Automatic construction and semantic comprehension enhancement method and device for large model knowledge base

Abstract

The invention provides an automatic construction and semantic understanding enhancement method and device for a large model knowledge base. Internet multi-type data sources are integrated through multi-source data collection, and text data related to the large model application field is obtained; a knowledge extraction and fusion technology is utilized to extract key knowledge elements from the collected data and eliminate redundant contradictory information to form a unified knowledge representation system; based on the unified knowledge representation system, automatically constructing a knowledge base with rich semantic association according to a predefined structure mode, and introducing a data quality evaluation mechanism to guarantee the reliability of the knowledge base; and based on the constructed knowledge base, the deep semantic analysis technology is used to enhance the semantic understanding ability of the large model, including semantic coding, knowledge graph embedding, semantic reasoning and knowledge completion. According to the scheme, efficient, accurate and automatic construction of the knowledge base can be achieved, the semantic understanding ability of a large model is remarkably improved, the large model can understand and apply knowledge more accurately, and powerful support is provided for artificial intelligence application.

Description

Technical Field

[0001] This invention relates to the field of large model technology, and in particular to a method and apparatus for automatically constructing and semantically enhancing large model knowledge bases. Background Technology

[0002] With the widespread application of artificial intelligence in many fields, large models, as a core technology, are becoming increasingly reliant on knowledge.

[0003] Traditional large-scale model knowledge base construction relies on manual labor, resulting in low efficiency, fragmented knowledge, and semantic understanding is limited by the quality of the knowledge base. Meanwhile, existing automatic construction methods suffer from problems such as single data source, poor accuracy of knowledge extraction, and insufficient depth of semantic understanding.

[0004] To address the aforementioned issues, this paper presents an automatic construction and semantic understanding enhancement method for large-scale model knowledge bases. This method enables efficient, accurate, and automatic construction of knowledge bases while significantly improving the semantic understanding capabilities of large-scale models, allowing them to more accurately understand and utilize knowledge, thus providing strong support for artificial intelligence applications. Summary of the Invention

[0005] This invention provides a method and apparatus for automatically constructing and enhancing the semantic understanding of a large model knowledge base. It can achieve efficient, accurate and automatic construction of the knowledge base, while significantly improving the semantic understanding ability of the large model, enabling it to understand and apply knowledge more accurately, and providing strong support for artificial intelligence applications.

[0006] According to one aspect of the present invention, a method for automatically constructing and semantically enhancing a large model knowledge base is provided, comprising:

[0007] By integrating multiple types of data sources from the Internet through multi-source data collection, text data related to the application domain of the large model can be obtained.

[0008] By using knowledge extraction and fusion technologies, key knowledge elements are extracted from the collected data and redundant and contradictory information is eliminated to form a unified knowledge representation system.

[0009] Based on the unified knowledge representation system, a knowledge base with rich semantic associations is automatically constructed according to a predefined structural pattern, while a data quality assessment mechanism is introduced to ensure the reliability of the knowledge base.

[0010] Based on the constructed knowledge base, deep semantic analysis technology is used to enhance the semantic understanding capabilities of large models, including semantic encoding, knowledge graph embedding, semantic reasoning, and knowledge completion.

[0011] Optionally, the multi-source data acquisition includes: determining the target application domain of the large model and selecting suitable data sources according to domain requirements;

[0012] Design an intelligent web crawler with domain adaptability. The crawler has built-in preset search strategies and crawling rules. The search strategies include multi-dimensional search expansion based on core domain terms, and the crawling rules include data source priority classification. The crawler also supports dynamically adjusting the crawling cycle according to the data source update frequency.

[0013] Optionally, the step of extracting key knowledge elements from the collected data and eliminating redundant and contradictory information to form a unified knowledge representation system includes:

[0014] Natural language processing (NLP) techniques are used to preprocess the collected multi-source text data, including word segmentation, part-of-speech tagging, and text cleaning; core entities in the domain are accurately extracted using entity recognition algorithms.

[0015] The relationship extraction model is used to mine the relationships between entities, and the key events in the text are extracted through the event extraction framework. The time, place, participants and core content of the events are recorded.

[0016] A fusion algorithm based on semantic similarity calculation is adopted to compare the same or similar knowledge from different data sources. When the semantic similarity between two pieces of knowledge is higher than a preset value, it is judged as redundant knowledge, and the knowledge with higher authority is retained. When there is a contradiction between two pieces of knowledge, arbitration is carried out by calling the domain standard knowledge base, and the knowledge representation system is output.

