Analysis method and system integrating big model and knowledge graph technology

Abstract

This application provides an analysis method and system that integrates large models and knowledge graph technology, involving the field of deep learning. It first extracts the text content of original documents related to industrial technology from patent databases, academic literature libraries and policy documents, and uses this to construct an industrial technology knowledge graph. Then, the relevant information nodes are located through the industrial technology knowledge graph to extract industrial technology query matching content, and a large language model is used to perform significant semantic interaction analysis, and then an intelligent answering robot based on a large language model is used to generate question answering texts that match user consultations, thereby performing semantic interaction analysis and automatic answer generation on industrial technology-related documents and user consultation questions, providing more personalized and comprehensive question answering texts related to industrial technology analysis for different user consultation questions, and improving the performance of the question-answering system.

Description

Technical Field

[0001] The present application relates to the field of deep learning, and in particular to an analysis method and system that integrates large models and knowledge graph technology. Background Art

[0002] In today's society, the development of industrial technology is one of the key factors driving economic growth and social progress. With the continuous advancement of science and technology, new technological achievements emerge in an endless stream. How to effectively manage and utilize these information resources has become an important challenge facing enterprises and research institutions. In this context, users may have a variety of consulting questions, such as seeking the development history of a specific technology, understanding the latest research results, and exploring technology development trends.

[0003] However, traditional information retrieval and consultation answering methods usually rely on keyword retrieval, which is inefficient when facing massive data and is prone to missing relevant information. Especially when dealing with highly professional documents such as patent documents and academic papers, since the content of these documents often involves a large number of specialized terms, complex professional knowledge, and cross-domain related information, simple keyword matching often cannot accurately reflect the key content in industrial technology-related information, nor can it meet the actual needs of users. In addition, this traditional retrieval method is mainly based on text surface matching and lacks deep semantic understanding and reasoning capabilities. When users ask more complex questions, traditional methods often find it difficult to give satisfactory answers and are difficult to adapt to users' personalized needs. For example, when asking about the application prospects or potential risks of a certain technology, surface text information alone is not enough to provide a comprehensive answer. At the same time, different users may have different concerns about the same problem, but traditional methods often find it difficult to distinguish these differences, resulting in the information provided not meeting the actual needs of users. Summary of the invention

[0004] In order to solve the above technical problems, this application is proposed.

[0005] According to one aspect of the present application, an analysis method integrating a large model and knowledge graph technology is provided, which includes:

[0006] Obtain original documents related to industrial technology from patent databases, academic literature libraries and policy documents;

[0007] Extracting text content from original documents related to the industrial technology;

[0008] Based on the text content, construct an industrial technology knowledge graph, wherein the industrial technology knowledge graph includes nodes and edges between nodes;

[0009] Obtain user consultation questions;

[0010] Based on the user's inquiry question, extracting industrial technology query matching content from the industrial technology knowledge graph;

[0011] Encoding the user consultation question and the industry technology query matching content to obtain semantic coding features of the user consultation question and semantic coding features of the industry technology query matching content;

[0012] The semantic coding features of the user consultation questions and the semantic coding features of the industrial technology query matching content are subjected to significant interaction analysis between semantic features to obtain a fusion representation of the user consultation questions and industrial technology query matching content semantic significant interaction, including: performing joint implicit feature activation processing on the semantic coding features of the user consultation questions and the semantic coding features of the industrial technology query matching content to obtain semantic conditional features of the user consultation questions and industrial technology query matching content; taking the semantic conditional features of the user consultation questions and industrial technology query matching content as conditional information, performing feature interaction fusion on the semantic coding features of the user consultation questions and the semantic coding features of the industrial technology query matching content to obtain a fusion representation of the user consultation questions and industrial technology query matching content semantic significant interaction;

[0013] Based on the semantically significant interactive fusion representation of the user consultation question-industry technology query matching content, an intelligent answering robot based on a large language model is used to generate question answering text.

[0014] Based on the above-mentioned analysis method of integrating the big model and knowledge graph technology, the user consultation questions and the industrial technology query matching content are encoded to obtain semantic coding features of the user consultation questions and semantic coding features of the industrial technology query matching content, including: semantic embedding coding is performed on the user consultation questions and the industrial technology query matching content to obtain a user consultation question semantic coding vector as the user consultation question semantic coding feature and an industrial technology query matching content semantic coding vector as the industrial technology query matching content semantic coding feature.

[0015] Based on the above analysis method of integrating the big model and knowledge graph technology, the user consultation question semantic coding features and the industry technology query matching content semantic coding features are jointly activated with implicit features to obtain user consultation question-industry technology query matching content semantic condition features, including:

[0016] Inputting the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content into a joint implicit feature capture network to obtain a user consultation question-industry technology query matching content semantic joint implicit feature vector;

[0017] The user consultation question - industry technology query matching content semantic joint implicit feature vector is subjected to feature activation based on the Sigmoid function to obtain the user consultation question - industry technology query matching content semantic condition feature vector as the user consultation question - industry technology query matching content semantic condition feature.

