiguated, extract the attribute information of the candidate entities (such as entity type,所属领域, associated entities, etc.), and calculate the attribute similarity between the scientific and technological entity pairs to be processed and the candidate entities;

[0056] Disambiguation decision: Combine semantic similarity (weight 0.6) and attribute similarity (weight 0.4) to obtain a comprehensive similarity. Select the candidate entity with the highest comprehensive similarity as the disambiguation result. If the comprehensive similarity is lower than a specified threshold (e.g., 0.5), it is determined to be a new entity and added to the standardized technology entity set.

[0057] In step S4, the relationships between entities are identified based on the syntactic analysis results and semantic role labeling (SRL) technology. Specific implementation includes:

[0058] Syntactic structure analysis: Through syntactic dependency trees, identify syntactic relationships between entities (such as subject-verb relations, verb-object relations, and modifier-head relations). For example, when deep learning is applied to image recognition, there is a verb-object relationship between deep learning and image recognition.

[0059] Semantic role labeling: Label the semantic role of each entity in a sentence (such as agent, patient, tool, purpose). For example, in the case of researchers using quantum computing technology to solve encryption problems, the researchers are the agents, quantum computing technology is the tool, and encryption problems are the patients.

[0060] Through the above steps, a candidate set of entity associations is generated, and each association includes a head entity, a tail entity, and an association type.

[0061] In step S5, to quantify the reliability and importance of entity associations, three core indicators are introduced: co-occurrence frequency, semantic similarity, and domain weight. A weighted summation formula is used to calculate the association strength.

[0062]

[0063] in: : The normalized value of the co-occurrence frequency of entity association in the target document set (range 0-1). The more times they co-occur, the stronger the association.

[0064] The semantic similarity between the head entity and the tail entity (range 0-1). The closer the semantic relationship, the stronger the relationship.

[0065] The domain weight of the association relationship (range 0-1) is set based on the experience of domain experts and literature statistics. For example, the weight of "R&D relationship" is higher than that of "comparison relationship".

[0066] , , Weighting coefficients, whose optimal values ​​are determined through cross-validation (e.g.) , , ).

[0067] The association strength ranges from 0 to 1, with higher values ​​indicating stronger reliability and importance of the entity association.

[0068] In step S6, an initial knowledge graph of scientific and technological literature is constructed using a graph database (such as Neo4j or ArangoDB). The graph structure includes:

[0069] Node: Corresponds to an entity in the standardized scientific and technological entity set. Each node contains the following attributes: entity ID, entity name, entity type, source document, and extraction time.

[0070] Edge: Corresponds to the association relationship in the entity association set. Each edge contains the following attributes: association ID, association type, association strength, and confidence (calculated based on the model prediction results).

[0071] Properties: Additional information about nodes and edges, used to support graph querying and analysis.

[0072] During the construction process, standardized entities and their relationships are imported into the graph database to establish an index of relationships between entities, thereby improving the efficiency of graph query.

[0073] In step S7, redundancy removal: set an association strength threshold (e.g., 0.2), delete edges with association strength below the threshold; merge duplicate nodes (nodes with identical entity names, types, and attributes) and duplicate edges (edges with identical head entities, tail entities, and association types), and retain the edge with the highest association strength;

[0074] Conflict detection: Construct a domain rule base (such as the same entity cannot belong to two different core domains at the same time, and causal relationships cannot be reversed), and traverse the graph through a logical reasoning engine to identify conflict information (such as conflicting entity attributes and contradictory relationships); for conflict information, a voting mechanism (combining model confidence, domain rule matching degree, and literature citation count) is used to determine the optimal solution, and if it cannot be resolved, it is marked as pending manual review;

[0075] Dynamic updates: An incremental update interface for literature is established. When new scientific and technological literature is added, preprocessing, entity extraction, and association identification processes are automatically triggered to integrate the new entities and new associations into the existing graph and update the association strength of related entities. At the same time, the graph is fully optimized regularly (e.g., monthly), outdated information (such as disproven associations) is deleted, and entities and associations corresponding to the latest research results are added.

[0076] like Figure 1 As shown, the present invention also provides a system for entity association extraction and graph generation in scientific and technological literature based on natural language processing, including the following functional modules:

[0077] Text preprocessing module 1: Implements format conversion, word segmentation, stop word removal, part-of-speech tagging, and syntactic dependency analysis for scientific and technological documents;

[0078] Entity extraction module 2: Based on a domain-adaptive pre-trained NLP model, extracts scientific and technological entities and outputs an initial entity set;

[0079] Entity disambiguation module 3: Integrates contextual semantics and domain knowledge graph to complete entity disambiguation and output a standardized entity set;

[0080] Association Recognition Module 4: Identifies entity associations through syntactic analysis and semantic role labeling, and outputs a candidate association set;

[0081] Association Strength Calculation Module 5: Calculates the association strength of entities and outputs a set of associations with strength labels;

[0082] Graph Construction Module 6: Based on entities and relationships, construct an initial knowledge graph and store it in a graph database;

[0083] Graph Optimization Module 7: Implements redundancy removal, conflict detection, and dynamic updating of the graph, and outputs the final knowledge graph.