[0017] Optionally, the knowledge base is constructed in the following manner:

[0018] Based on the knowledge representation system output from the knowledge extraction and fusion stage, a predefined structural pattern for the knowledge base is determined, which includes a knowledge classification framework and knowledge association rules.

[0019] A storage approach combining graph databases and RDF is adopted, storing knowledge elements in semantic triples of subject-prediction-object. The graph database is used to store highly correlated knowledge networks, while RDF is used to store structured and standardized knowledge. Both storage methods support real-time data synchronization and bidirectional queries. During the knowledge base construction process, a multi-dimensional data quality assessment model is introduced to monitor knowledge in real time from three dimensions: accuracy, completeness, and consistency. When a piece of knowledge triggers a quality warning, it is automatically marked as pending review and pushed to the manual review module. Only after passing the review can it enter the formal storage of the knowledge base. If the review fails, it returns to the knowledge extraction and fusion stage for reprocessing.

[0020] Optionally, the semantic encoding includes:

[0021] Select a pre-trained language model that fits the domain requirements, use the completed knowledge base as the training data source, and fine-tune the pre-trained language model for domain adaptability. The fine-tuning process uses the mini-batch gradient descent algorithm and sets a reasonable learning rate.

[0022] The pre-trained language model with fine-tuning performs semantic encoding on the text knowledge in the knowledge base, mapping the text information to a 512-dimensional high-dimensional semantic space, enabling the model to capture deep semantic relationships and sentiment tendencies between knowledge. The semantic vectors output after encoding must meet the requirements that the semantic similarity of knowledge of the same category is higher than 90% and the semantic similarity of knowledge of different categories is lower than 30%.

[0023] Optionally, the knowledge graph embedding and the semantic reasoning include:

[0024] Knowledge graph embedding technology is introduced, and TransE or DistMult algorithms are selected to embed knowledge elements such as entities and relationships in the knowledge base into a 128-dimensional low-dimensional vector space by optimizing the objective function.

[0025] Semantic reasoning is achieved based on embedded semantic vectors. When the large model receives a user query, it encodes the query text into a semantic vector. By calculating the similarity between this vector and the semantic vectors of knowledge in the knowledge base, relevant knowledge is quickly located. At the same time, logical reasoning is performed in combination with domain knowledge rules to generate accurate knowledge association chains.

[0026] When a knowledge gap is detected in the knowledge base, the knowledge completion function is activated, automatically calling the multi-source data acquisition module to supplement and crawl relevant information.

[0027] Optionally, the method further includes:

[0028] Scheduled incremental updates are executed at fixed intervals to collect, extract, merge, and store newly added or updated data source content, thereby reducing the impact of updates on knowledge base services.

[0029] Triggered full updates are initiated when a major event occurs in the domain. The initiation conditions include manual and automatic triggering. The system switches from the original knowledge base to ensure that the large model can call the knowledge base normally during the update and avoid knowledge service interruption.

[0030] After each update is completed, an update report is generated, recording information such as the amount of updated data, the updated knowledge categories, and the quality assessment results, which facilitates subsequent traceability and optimization.

[0031] According to another aspect of the present invention, an apparatus for automatically constructing and semantically enhancing a large model knowledge base is provided, comprising:

[0032] The acquisition unit is used to integrate multiple types of data sources from the Internet through multi-source data collection to acquire text data related to the application domain of the large model.

[0033] The extraction unit is used to extract key knowledge elements from the collected data and eliminate redundant and contradictory information by using knowledge extraction and fusion technology to form a unified knowledge representation system.

[0034] The construction unit is used to automatically build a semantically rich knowledge base based on the unified knowledge representation system and according to a predefined structural pattern, while introducing a data quality assessment mechanism to ensure the reliability of the knowledge base.

[0035] The enhancement unit is used to enhance the semantic understanding capabilities of large models based on the constructed knowledge base using deep semantic analysis techniques, including semantic encoding, knowledge graph embedding, semantic reasoning, and knowledge completion.

[0036] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0037] At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores a computer program executable by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to execute the automatic construction and semantic understanding enhancement method for a large model knowledge base according to any embodiment of the present invention.

[0038] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the automatic construction and semantic understanding enhancement method for a large model knowledge base as described in any embodiment of the present invention.

[0039] This invention provides a method and apparatus for automatically constructing and enhancing the semantic understanding of a large model knowledge base. It can achieve efficient, accurate and automatic construction of the knowledge base, while significantly improving the semantic understanding ability of the large model, enabling it to understand and apply knowledge more accurately, and providing strong support for artificial intelligence applications.