[0018] Based on the above analysis method integrating the big model and knowledge graph technology, the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content are input into the joint implicit feature capture network to obtain the user consultation question-industry technology query matching content semantic joint implicit feature vector, including:

[0019] After adding the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content by position point, the obtained user consultation question-industrial technology query matching content semantic sum vector is multiplied by the weight matrix and then added by position with the bias vector to obtain the user consultation question-industrial technology query matching content semantic joint interaction vector;

[0020] The user consultation question-industry technology query matching content semantic joint interaction vector is processed using a tanh function to obtain the user consultation question-industry technology query matching content semantic joint implicit feature vector.

[0021] Based on the above-mentioned analysis method of the fusion big model and knowledge graph technology, the semantic conditional features of the user consultation question-industry technology query matching content are used as conditional information, and the semantic coding features of the user consultation question and the semantic coding features of the industry technology query matching content are interactively fused to obtain the semantically significant interactive fusion representation of the user consultation question-industry technology query matching content, including:

[0022] Calculate the user consultation question semantic contribution of the user consultation question semantic coding feature relative to the user consultation question-industry technology query matching content semantic condition feature vector;

[0023] Calculating the semantic contribution of the semantic coding feature of the industrial technology query matching content to the semantic condition feature vector of the user consultation question-industrial technology query matching content;

[0024] Normalizing the semantic contribution of the user consultation question and the semantic contribution of the industrial technology query matching content, and using the normalized semantic contribution of the user consultation question and the normalized semantic contribution of the industrial technology query matching content to weighted modulate the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content to obtain a modulated user consultation question semantic coding vector and a modulated industrial technology query matching content semantic coding vector;

[0025] Using the modulated user consultation question semantic coding vector as the query vector, the modulated industry technology query matching content semantic coding vector as the key vector and the user consultation question-industry technology query matching content semantic condition feature vector as the value vector, the modulated user consultation question semantic coding vector, the modulated industry technology query matching content semantic coding vector and the user consultation question-industry technology query matching content semantic condition feature vector are input into a feature-to-feature significant guided interaction module based on a converter structure to obtain a user consultation question-industry technology query matching content semantic significant interaction fusion representation vector as the user consultation question-industry technology query matching content semantic significant interaction fusion representation.

[0026] Based on the above analysis method integrating the big model and the knowledge graph technology, the user consultation question semantic contribution of the user consultation question semantic encoding feature relative to the user consultation question-industry technology query matching content semantic condition feature vector is calculated, including:

[0027] Calculate the positional division of the user consultation question semantic coding feature and the user consultation question-industry technology query matching content semantic condition feature vector to obtain the user consultation question semantic contribution vector;

[0028] Calculating the logarithmic function value with base 2 of the absolute value of each eigenvalue of the user consultation question semantic contribution vector to obtain the user consultation question semantic contribution logarithmic vector;

[0029] Calculate the positional dot product of the semantic coding feature of the user consultation question and the logarithmic vector of the semantic contribution of the user consultation question, and perform positional multiplication on the obtained dot product vector to obtain the semantic contribution value of the user consultation question;

[0030] An exponential function with the natural constant e as the base and the semantic contribution value of the user consultation question as the exponent is calculated to obtain the semantic contribution degree of the user consultation question.

[0031] Based on the above-mentioned analysis method of integrating the big model and the knowledge graph technology, the semantic contribution of the user consultation question and the semantic contribution of the industrial technology query matching content are normalized, and the normalized semantic contribution of the user consultation question and the normalized semantic contribution of the industrial technology query matching content are used to weighted modulate the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content to obtain the modulated user consultation question semantic coding vector and the modulated industrial technology query matching content semantic coding vector, including:

[0032] Calculate the sum of the semantic contribution of the user consultation question and the semantic contribution of the industry technology query matching content to obtain the total sum of the semantic contribution of the user consultation question-industry technology query matching content;

[0033] Respectively dividing the user consultation question semantic contribution and the industry technology query matching content semantic contribution by the total sum of the user consultation question-industry technology query matching content semantic contribution to obtain the normalized user consultation question semantic contribution and the normalized industry technology query matching content semantic contribution;

[0034] Multiplying the user consultation question semantic coding feature by the normalized user consultation question semantic contribution by position point to obtain the modulated user consultation question semantic coding vector;

[0035] The semantic coding feature of the industrial technology query matching content and the normalized semantic contribution of the industrial technology query matching content are multiplied by position point to obtain the modulated industrial technology query matching content semantic coding vector.

[0036] Based on the above-mentioned analysis method of the fusion big model and knowledge graph technology, the modulated user consultation question semantic coding vector is used as the query vector, the modulated industry technology query matching content semantic coding vector is used as the key vector, and the user consultation question-industry technology query matching content semantic condition feature vector is used as the value vector, and the modulated user consultation question semantic coding vector, the modulated industry technology query matching content semantic coding vector and the user consultation question-industry technology query matching content semantic condition feature vector are input into the feature-to-feature significant guided interaction module based on the converter structure to obtain the user consultation question-industry technology query matching content semantic significant interaction fusion representation vector as the user consultation question-industry technology query matching content semantic significant interaction fusion representation, including:

[0037] Perform vector multiplication of the modulated user consultation question semantic coding vector and the transposed vector of the modulated industry technology query matching content semantic coding vector, and divide the obtained modulated user consultation question-industry technology query matching content semantic association matrix by the square root of the length of the modulated industry technology query matching content semantic coding vector by position to obtain a modulated user consultation question-industry technology query matching content semantic association scaling matrix;

[0038] The modulated user consultation question-industry technology query matching content semantic association scaling matrix is ​​activated using a softmax function, and the obtained modulated user consultation question-industry technology query matching content semantic association scaling activation matrix is ​​multiplied by the user consultation question-industry technology query matching content semantic conditional feature vector to obtain the user consultation question-industry technology query matching content semantic significant interaction fusion representation vector.