[0084] Each module works collaboratively through a data interface, supporting both batch processing and incremental processing modes to meet application needs in different scenarios.

[0085] Process of entity association extraction and graph generation in scientific literature:

[0086] This embodiment provides a process for entity association extraction and graph generation from scientific and technological literature using the above method, including:

[0087] Data input: Select 1000 core journal articles (PDF format) in the field of artificial intelligence as the target scientific literature collection;

[0088] Preprocessing: The PDF paper was converted into text format, and the BiLSTM-CRF model was used for word segmentation. After removing stop words, part-of-speech tagging and syntactic dependency analysis were performed to obtain standardized text data.

[0089] Entity extraction: Using the BERT model fine-tuned based on literature in the field of artificial intelligence, 3200 scientific and technological entities from deep learning, convolutional neural networks, Google, ImageNet datasets, etc. were extracted to form an initial entity set;

[0090] Entity disambiguation: For the entity with the same name Transformer, through contextual semantic analysis and domain knowledge matching, three different concepts are distinguished: Transformer model, Transformer architecture, and Transformer company, thus completing entity disambiguation and obtaining 2800 standardized entities;

[0091] Association recognition: Identifies 4,500 entity associations in deep learning applications such as image classification and convolutional neural networks based on the Transformer architecture, forming a candidate association set;

[0092] Association strength calculation: Calculate the co-occurrence frequency, semantic similarity and domain weight of each association to obtain a set of associations with strength labels that are distributed between 0.1 and 0.9;

[0093] Graph Construction: Import 2800 entities as nodes and 4500 associations as edges into the Neo4j graph database to construct an initial knowledge graph;

[0094] Graph optimization: 1200 edges with an association strength below 0.2 were deleted, 50 duplicate nodes and 80 duplicate edges were merged, and 30 conflict information was detected and resolved; when 50 new papers are added later, 150 entities and 300 associations are added through the incremental update interface to complete the dynamic update of the graph.

[0095] To verify the effectiveness of the above methods, existing mainstream entity extraction tools (such as Stanford NER and spaCy) and graph generation tools were selected for comparison. The verification metrics included entity extraction accuracy, association recognition accuracy, graph construction efficiency, and graph update latency.

[0096] Validation metrics The method of the invention Comparison tool 1 Comparison tool 2 Entity extraction accuracy (%) 92.3 78.5 81.2 Relation recognition accuracy (%) 89.7 65.3 70.1 Atlas construction efficiency (articles / hour) 200 80 100 Atlas update latency (minutes) 5 30 25

[0097] The verification results show that the method of the present invention is significantly better than existing comparative tools in terms of entity extraction accuracy, association recognition accuracy, map construction efficiency and dynamic update speed, and can effectively meet the needs of scientific and technological literature knowledge mining.

[0098] In a preferred embodiment, a scientific and technological literature entity association extraction and graph generation system based on natural language processing includes: a text preprocessing module for standardizing scientific and technological literature and outputting standardized text data; an entity extraction module for extracting scientific and technological entities based on a pre-trained natural language processing model and outputting an initial set of scientific and technological entities; an entity disambiguation module for disambiguating entities by combining contextual semantics and domain knowledge graphs and outputting a standardized set of scientific and technological entities; an association recognition module for identifying the association relationships between entities through syntactic analysis and semantic role labeling and outputting a candidate set of entity associations; an association strength calculation module for calculating the association strength of entity associations and outputting a set of entity associations with strength labels; a graph construction module for constructing an initial scientific and technological literature knowledge graph based on entities and association relationships; and a graph optimization module for performing redundancy removal, conflict detection, and dynamic updating on the initial graph and outputting a final knowledge graph.

[0099] The above-mentioned unit modules can be embedded in the processor of the computer device in hardware form or independent of it, or they can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of the above modules.

[0100] In one embodiment, a computer device is provided, which may be a terminal. The computer device includes a processor, memory, a communication interface, a display screen, and an input device connected via a system bus. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The communication interface of the computer device is used for wired or wireless communication with external terminals. Wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen of the computer device may be an LCD screen or an e-ink display screen. The input device of the computer device may be a touch layer covering the display screen, or buttons, a trackball, or a touchpad located on the casing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0101] In summary, this invention addresses the scientific literature scenario by constructing a high-precision, high-efficiency, and high-security knowledge graph construction and optimization system. By integrating domain-adapted pre-trained language models with specialized dictionaries, it significantly improves the accuracy of identifying scientific entities such as technical terms, patent names, and experimental equipment. It achieves efficient entity disambiguation by combining contextual semantics with the domain knowledge graph, ensuring standardized consistency of entities. Syntactic analysis and semantic role labeling accurately capture complex semantic relationships such as causality, application, and attribution, and the strength of these relationships is evaluated through weighted quantification, supporting accurate decision-making in scientific and technological intelligence. Simultaneously, the system has a dynamic update mechanism that automatically removes redundancy, detects conflicts, and incrementally integrates new literature, ensuring the timeliness and vitality of the knowledge graph. Relying on localized deployment and offline operation capabilities, it significantly reduces dependence on external resources while meeting the confidentiality requirements of scientific data. The overall process is highly automated, with structured transformation efficiency more than 50 times higher than manual methods, effectively freeing up researchers' energy and allowing them to focus on core innovation.