[0040] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1This is a flowchart of an automatic construction and semantic understanding enhancement method for a large model knowledge base provided in an embodiment of the present invention;

[0043] Figure 2 This is a schematic diagram of the structure of an automatic construction and semantic understanding enhancement device for a large model knowledge base provided in an embodiment of the present invention;

[0044] Figure 3 This is a schematic diagram of the structure of an electronic device that implements the automatic construction and semantic understanding enhancement method of the large model knowledge base in the embodiments of the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0046] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0047] like Figure 1 As shown, this embodiment of the invention provides a method for automatically constructing and enhancing the semantic understanding of a large model knowledge base. This method may include the following steps:

[0048] S110. By integrating multiple types of data sources from the Internet through multi-source data collection, text data related to the application field of the large model can be obtained.

[0049] It integrates various data sources on the Internet, including but not limited to web page text, professional databases, academic literature, and news information. By designing intelligent web crawlers, it extensively collects various types of text data related to the application field of large models according to preset search strategies and crawling rules, ensuring the richness and diversity of the data.

[0050] In this embodiment of the invention, multi-source data acquisition includes: determining the target application domain of the large model and selecting suitable data sources according to the domain requirements;

[0051] The design incorporates a domain-adaptive intelligent web crawler with built-in preset search strategies and crawling rules. The search strategies include multi-dimensional retrieval extensions based on core domain terms, and the crawling rules include data source priority classification. Furthermore, the crawler supports dynamically adjusting the crawling cycle according to the data source update frequency.

[0052] Taking the construction of a large-scale knowledge base in the medical field as an example, web crawlers can collect text data from multiple sources, such as authoritative medical websites, such as WebMD and MayoClinic, medical academic databases, online platforms for medical professional books, and medical news websites, covering a full range of content including disease symptoms, treatment methods, drug information, and medical research progress.

[0053] S120. Using knowledge extraction and fusion technology, key knowledge elements are extracted from the collected data and redundant and contradictory information is eliminated to form a unified knowledge representation system.

[0054] Utilizing advanced natural language processing technology, the collected multi-source text data is processed in depth to accurately identify and extract key knowledge elements such as entities, relationships, and events. Entities include disease names, drug names, and symptom names; relationships cover the association between diseases and symptoms, and the correspondence between drugs and indications; events include medical research findings.

[0055] Fusion algorithms are employed to integrate identical or similar knowledge from different data sources, eliminating redundant and contradictory information and constructing a unified and standardized knowledge representation system. For example, to address the differences in disease symptom descriptions from different data sources, semantic similarity calculation and data mining techniques are used to integrate similar symptoms into a unified entity and clarify their association with the disease.

[0056] In this embodiment of the invention, key knowledge elements are extracted from the collected data and redundant and contradictory information is eliminated to form a unified knowledge representation system, including:

[0057] Natural language processing (NLP) techniques are used to preprocess the collected multi-source text data, including word segmentation, part-of-speech tagging, and text cleaning; core entities in the domain are accurately extracted using entity recognition algorithms.

[0058] The relationship extraction model is used to mine the relationships between entities, and the key events in the text are extracted through the event extraction framework. The time, place, participants and core content of the events are recorded.

[0059] A fusion algorithm based on semantic similarity calculation is adopted to compare the same or similar knowledge from different data sources. When the semantic similarity between two pieces of knowledge is higher than a preset value, it is judged as redundant knowledge, and the knowledge with higher authority is retained. When there is a contradiction between two pieces of knowledge, arbitration is carried out by calling the domain standard knowledge base, and the knowledge representation system is output.

[0060] Natural Language Processing (NLP) techniques are used to preprocess multi-source text data (such as web page text, academic literature, etc.). Word segmentation breaks down continuous text into word units that conform to domain semantics; part-of-speech tagging adds grammatical attributes to the segmented words. Text cleaning removes invalid information from the data to ensure the quality of subsequent text processing.

[0061] Entity recognition uses specialized entity recognition algorithms to locate and extract core entities within the domain, ensuring the accuracy of entity extraction and laying the foundation for subsequent knowledge association. It employs a relation extraction model to uncover the relationships between identified entities. With the help of an event extraction framework, it extracts key events from the text and fully records the event's time, location, participating entities, core content, and other attributes to ensure the integrity of the event information.