[0039] Based on the above-mentioned analysis method of integrating large models and knowledge graph technology, based on the semantically significant interactive fusion representation of the user consultation question-industry technology query matching content, an intelligent answering robot based on a large language model is used to generate question answering text, including: inputting the semantically significant interactive fusion representation vector of the user consultation question-industry technology query matching content into the intelligent answering robot based on a large language model to obtain the question answering text.

[0040] According to another aspect of the present application, an analysis system integrating a large model and knowledge graph technology is provided, which includes:

[0041] The module for acquiring original documents related to industrial technology is used to acquire original documents related to industrial technology from patent databases, academic literature libraries and policy documents;

[0042] A text content extraction module, used to extract text content from the original documents related to the industrial technology;

[0043] An industrial technology knowledge graph construction module, used to construct an industrial technology knowledge graph based on the text content, wherein the industrial technology knowledge graph includes nodes and edges between nodes;

[0044] User consultation question acquisition module, used to obtain user consultation questions;

[0045] An industrial technology query matching content extraction module is used to extract industrial technology query matching content from the industrial technology knowledge graph based on the user's inquiry question;

[0046] A text content encoding module, used for encoding the user consultation question and the industry technology query matching content to obtain semantic encoding features of the user consultation question and semantic encoding features of the industry technology query matching content;

[0047] A significant interaction analysis module between semantic features, used to perform significant interaction analysis between semantic features on the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content to obtain a fusion representation of the significant interaction between the semantic features of the user consultation question and the industrial technology query matching content;

[0048] The question answering text generation module is used to generate question answering text based on the semantically significant interactive fusion representation of the user consultation question-industry technology query matching content, using an intelligent answering robot based on a large language model.

[0049] Compared with the prior art, the present application provides an analysis method and system that integrates large models and knowledge graph technology. It first extracts the text content of original documents related to industrial technology from patent databases, academic literature libraries and policy documents, and uses this to construct an industrial technology knowledge graph. Then, the relevant information nodes are located through the industrial technology knowledge graph to extract industrial technology query matching content, and a large language model is used to perform significant semantic interaction analysis. Then, an intelligent answering robot based on a large language model is used to generate question answering texts that match user consultations, thereby performing semantic interaction analysis and automatic answer generation on industrial technology-related documents and user consultation questions, providing more personalized and comprehensive question answering texts related to industrial technology analysis for different user consultation questions, and improving the performance of the question-answering system. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0051] Figure 1 It is a flowchart of the analysis method integrating the big model and the knowledge graph technology according to the embodiment of the present application.

[0052] Figure 2 This is a data flow diagram of the analysis method that integrates the big model and knowledge graph technology according to an embodiment of the present application.

[0053] Figure 3 A flowchart of performing significant interaction analysis between the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content in the analysis method of the fusion big model and knowledge graph technology according to the embodiment of the present application to obtain a fusion representation of the significant interaction of the semantics of the user consultation question and the industrial technology query matching content.

[0054] Figure 4 A flowchart for performing joint implicit feature activation processing on the semantic coding features of user consultation questions and the semantic coding features of industrial technology query matching content in the analysis method of the fusion big model and knowledge graph technology according to the embodiment of the present application to obtain the semantic conditional features of user consultation questions-industry technology query matching content.

[0055] Figure 5 This is a system block diagram of an analysis system that integrates a large model and knowledge graph technology according to an embodiment of the present application. DETAILED DESCRIPTION

[0056] Various exemplary embodiments, features and aspects of the present application will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0057] In today's context of the emergence of new technological achievements, traditional information resource keyword retrieval methods are difficult to efficiently and accurately meet users' diverse consulting needs when dealing with highly professional and complex documents, especially when it comes to deep semantic understanding and personalized needs, and there are obvious limitations.

[0058] In recent years, with the rapid development of big data and artificial intelligence technology, how to efficiently extract valuable knowledge from massive information has become an important research topic. Especially in the field of industrial technology, facing multi-source heterogeneous data from patent databases, academic literature libraries, policy documents, etc., how to extract useful information and build a corresponding knowledge system has become an important link in promoting technological innovation and development.

[0059] Traditional information retrieval and analysis methods often seem inadequate when faced with complex and diverse data sources. Multi-source heterogeneous knowledge graph technology can provide a comprehensive view of industrial technology at a macro level by integrating data from different sources, such as public information on the Internet, technical details in patent databases, research results in academic literature, and policy documents. This technology relies on advanced natural language processing (NLP) capabilities and data mining techniques to automatically or semi-automatically extract structured information from unstructured text and construct a knowledge graph containing entities and their relationships. The combination of mapping and extraction workflows enables a comprehensive analysis of industrial technology and improves a comprehensive understanding of industrial technology, which in turn helps to ensure the accuracy and completeness of the knowledge graph.