[0102] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for entity association extraction and graph generation in scientific and technological documents based on natural language processing, characterized in that, include: The target collection of scientific and technological documents is preprocessed to obtain standardized text data; A pre-trained natural language processing model is used to extract scientific and technological entities from standardized text data to obtain an initial set of scientific and technological entities. Based on contextual semantic similarity and domain knowledge graph, entity disambiguation is performed on the initial set of scientific and technological entities to obtain a standardized set of scientific and technological entities. By using syntactic analysis and semantic role labeling techniques, the relationships between entities in a standardized set of scientific and technological entities are identified, forming a candidate set of entity associations. By introducing co-occurrence frequency, semantic similarity, and domain weight parameters, the association strength of each group of entity associations in the entity association candidate set is calculated. Based on the standardized set of scientific and technological entities, entity relationships, and relationship strength, an initial knowledge graph of scientific and technological literature is constructed. Redundancy removal, conflict detection, and dynamic updating optimization are performed on the initial scientific and technological literature knowledge graph to generate a complete scientific and technological literature knowledge graph.

2. The method for entity association extraction and graph generation of scientific and technological documents based on natural language processing according to claim 1, characterized in that, The preprocessing of the target scientific and technological literature collection includes text format standardization, word segmentation, stop word removal, part-of-speech tagging, and syntactic dependency analysis. Among them, word segmentation is implemented using a model that combines bidirectional long short-term memory network and conditional random field.

3. The method for entity association extraction and graph generation of scientific and technological documents based on natural language processing according to claim 1, characterized in that, The pre-trained natural language processing model is used to adapt the model to the domain based on the Transformer architecture. After fine-tuning through scientific and technological literature corpus, the extracted scientific and technological entities are provided, with each entity accompanied by a type label and text location information.

4. The method for entity association extraction and graph generation of scientific and technological documents based on natural language processing according to claim 1, characterized in that, The entity disambiguation process for the initial set of technological entities includes: The semantic similarity between the entity to be disambiguated and the context entities is calculated using the cosine similarity formula: in, , Entities , The pre-trained semantic vectors; Query candidate matching entities of the entity to be disambiguated in the domain knowledge graph, and calculate the attribute similarity between the technology entity pair to be processed and the candidate matching entities; By integrating semantic similarity and attribute similarity, the candidate entity with the highest similarity is selected as the disambiguation result, forming a standardized set of scientific and technological entities.

5. The method for entity association extraction and graph generation of scientific and technological documents based on natural language processing according to claim 1, characterized in that, The association strength of each entity association in the candidate set of entity associations is calculated using a weighted summation formula: in, Entity association The strength of the association, This represents the normalized value of the co-occurrence frequency of entity associations. For the semantic similarity of related entity pairs, Domain weights for relationships, , , These are the weighting coefficients, and .

6. The method for entity association extraction and graph generation of scientific and technological documents based on natural language processing according to claim 1, characterized in that, The conflict detection and dynamic update optimization includes redundancy removal, conflict detection, and dynamic update: The redundancy removal includes deleting entity associations with association strength below a specified threshold and merging duplicate entity nodes and association relationships. The conflict detection includes identifying entity attribute conflicts and contradictory relationship issues through a domain rule base and a logical reasoning engine, and determining conflict solutions using a voting mechanism; The dynamic update includes establishing an incremental update interface for documents. When new scientific and technological documents are added, steps 1-6 are performed to integrate the newly extracted entities and relationships into the existing knowledge graph and update the relationship strength.

7. A system for entity association extraction and graph generation in scientific and technological documents based on natural language processing, characterized in that, The method described by any one of claims 1-6 includes: The text preprocessing module is used to standardize scientific and technological documents and output standardized text data. Entity extraction module: Extracts scientific and technological entities based on a pre-trained natural language processing model and outputs an initial set of scientific and technological entities; The entity disambiguation module combines contextual semantics with domain knowledge graphs to achieve entity disambiguation and output a standardized set of scientific and technological entities. The association recognition module identifies the association relationships between entities through syntactic analysis and semantic role labeling, and outputs a candidate set of entity associations; The association strength calculation module calculates the association strength of entities and outputs a set of entity associations with strength labels. The graph construction module builds an initial knowledge graph of scientific and technological literature based on entities and relationships. The graph optimization module performs redundancy removal, conflict detection, and dynamic updates on the initial graph, outputting the final knowledge graph.

Images