[0062] A fusion algorithm based on semantic similarity calculation is used to integrate knowledge from different data sources. The semantic similarity of two pieces of knowledge is compared; if the similarity exceeds a preset value, it is considered redundant knowledge, and knowledge from the more authoritative source is retained first. If two pieces of knowledge contradict each other, arbitration is performed by calling a domain standard knowledge base to determine the unique and accurate knowledge. After redundancy elimination and contradiction arbitration, a unified and standardized knowledge representation system is finally formed, providing standardized knowledge elements for the subsequent automatic construction of the knowledge base.

[0063] S130. Based on a unified knowledge representation system, a knowledge base with rich semantic associations is automatically constructed according to a predefined structural pattern, while a data quality assessment mechanism is introduced to ensure the reliability of the knowledge base.

[0064] Based on the extracted and integrated knowledge elements, the framework and content of the knowledge base are automatically generated according to a predefined knowledge base structure and pattern. The knowledge base structure adopts a storage method suitable for semantic representation, such as graph database or RDF, and stores entities, relations, and events in the form of semantic triples (such as subject-prediction-object), forming a semantically rich knowledge network.

[0065] During the knowledge base construction process, a data quality assessment model is used to monitor the accuracy, completeness, and consistency of knowledge in real time, automatically filtering out low-quality knowledge to ensure the reliability and effectiveness of the knowledge base. For example, for disease treatment plan knowledge extracted from different data sources, unreasonable treatment suggestions are removed and authoritatively verified plans are retained through comparative analysis and algorithm verification.

[0066] In this embodiment of the invention, the knowledge base is constructed in the following manner:

[0067] Based on the knowledge representation system output from the knowledge extraction and fusion stage, a predefined structural pattern for the knowledge base is determined. The structural pattern includes a knowledge classification framework and knowledge association rules.

[0068] A storage approach combining graph databases and RDF is adopted, storing knowledge elements in semantic triples of subject-prediction-object. The graph database is used to store highly correlated knowledge networks, while RDF is used to store structured and standardized knowledge. Both storage methods support real-time data synchronization and bidirectional queries. During the knowledge base construction process, a multi-dimensional data quality assessment model is introduced to monitor knowledge in real time from three dimensions: accuracy, completeness, and consistency. When a piece of knowledge triggers a quality warning, it is automatically marked as pending review and pushed to the manual review module. Only after passing the review can it enter the formal storage of the knowledge base. If the review fails, it returns to the knowledge extraction and fusion stage for reprocessing.

[0069] Taking the unified and standardized knowledge representation system output from the knowledge extraction and fusion stage as the core input, the predefined structural pattern of the knowledge base is first clarified. The knowledge classification framework divides categories according to the attribute characteristics of domain knowledge. For example, the medical field can be divided into categories such as "basic medical concepts, disease knowledge, treatment plans, and medical research progress", and the education field can be divided into categories such as "subject terminology, knowledge point system, teaching methods, and assessment standards", to ensure the classification logic of knowledge storage.

[0070] Knowledge association rules define the types of relationships between different categories of knowledge, such as citation relationships, derivation relationships, and supplementary relationships, providing a basis for subsequent knowledge network construction.

[0071] A storage method combining graph data and RDF is adopted, using semantic triples of subject-prediction-object as a unified storage form for knowledge elements to achieve accurate carrying of different types of knowledge: The role of graph database is to focus on storing "highly correlated knowledge networks, such as the chain-linked network of disease-symptom-drug-treatment plan in the medical field, and the radial network of policy-industry-enterprise-stock in the financial field. Relying on the efficient storage and query capabilities of graph database for complex relationships, it supports large models to quickly trace knowledge relationships;

[0072] RDF focuses on storing structured and standardized knowledge, such as standardized attributes of drugs, such as ingredients, dosage, and contraindications, as well as fixed content such as definitions, core points, and assessment requirements of knowledge points. By leveraging the standardization of structured data semantic representation, RDF ensures the uniformity and accuracy of the format of this type of knowledge.

[0073] Both storage methods support real-time data synchronization and bidirectional querying, ensuring the continuity of knowledge retrieval.

[0074] To prevent low-quality knowledge from entering the knowledge base, a multi-dimensional data quality assessment model is introduced during the construction process, forming a closed-loop quality control system of real-time monitoring, early warning, review, and rework. The assessment dimensions dynamically monitor knowledge from three core dimensions: Accuracy: comparing knowledge content with authoritative domain databases to ensure error-free information; Completeness: checking for missing key information in knowledge elements to avoid fragmented knowledge; Consistency: verifying the absence of logical contradictions between knowledge items.