[0060] Although knowledge graphs provide rich structured information, they are limited in processing natural language queries and generating natural language responses. In contrast, large language models have powerful language understanding and generation capabilities after being trained with large-scale corpora, but lack expertise in specific fields. Therefore, integrating knowledge graphs with large language models can give the system both the flexibility of language models and the professionalism of knowledge graphs. By embedding knowledge graph data into the large model training process, or as an external tool working with the large model in the inference phase, this solution can achieve a deep understanding and accurate response to industry technology-related information.

[0061] Based on this, in the technical solution of this application, an analysis method integrating a large model and knowledge graph technology is proposed, which extracts the text content of original documents related to industrial technology from patent databases, academic literature libraries and policy documents, and constructs an industrial technology knowledge graph based on this. In this way, when facing user consultations, the relevant information nodes are located through the industrial technology knowledge graph to extract industrial technology query matching content, and the large language model is used to perform significant semantic interaction analysis on the industrial technology query matching content and the user consultation questions, and then the intelligent answer robot based on the large language model is used to generate question answer texts that match the user consultation. In this way, the large model and knowledge graph technology can be integrated to perform semantic interaction analysis and automatic answer generation on industrial technology related documents and user consultation questions, thereby providing more personalized and comprehensive question answer texts related to industrial technology analysis for different user consultation questions, which can further improve the performance of the question-answering system by intelligent analysis and generation.

[0062] Specifically, this application proposes an analysis method that integrates large models and knowledge graph technology. Figure 1 It is a flowchart of the analysis method integrating the big model and the knowledge graph technology according to the embodiment of the present application. Figure 2 Schematic diagram of data flow of the analysis method integrating large model and knowledge graph technology according to the embodiment of the present application. Figure 1 and Figure 2As shown, according to the analysis method of the fusion big model and knowledge graph technology of the embodiment of the present application, it includes: step S110, obtaining original documents related to industrial technology from patent databases, academic literature libraries and policy documents; step S120, extracting text content from the original documents related to industrial technology; step S130, based on the text content, constructing an industrial technology knowledge graph, wherein the industrial technology knowledge graph includes nodes and edges between nodes; step S140, obtaining user consultation questions; step S150, extracting industrial technology query matching content from the industrial technology knowledge graph based on the user consultation questions; step S1 Step 160, encode the user consultation question and the industrial technology query matching content to obtain the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content; step S170, perform significant interaction analysis between the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content to obtain a fusion representation of the user consultation question-industry technology query matching content semantic significant interaction; step S180, based on the fusion representation of the user consultation question-industry technology query matching content semantic significant interaction, use an intelligent answering robot based on a large language model to generate a question answering text.

[0063] In the above-mentioned analysis method integrating the big model and the knowledge graph technology, the step 110 obtains original documents related to industrial technology from patent databases, academic literature libraries and policy documents. It should be understood that the patent application and grant information recorded in the patent database not only shows the latest technological development trends, but also reveals the technological layout and market trends of competitors. The academic literature library provides cutting-edge research results and theoretical basis. Policy documents reflect the policy orientation and development plans of the country and the industry. These resources contain rich industry information and technical details, which can provide basic support for technological research and development and innovation.

[0064] In the above analysis method integrating the big model and the knowledge graph technology, the step 120 extracts text content from the original file related to the industrial technology. It should be understood that considering the rapid development of industrial technology, which results in a huge amount of information, extracting text content can transform massive data into actionable information, helping to quickly grasp the current state of technology and cutting-edge trends.

[0065] In the above-mentioned analysis method integrating the large model and the knowledge graph technology, the step 130 constructs an industrial technology knowledge graph based on the text content, and the industrial technology knowledge graph includes nodes and edges between nodes. It should be understood that since the text content extracted from the original file related to the industrial technology usually contains a large amount of complex and professional information, this information is distributed in different documents and has complex associations with each other. In order to combine these semantic information about the text content of the original file to form a structured network, it is convenient for the subsequent understanding of user consultation questions and related information query, in the technical solution of the present application, the industrial technology knowledge graph is further constructed based on the text content, and the industrial technology knowledge graph includes nodes and edges between nodes. In particular, the core of the industrial technology knowledge graph is to construct entities (nodes) and the relationships (edges) between them. Through natural language processing technologies such as entity recognition, relationship extraction, etc., key knowledge points such as inventors, patent numbers, technical classifications, etc. can be automatically extracted from the text, and logical relationships between them can be established. In other words, the industrial technology knowledge graph not only stores the information of the entity itself, but more importantly, it stores the relationship between entities. Such structured data supports more complex semantic queries, and users can obtain deeper information by querying specific relationships rather than just keyword matching.

[0066] In the above analysis method integrating the big model and the knowledge graph technology, the step 140 is to obtain user consultation questions. It should be understood that obtaining user consultation questions can directly reflect the needs, concerns and questions of users, thereby providing a basis for product optimization and service improvement. By analyzing user questions, common problems and potential market needs can be identified, thereby guiding subsequent technology development and customer service strategies.

[0067] In the above-mentioned analysis method integrating the large model and the knowledge graph technology, the step 150 extracts the industrial technology query matching content from the industrial technology knowledge graph based on the user consultation question. It should be understood that matching with the content in the knowledge graph can more accurately understand the user's needs under a specific industrial technology background and related information, thereby providing more appropriate answers that meet the user's needs. This precise understanding is incomparable to traditional keyword search because it takes into account the context of the question and the industrial technology context. In other words, the knowledge graph not only stores entity information, but more importantly, it stores the relationship between entities. By extracting industrial technology query matching content that matches the user's question from the knowledge graph, an answer that is more relevant to the user's consultation question can be provided. Because the information in the knowledge graph is organized in a structured manner, it can better reflect the connection between various industrial technology-related knowledge points, thereby providing more complete and industrial technology-related background information when answering user questions.

[0068] In the above-mentioned analysis method integrating the big model and the knowledge graph technology, the step 160 encodes the user consultation question and the industrial technology query matching content to obtain the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content. It should be understood that considering that the user consultation question and the query matching content usually exist in the form of natural language, their structure is complex and contains a variety of semantic information. Directly comparing them may lead to ambiguity in understanding and information loss. Therefore, in the technical solution of the present application, by encoding the user consultation question and the industrial technology query matching content to convert them into a unified numerical vector representation, the diversity of language expressions is effectively eliminated, ensuring that the comparison between different texts is more direct and objective.

[0069] In an embodiment of the present application, the step 160, encoding the user consultation question and the industrial technology query matching content to obtain the user consultation question semantic coding features and the industrial technology query matching content semantic coding features, includes: semantically embedding the user consultation question and the industrial technology query matching content to obtain the user consultation question semantic coding vector as the user consultation question semantic coding features and the industrial technology query matching content semantic coding vector as the industrial technology query matching content semantic coding features. In this way, when processing user consultation questions, the semantic coding method can capture the question intent of the user consultation questions, and when processing industrial technology query matching content, it can better understand the main semantics of the matching content and the industrial technology-related information, thereby improving the accuracy and relevance of answering questions.

[0070] In the above-mentioned analysis method integrating the large model and the knowledge graph technology, the step 170 performs a significant interaction analysis between the semantic features of the user consultation question semantic coding features and the semantic coding features of the industrial technology query matching content to obtain a user consultation question-industry technology query matching content semantic significant interaction fusion representation. It should be understood that since the user consultation question semantic coding features and the industrial technology query matching content semantic coding features respectively express the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content, in order to further capture the semantic feature information related to the user consultation at a higher level from the industrial technology query matching content semantics, so as to subsequently use the large language model to generate an answer text that is more in line with the user consultation question and meets the user's needs, it is necessary to perform a significant interaction analysis between the semantic features of the user consultation question semantic coding features and the industrial technology query matching content semantic coding features to obtain a user consultation question-industry technology query matching content semantic significant interaction fusion representation. It is worth mentioning that the user consultation question-industry technology query matching content semantic significant interaction fusion representation can better reflect the essential relationship and significant interaction fusion characteristics between the user consultation question and the industrial technology query matching content. It not only contains the original semantic feature information of user consultation questions and the semantic feature information of industrial technology query matching content, but also integrates the potential complex semantic associations and interactive information between the two, strengthens the semantic feature expression of industrial technology query matching content related to user questions and the importance semantics related to the subsequent answer text generation task, and reduces the interference of irrelevant information, thus providing a more accurate basis for the subsequent question answer text generation. This is crucial to improving the accuracy and comprehensiveness of question answer text generation, because it ensures that the answer text of the large language model can be generated based on the most relevant user consultation question-industry technology query matching content interactive semantic information.

[0071] Figure 3 This is a flowchart of performing significant interaction analysis between the semantic coding features of the user consultation question and the semantic coding features of the industry technology query matching content in the analysis method of the fusion big model and knowledge graph technology according to the embodiment of the present application to obtain the user consultation question-industry technology query matching content semantic significant interaction fusion representation. Figure 3As shown, in an embodiment of the present application, a significant interaction analysis between semantic features of the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content is performed to obtain a fusion representation of the user consultation question-industry technology query matching content semantic significant interaction, including: step S171, performing a joint implicit feature activation process on the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content to obtain the semantic conditional features of the user consultation question-industry technology query matching content; step S172, taking the semantic conditional features of the user consultation question-industry technology query matching content as conditional information, performing feature interaction fusion on the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content to obtain the fusion representation of the user consultation question-industry technology query matching content semantic significant interaction.

[0072] Figure 4 This is a flowchart of performing joint implicit feature activation processing on the semantic coding features of user consultation questions and the semantic coding features of industrial technology query matching content in the analysis method of the fusion big model and knowledge graph technology according to the embodiment of the present application to obtain the semantic conditional features of user consultation questions-industry technology query matching content. Figure 4 As shown, in an embodiment of the present application, the step S171, performs a joint implicit feature activation process on the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content to obtain the user consultation question-industry technology query matching content semantic conditional features, including: step S1711, inputting the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content into a joint implicit feature capture network to obtain a user consultation question-industry technology query matching content semantic joint implicit feature vector; step S1712, performs a Sigmoid function-based feature activation on the user consultation question-industry technology query matching content semantic joint implicit feature vector to obtain a user consultation question-industry technology query matching content semantic conditional feature vector as the user consultation question-industry technology query matching content semantic conditional features.

[0073] Specifically, in step S1711, the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content are input into the joint implicit feature capture network to obtain the user consultation question-industry technology query matching content semantic joint implicit feature vector. It should be understood that the user consultation question semantic coding vector and the industrial technology query matching content semantic coding vector are input into the joint implicit feature capture network, and the potential association between the two can be extracted. Among them, the joint implicit feature capture network can learn the intrinsic connection between the two sets of input features through deep learning technology, and identify and integrate the shared implicit structure in the input features. Specifically, when the semantic coding vectors of user consultation and industrial technology content are input, the network will analyze the similarity and correlation between them, and extract representative features to obtain the user consultation question-industry technology query matching content semantic joint implicit feature vector. This new vector not only contains the original input information, but also can capture the intrinsic connection between user consultation questions and industrial technology queries, revealing information that is not easy to detect when it exists alone, and enhancing the relevance of information.