[0075] When a piece of knowledge triggers a quality warning in any dimension, it is automatically marked as pending review and pushed to the manual review module. Only after the reviewer approves it can the knowledge be formally stored in the knowledge base. If it fails to meet the requirements, the knowledge is returned to the "knowledge extraction and fusion stage" and the extraction and fusion process is repeated until the quality requirements are met.

[0076] S140. Based on the constructed knowledge base, deep semantic analysis technology is used to enhance the semantic understanding ability of large models, including semantic encoding, knowledge graph embedding, semantic reasoning and knowledge completion.

[0077] Based on the constructed knowledge base, pre-trained language models in deep learning (such as BERT and GPT) are used for semantic encoding and representation, mapping textual information to a high-dimensional semantic space to capture deep semantic connections and logical relationships between knowledge. Through model training and optimization, the depth and breadth of the large model's understanding of knowledge semantics are improved, enabling it to accurately grasp the inherent meaning, implicit semantics, and sentiment of knowledge.

[0078] By introducing knowledge graph embedding technology, entities, relationships, and other knowledge elements from the knowledge base are embedded into a low-dimensional vector space, further enhancing the compactness and computability of semantic representation. These semantic vectors are then used for semantic reasoning and knowledge completion. When users pose questions or large models perform knowledge retrieval, the system can quickly and accurately locate knowledge related to the question, and perform deep knowledge mining and reasoning based on semantic associations to generate accurate and reasonable answers or decision suggestions.

[0079] In this embodiment of the invention, semantic encoding includes:

[0080] Select a pre-trained language model that fits the domain requirements, use the completed knowledge base as the training data source, and fine-tune the pre-trained language model for domain adaptability. The fine-tuning process uses the mini-batch gradient descent algorithm and sets a reasonable learning rate.

[0081] The pre-trained language model with fine-tuning performs semantic encoding on the text knowledge in the knowledge base, mapping the text information to a 512-dimensional high-dimensional semantic space, enabling the model to capture deep semantic relationships and sentiment tendencies between knowledge. The semantic vectors output after encoding must meet the requirements that the semantic similarity of knowledge of the same category is higher than 90% and the semantic similarity of knowledge of different categories is lower than 30%.

[0082] Prioritize selecting pre-trained language models that are suitable for the target application domain of the large model, such as BioBERT for the medical field and BERT-Large or GPT-3.5 for the general domain, to ensure that the basic capabilities of the model match the characteristics of the domain knowledge; use the pre-built knowledge base as the data source for data input, so that the model can learn standardized knowledge in the domain; fine-tuning is carried out by using the mini-batch gradient descent algorithm to optimize the model parameters, while setting an appropriate learning rate, and through multiple rounds of training, the model adapts to the characteristics of domain knowledge and improves the accuracy of understanding text in the domain.

[0083] The encoding operation uses a fine-tuned pre-trained language model to semantically encode textual knowledge in the knowledge base, transforming textual information into 512-dimensional high-dimensional semantic vectors. Through high-dimensional vector mapping, the model captures deep semantic relationships between knowledge and textual sentiment. The encoded semantic vectors must meet strict similarity standards: semantic similarity of knowledge of the same category is higher than 90%, and semantic similarity of knowledge of different categories is lower than 30%, laying the foundation for subsequent semantic reasoning and knowledge matching.

[0084] In this embodiment of the invention, knowledge graph embedding and semantic reasoning include:

[0085] Knowledge graph embedding technology is introduced, and TransE or DistMult algorithms are selected to embed knowledge elements such as entities and relationships in the knowledge base into a 128-dimensional low-dimensional vector space by optimizing the objective function.

[0086] Semantic reasoning is achieved based on embedded semantic vectors. When the large model receives a user query, it encodes the query text into a semantic vector. By calculating the similarity between this vector and the semantic vectors of knowledge in the knowledge base, relevant knowledge is quickly located. At the same time, logical reasoning is performed in combination with domain knowledge rules to generate accurate knowledge association chains.

[0087] When a knowledge gap is detected in the knowledge base, the knowledge completion function is activated, automatically calling the multi-source data acquisition module to supplement and crawl relevant information.