[0074] In an embodiment of the present application, the step S1711, inputs the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content into a joint implicit feature capture network to obtain a user consultation question-industry technology query matching content semantic joint implicit feature vector, including: adding the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content by position point, multiplying the obtained user consultation question-industry technology query matching content semantic sum vector with the weight matrix and then adding it by position with the bias vector to obtain a user consultation question-industry technology query matching content semantic joint interaction vector; using a tanh function to process the user consultation question-industry technology query matching content semantic joint interaction vector to obtain the user consultation question-industry technology query matching content semantic joint implicit feature vector.

[0075] Specifically, in step S1712, the user consultation question-industry technology query matching content semantic joint implicit feature vector is activated based on the Sigmoid function to obtain the user consultation question-industry technology query matching content semantic condition feature vector as the user consultation question-industry technology query matching content semantic condition feature. It should be understood that the user consultation question semantic encoding vector and the industry technology query matching content semantic encoding vector are input into the joint implicit feature capture network, which can effectively capture the potential association between the two features. However, the obtained user consultation question-industry technology query matching content semantic joint implicit feature vector may contain a large amount of redundant information and noise, so it is necessary to perform feature activation based on the Sigmoid function to suppress irrelevant noise information and refine implicit information. Among them, the Sigmoid function maps the feature value to between 0 and 1, so that the relative importance of each feature can be highlighted. Through this activation, the model can evaluate the impact of the feature at a more detailed level, thereby emphasizing those parts that are most critical to the user consultation and industry technology query matching. This activation mechanism can not only suppress the interference of irrelevant features, but also effectively reduce the negative impact of extreme values ​​on the model, ensuring the stability of the feature in subsequent processing. Through this activation process, important features will be emphasized, thereby enhancing the model's understanding and expression of the relationship between user inquiries and industry technical content.

[0076] In an embodiment of the present application, the step S172, taking the semantic conditional features of the user consultation question-industry technology query matching content as conditional information, performs feature interactive fusion on the semantic coding features of the user consultation question and the semantic coding features of the industry technology query matching content to obtain the semantically significant interactive fusion representation of the user consultation question-industry technology query matching content, including: step S1721, calculating the user consultation question semantic contribution of the user consultation question semantic coding features relative to the user consultation question-industry technology query matching content semantic conditional feature vector; step S1722, calculating the industry technology query matching content semantic contribution of the industry technology query matching content semantic coding features relative to the user consultation question-industry technology query matching content semantic conditional feature vector; step S1723, normalizing the user consultation question semantic contribution and the industry technology query matching content semantic contribution, and using the normalized user consultation question semantic contribution and the normalized The semantic contribution of the industrial technology query matching content is weightedly modulated on the user consultation question semantic coding features and the industrial technology query matching content semantic coding features to obtain the modulated user consultation question semantic coding vector and the modulated industrial technology query matching content semantic coding vector; step S1724, using the modulated user consultation question semantic coding vector as the query vector, the modulated industrial technology query matching content semantic coding vector as the key vector and the user consultation question-industrial technology query matching content semantic condition feature vector as the value vector, the modulated user consultation question semantic coding vector, the modulated industrial technology query matching content semantic coding vector and the user consultation question-industrial technology query matching content semantic condition feature vector are input into the feature-significant guided interaction module based on the converter structure to obtain the user consultation question-industrial technology query matching content semantic significant interaction fusion representation vector as the user consultation question-industrial technology query matching content semantic significant interaction fusion representation.

[0077] Specifically, in step S1721, the user consultation question semantic contribution of the user consultation question semantic coding feature relative to the user consultation question-industry technology query matching content semantic condition feature vector is calculated. It should be understood that in the technical solution of the present application, in order to determine the importance of the semantic features of the user consultation question to the final representation, the user consultation question semantic contribution of the user consultation question semantic coding vector relative to the user consultation question-industry technology query matching content semantic condition feature vector is calculated. In this way, the importance of user questions in information matching can be evaluated.

[0078] In an embodiment of the present application, the step S1721, calculating the user consultation question semantic contribution of the user consultation question semantic coding feature relative to the user consultation question-industry technology query matching content semantic condition feature vector, includes: calculating the positional division of the user consultation question semantic coding feature and the user consultation question-industry technology query matching content semantic condition feature vector to obtain the user consultation question semantic contribution vector; calculating the logarithmic function value with base two of the absolute value of each eigenvalue of the user consultation question semantic contribution vector to obtain the user consultation question semantic contribution logarithmic vector; calculating the positional dot product of the user consultation question semantic coding feature and the user consultation question semantic contribution logarithmic vector, and performing positional multiplication on the obtained dot product vector to obtain the user consultation question semantic contribution value; calculating an exponential function with the natural constant e as the base, with the user consultation question semantic contribution value as the exponent, to obtain the user consultation question semantic contribution degree.

[0079] Similarly, in step S1722, the semantic contribution of the industrial technology query matching content semantic coding features relative to the user consultation question-industry technology query matching content semantic condition feature vector is calculated, which can quantify the relationship between industrial technology queries and user consultation questions-industry technology query matching content semantic condition feature vectors, and determine which queries contribute the most to the relevance and accuracy of user consultations. In this way, the system can capture the user's intentions more effectively. It is worth mentioning that in order to obtain a more accurate weight distribution, in other implementation schemes of the present application, it is considered to introduce techniques such as mutual information estimation and canonical correlation analysis (CCA) to improve the quality of the final representation and enhance the model's ability to resist noise interference.

[0080] Specifically, in step S1723, the semantic contribution of the user consultation question and the semantic contribution of the industrial technology query matching content are normalized, and the normalized semantic contribution of the user consultation question and the normalized semantic contribution of the industrial technology query matching content are used to weighted modulate the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content to obtain the modulated user consultation question semantic coding vector and the modulated industrial technology query matching content semantic coding vector. It should be understood that, considering that the original semantic contribution of the user consultation question and the semantic contribution of the industrial technology query matching content may not be comparable due to different scales, normalization is required to ensure that the weights are reasonably allocated in the subsequent weighted modulation, so that the more important semantic features of the user consultation question and the semantic features of the industrial technology query matching content occupy a larger weight in the final feature representation, and appropriately reduce the influence of less relevant components.

[0081] In an embodiment of the present application, the step S1723 normalizes the semantic contribution of the user consultation question and the semantic contribution of the industrial technology query matching content, and uses the normalized semantic contribution of the user consultation question and the normalized semantic contribution of the industrial technology query matching content to weighted modulate the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content to obtain a modulated user consultation question semantic coding vector and a modulated industrial technology query matching content semantic coding vector, including: calculating the sum of the semantic contribution of the user consultation question and the semantic contribution of the industrial technology query matching content to obtain the total sum of the semantic contribution of the user consultation question-industry technology query matching content value; respectively divide the user consultation question semantic contribution and the industry technology query matching content semantic contribution by the total sum of the user consultation question-industry technology query matching content semantic contribution to obtain the normalized user consultation question semantic contribution and the normalized industry technology query matching content semantic contribution; perform position point multiplication on the user consultation question semantic coding feature and the normalized user consultation question semantic contribution to obtain the modulated user consultation question semantic coding vector; perform position point multiplication on the industry technology query matching content semantic coding feature and the normalized industry technology query matching content semantic contribution to obtain the modulated industry technology query matching content semantic coding vector.

[0082] Specifically, in the step S1724, the modulated user consultation question semantic coding vector is used as the query vector, the modulated industry technology query matching content semantic coding vector is used as the key vector, and the user consultation question-industry technology query matching content semantic condition feature vector is used as the value vector. The modulated user consultation question semantic coding vector, the modulated industry technology query matching content semantic coding vector and the user consultation question-industry technology query matching content semantic condition feature vector are input into the feature-significant guided interaction module based on the converter structure to obtain the user consultation question-industry technology query matching content semantic significant interaction fusion representation vector as the user consultation question-industry technology query matching content semantic significant interaction fusion representation. It should be understood that weighted modulation alone is not enough to fully reflect the deep relationship between the query and the matching content. Therefore, by using the modulated user consultation question semantic coding vector as the query vector, the industry technology query matching content semantic coding vector as the key vector, and the user consultation question-industry technology query matching content semantic condition feature vector as the value vector, an effective information retrieval framework is constructed to enable the system to better understand and process the relationship between user intentions and industry information. Among them, the query vector reflects the user's needs through the user's questions, while the key vector represents the characteristics of the industrial technology query. This combination allows the system to quickly locate relevant information during retrieval. At the same time, the conditional feature vector as the value vector provides contextual information and enhances the semantic understanding ability. By inputting these vectors into the feature-to-feature significant guided interaction module based on the converter structure, the system can realize complex feature interactions and extract deeper correlations, thereby generating a significant interaction fusion representation vector of the user consultation question-industrial technology query matching content, so as to examine the interaction between different features from multiple perspectives at the same time, automatically discover the most valuable semantic features, and thus improve the overall expression ability.

[0083] In an embodiment of the present application, the step S1724 uses the modulated user consultation question semantic coding vector as the query vector, the modulated industry technology query matching content semantic coding vector as the key vector and the user consultation question-industry technology query matching content semantic condition feature vector as the value vector, and inputs the modulated user consultation question semantic coding vector, the modulated industry technology query matching content semantic coding vector and the user consultation question-industry technology query matching content semantic condition feature vector into a feature-to-feature significant guided interaction module based on a converter structure to obtain a user consultation question-industry technology query matching content semantic significant interaction fusion representation vector as the user consultation question-industry technology query matching content semantic significant interaction fusion representation, including: inputting the modulated user consultation question semantic coding vector with The transposed vector of the modulated industrial technology query matching content semantic coding vector is vector-multiplied, and the obtained modulated user consultation question-industrial technology query matching content semantic association matrix is ​​divided by the square root of the length of the modulated industrial technology query matching content semantic coding vector by position to obtain the modulated user consultation question-industrial technology query matching content semantic association scaling matrix; the modulated user consultation question-industrial technology query matching content semantic association scaling matrix is ​​activated using a softmax function, and the obtained modulated user consultation question-industrial technology query matching content semantic association scaling activation matrix is ​​multiplied by the user consultation question-industrial technology query matching content semantic conditional feature vector to obtain the user consultation question-industrial technology query matching content semantic significant interaction fusion representation vector.