[0088] Introducing knowledge graph embedding technology transforms abstract knowledge elements such as entities and relationships in a knowledge base into computable low-dimensional vectors. Mainstream knowledge graph embedding algorithms such as TransE or DistMult are selected to adapt to different types of entity relationships in the knowledge base. The algorithms embed entities and relationships from the knowledge base into a 128-dimensional low-dimensional vector space, breaking down the abstractness of knowledge elements. By optimizing the objective function, it is ensured that the low-dimensional vectors retain the core semantic features of the knowledge to the greatest extent, while improving vector computation efficiency, laying a computable foundation for subsequent semantic reasoning.

[0089] Based on embedded low-dimensional semantic vectors, a large model can accurately respond to user queries and derive knowledge associations. When the large model receives a user query, it first encodes the query text into a semantic vector of the same dimension as the knowledge elements in the knowledge base. By calculating the similarity between the query semantic vector and the semantic vectors of knowledge in the knowledge base, highly relevant knowledge is quickly filtered out. Combined with domain knowledge rules, logical deduction is performed on the matched knowledge to generate a complete knowledge association chain, ensuring the accuracy and coherence of the reasoning results.

[0090] When a knowledge gap is detected in the knowledge base, a knowledge completion mechanism is activated. This automatically invokes previously collected multi-source data and, based on the content of the knowledge gap, crawls relevant information from authoritative data sources in the domain. The collected supplementary information undergoes verification by a multi-dimensional data quality assessment model during the knowledge base construction phase. Once verified, it is integrated into the knowledge base to fill the original gap, ensuring the integrity of the knowledge base's knowledge system and providing more comprehensive knowledge support for subsequent semantic reasoning.

[0091] In this embodiment of the invention, the method may further include the following steps:

[0092] Scheduled incremental updates are executed at fixed intervals to collect, extract, merge, and store newly added or updated data source content, thereby reducing the impact of updates on knowledge base services.

[0093] Triggered full updates are initiated when a major event occurs in the domain. The initiation conditions include manual and automatic triggering. The system switches from the original knowledge base to ensure that the large model can call the knowledge base normally during the update and avoid knowledge service interruption.

[0094] After each update is completed, an update report is generated, recording information such as the amount of updated data, the updated knowledge categories, and the quality assessment results, which facilitates subsequent traceability and optimization.

[0095] Executed on a fixed schedule, operations are only performed on newly added or updated content in the data source. The process covers multi-source data collection, knowledge extraction and fusion, and knowledge storage (writing compliant knowledge into the knowledge base). This avoids the resource consumption and time costs of full updates, minimizes the impact of the update process on the normal service of the knowledge base, and ensures that the large model can still stably call existing knowledge during the update period.

[0096] Triggered when a major event occurs in the target application area of ​​the large model, such as the release of new treatment guidelines in the medical field or the introduction of curriculum reform policies in the education field, ensuring that the knowledge base can quickly synchronize core domain knowledge; including both manually triggered domains and automatically triggered domains, taking into account the accuracy of human decision-making and the timeliness of system response; adopting a dual-database parallel strategy, that is, a new temporary knowledge base is created during the update to store the fully updated knowledge, and after the temporary knowledge base passes the quality assessment, it is seamlessly switched with the original knowledge base, completely avoiding knowledge service interruption during the update.

[0097] After each update, an update report is automatically generated. The core recorded content includes: Update data volume: the number of new / updated knowledge entries and the number of data sources; Update knowledge category: the category to which the updated knowledge belongs according to the predefined classification framework of the knowledge base; Quality assessment results: the assessment scores of the updated knowledge in terms of accuracy, completeness, and consistency, and the percentage of knowledge that passed / failed manual review; It provides data support for subsequent knowledge base optimization (such as adjusting the update cycle and optimizing the priority of data sources), and at the same time enables the traceability of the update process, making it easier to identify potential problems.

[0098] Compared with the prior art, the present invention has the following advantages:

[0099] 1. Improve knowledge base construction efficiency: The automated multi-source data collection, knowledge extraction and fusion process replaces the traditional manual collection and organization work, which greatly shortens the knowledge base construction cycle and improves the timeliness of knowledge base updates, enabling it to quickly adapt to the dynamic changes and expansion of knowledge.

[0100] 2. Improve the quality and completeness of the knowledge base: Through multi-source data fusion and quality assessment, rich and comprehensive knowledge resources are integrated, redundant and contradictory information is eliminated, and a semantically rich and standardized knowledge base is constructed, providing a high-quality knowledge foundation for large models.

[0101] 3. Enhance the semantic understanding capabilities of large models: The application of deep semantic analysis and knowledge graph embedding technology enables large models to more accurately understand text semantics, knowledge associations and user intent, improving their performance in application scenarios such as intelligent question answering, text generation and semantic search, and providing users with a more intelligent and personalized service experience.

[0102] 4. Promoting the development of artificial intelligence applications: The method of this invention provides strong support for knowledge-driven large models, which helps to break through the bottleneck of artificial intelligence in knowledge-intensive tasks, expand the depth and breadth of its application in many fields such as medical care, education, finance, and industry, and promote the innovative development and industrialization of artificial intelligence technology.

[0103] like Figure 2 As shown, this embodiment of the invention provides an automatic construction and semantic understanding enhancement device for a large model knowledge base, including:

[0104] The acquisition unit 210 is used to integrate multiple types of data sources from the Internet through multi-source data acquisition to acquire text data related to the application domain of the large model.

[0105] Extraction unit 220 is used to extract key knowledge elements from collected data and eliminate redundant and contradictory information by using knowledge extraction and fusion technology to form a unified knowledge representation system.

[0106] Building unit 230 is used to automatically build a semantically rich knowledge base based on a unified knowledge representation system and a predefined structural pattern, while introducing a data quality assessment mechanism to ensure the reliability of the knowledge base.

[0107] Enhancement Unit 240 is used to enhance the semantic understanding capabilities of large models based on the constructed knowledge base using deep semantic analysis techniques, including semantic encoding, knowledge graph embedding, semantic reasoning, and knowledge completion.

[0108] It is understood that the structures illustrated in the embodiments of the present invention do not constitute a specific limitation on the automatic construction and semantic understanding enhancement device for large model knowledge bases. In other embodiments of the present invention, the automatic construction and semantic understanding enhancement device for large model knowledge bases may include more or fewer components than illustrated, or combine some components, split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0109] The information interaction and execution process between the various units in the above-mentioned device are based on the same concept as the method embodiment of the present invention, and the specific details can be found in the description of the method embodiment of the present invention, and will not be repeated here.

[0110] Figure 3A schematic diagram of an electronic device 10 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0111] like Figure 3 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 may also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0112] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0113] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as the automatic construction of large model knowledge bases and semantic understanding enhancement methods.

[0114] In some embodiments, the method for automatically constructing and enhancing the semantic understanding of a large model knowledge base can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the method for automatically constructing and enhancing the semantic understanding of a large model knowledge base described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the method for automatically constructing and enhancing the semantic understanding of a large model knowledge base by any other suitable means (e.g., by means of firmware).

[0115] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0116] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0117] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0118] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0119] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0120] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0121] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0122] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. An automatic construction of large model knowledge base and semantic understanding enhancement method, characterized in that, The application relates to a method for constructing a knowledge base for a large model, and belongs to the technical field of artificial intelligence. The method comprises the following steps: Collecting and integrating multiple types of Internet data sources through multi-source data collection to obtain text data related to the application field of the large model; Extracting key knowledge elements and eliminating redundant and contradictory information from the collected data by using knowledge extraction and fusion technology to form a unified knowledge representation system; Based on the unified knowledge representation system, a knowledge base with rich semantic association is automatically constructed according to a pre-defined structure mode, and a data quality evaluation mechanism is introduced to ensure the reliability of the knowledge base; 2. The method of claim 1, wherein, Based on the constructed knowledge base, the deep semantic analysis technology is used to realize the semantic understanding ability enhancement of the large model, including semantic coding, knowledge graph embedding, semantic reasoning and knowledge completion. The multi-source data collection comprises the following steps: Determining the target application field of the large model, and screening the adaptive data sources according to the field requirements; 3. The method of claim 1, wherein, Designing an intelligent network crawler with field adaptability, wherein the crawler is built-in with preset search strategies and crawling rules, the search strategies include multi-dimensional retrieval expansion based on field core terms, the crawling rules include data source priority division, and the crawler supports dynamic adjustment of the crawling cycle according to the data source update frequency. The method for extracting key knowledge elements and eliminating redundant and contradictory information from the collected data to form a unified knowledge representation system comprises the following steps: Using natural language processing technology to preprocess the collected multi-source text data, including word segmentation, part-of-speech tagging and text cleaning; accurately extracting field core entities through entity recognition algorithm; Using a relationship extraction model to mine the association between entities, and extracting key events in the text through an event extraction framework, and recording the time, place, participating subjects and core content of the event; 4. The method of claim 1, wherein, Using a fusion algorithm based on semantic similarity calculation to compare the same or similar knowledge from different data sources, when the semantic similarity of two pieces of knowledge is higher than a preset value, the knowledge with higher source authority is retained, and when there is a contradiction between the two pieces of knowledge, the field standard knowledge base is called to arbitrate, and the knowledge representation system is output. The knowledge base is constructed by the following method: Based on the knowledge representation system output in the knowledge extraction and fusion stage, the pre-defined structure mode of the knowledge base is determined, and the structure mode includes a knowledge classification framework and a knowledge association rule; 5. The method of claim 1, wherein, Using a storage method combining a graph database and RDF, the knowledge elements are stored in the form of semantic triples of subject-predicate-object, wherein the graph database is used for storing high-correlation knowledge network, and the RDF is used for storing structured standard knowledge, and the two storage methods support real-time data synchronization and bidirectional query; during the construction of the knowledge base, a multi-dimensional data quality evaluation model is introduced to monitor the knowledge in real time from three dimensions of accuracy, integrity and consistency, when a piece of knowledge triggers a quality warning, it is automatically marked as a state to be audited and pushed to an artificial auditing module, after the auditing is passed, the piece of knowledge can be formally stored in the knowledge base, and if the auditing is not passed, the piece of knowledge is returned to the knowledge extraction and fusion stage for reprocessing. The semantic coding comprises the following steps: Selecting a pre-trained language model adaptive to the field requirements, using the completed knowledge base as a training data source, and performing field adaptation fine-tuning on the pre-trained language model, wherein a small-batch gradient descent algorithm is used in the fine-tuning process, and a reasonable learning rate is set. The text knowledge in the knowledge base is semantically encoded by the fine-tuned pre-trained language model, and the text information is mapped to a 512-dimensional high-dimensional semantic space, so that the model can capture the deep semantic association and emotional tendency between knowledge. The semantic vector output after encoding needs to meet the requirements that the semantic similarity of similar category knowledge is higher than 90% and the semantic similarity of different category knowledge is lower than 30%.

6. The method of claim 1, wherein, The knowledge graph embedding and the semantic reasoning include: The knowledge graph embedding technology is introduced, and the TransE or DistMult algorithm is selected to embed the entity, relationship and other knowledge elements in the knowledge base into a 128-dimensional low-dimensional vector space, and the objective function is optimized; Based on the embedded semantic vector, the semantic reasoning function is realized. When the large model receives a user query, the query text is encoded into a semantic vector, the similarity between the vector and the semantic vector of the knowledge in the knowledge base is calculated, the related knowledge is quickly located, and the logic reasoning is combined with the domain knowledge rules to generate an accurate knowledge association chain; When it is detected that there is a knowledge gap in the knowledge base, the knowledge completion function is started, and the multi-source data acquisition module is automatically called to supplement the crawled information.

7. The method of claim 1, wherein, The method further includes: The fixed period incremental update is performed at a fixed period, and the newly added or updated data source content is collected, extracted, fused and stored in the database, so that the influence of the update on the knowledge base service is reduced; The trigger type full update is started when a major event occurs in the field, the starting conditions include manual triggering and automatic triggering, the original knowledge base is switched, the large model can normally call the knowledge base during the update, and the interruption of the knowledge service is avoided; After each update is completed, an update report is generated, and information such as the update data volume, the updated knowledge category, the quality evaluation result and the like is recorded, so that subsequent tracing and optimization are facilitated.

8. The device for automatic construction of large model knowledge base and semantic understanding enhancement, characterized in that, It includes: An acquisition unit is configured to acquire text data related to the application field of the large model by integrating multiple types of data sources on the Internet through a multi-source data acquisition; An extraction unit is configured to extract key knowledge elements from the collected data and eliminate redundant and contradictory information by using knowledge extraction and fusion technology, and form a unified knowledge representation system; A construction unit is configured to automatically construct a knowledge base with rich semantic association based on the unified knowledge representation system according to a predefined structure mode, and introduce a data quality evaluation mechanism to ensure the reliability of the knowledge base; An enhancement unit is configured to enhance the semantic understanding ability of the large model by using deep semantic analysis technology based on the constructed knowledge base, including semantic encoding, knowledge graph embedding, semantic reasoning and knowledge completion.

9. An electronic device, characterized by It includes: At least one processor; and a memory connected in communication with the at least one processor; wherein the memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the automatic construction and semantic understanding enhancement method of the large model knowledge base in any one of claims 1-7.

10. A computer readable medium characterized by The computer readable storage medium stores computer instructions for enabling the processor to implement the automatic construction and semantic understanding enhancement method of the large model knowledge base in any one of claims 1-7 when executed.

Images