[0084] Specifically, the following semantic feature significant interaction formula is used to perform significant interaction analysis between the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content to obtain the semantic significant interaction fusion representation of the user consultation question-industrial technology query matching content;

[0085] Among them, the semantic feature significant interaction formula is: ; ; ; ; ; ; ; ; ;in, and are the semantic coding features of the user's consultation question and the semantic coding features of the industry technology query matching content, respectively. To add by location, and are the weight matrix and bias vector respectively, for function, Matching the content semantics and implicit feature vector of the user's inquiry question-industry technology query, for function, represents the semantic condition feature vector of the user consultation question-industry technology query matching content, The feature values ​​of each position in the semantic coding feature of the user's inquiry question, matching the feature values ​​of each position in the semantic coding feature of the content for the industrial technology query, The feature values ​​of each position in the semantic condition feature vector of the user's inquiry question-industry technology query match content, The number of feature values ​​in the semantic coding feature of the user's inquiry question and the semantic coding feature of the industry technology query matching content, represents the logarithmic function value with base 2, represents an exponential function with the natural constant e as the base, Semantic contribution to user consultation questions, Matching content semantic contribution for industrial technology queries, and They are the normalized semantic contribution of user consultation questions and the normalized semantic contribution of industrial technology query matching content. and They are the modulated user consultation question semantic coding vector and the modulated industry technology query matching content semantic coding vector. yes The transposed vector of yes Length, represents vector multiplication, yes function, It is the semantically significant interaction fusion representation vector of the user consultation question-industry technology query matching content.

[0086] In the above analysis method integrating large models and knowledge graph technology, the step 180 generates question answer texts using an intelligent answer robot based on a large language model based on the semantically significant interaction fusion representation of the user consultation question-industry technology query matching content. In this way, by effectively combining user intentions with industry knowledge, the intelligent answer robot becomes more intelligent in information services, improving the overall user experience.

[0087] In an embodiment of the present application, the step 180, based on the semantic significant interaction fusion representation of the user consultation question-industry technology query matching content, generates a question answer text using an intelligent answer robot based on a large language model, including: inputting the user consultation question-industry technology query matching content semantic significant interaction fusion representation vector into the intelligent answer robot based on a large language model to obtain the question answer text. That is, the intelligent answer robot based on the large language model is used to process the significant interaction fusion feature representation between the user consultation question semantics and the industry technology query matching content semantics, so as to automatically generate a question answer text that matches the user consultation question using the large language generation technology. In this way, the large model and knowledge graph technology can be integrated to perform semantic interaction analysis and automatic answer generation on industrial technology related documents and user consultation questions, thereby providing different user consultation questions with more personalized and comprehensive question answer texts related to industrial technology analysis, which can further improve the performance of the question-answering system using intelligent analysis and generation methods.

[0088] Preferably, the user consultation question-industry technology query matching content semantically significant interaction fusion representation vector is input into an intelligent answering robot based on a large language model to obtain a question answer text, including: determining a user consultation question-industry technology query matching content semantically significant interaction fusion association response matrix and a user consultation question-industry technology query matching content semantically significant interaction fusion distance response matrix based on the square root of the association value of the eigenvalue of the user consultation question-industry technology query matching content semantically significant interaction fusion representation vector, that is: ;in, The first one represents the semantically significant interaction fusion representation vector of the user consultation question-industry technology query matching content eigenvalues, The first one represents the semantically significant interaction fusion representation vector of the user consultation question-industry technology query matching content eigenvalues, Indicates the first ( , ) eigenvalues, Indicates the user consultation question-industry technology query matching content semantic significant interaction fusion distance response matrix ( , ) eigenvalues, ;

[0089] The user consultation question-industry technology query matching content semantically significant interaction fusion representation vector is matrix-multiplied with the user consultation question-industry technology query matching content semantically significant interaction fusion association response matrix to obtain the user consultation question-industry technology query matching content semantically significant interaction fusion association response vector; the user consultation question-industry technology query matching content semantically significant interaction fusion distance response matrix is ​​matrix-multiplied with the transposed vector of the user consultation question-industry technology query matching content semantically significant interaction fusion representation vector and then transposed to obtain the user consultation question-industry technology query matching content semantically significant interaction fusion distance response vector; the user consultation question-industry technology query matching content semantically significant interaction fusion association response matrix and the user consultation question-industry technology query matching content semantically significant interaction fusion distance response matrix are matrix-multiplied with the transposed vector of the user consultation question-industry technology query matching content semantically significant interaction fusion representation vector and then transposed to obtain the user consultation question-industry technology query matching content semantically significant interaction fusion distance response vector. After performing dot multiplication on the distance response matrix, perform matrix multiplication with the transposed vector of the user consultation question-industry technology query matching content semantically significant interaction fusion representation vector to obtain the user consultation question-industry technology query matching content semantically significant interaction fusion bias vector; perform dot multiplication on the user consultation question-industry technology query matching content semantically significant interaction fusion associated response vector, the user consultation question-industry technology query matching content semantically significant interaction fusion distance response vector and the user consultation question-industry technology query matching content semantically significant interaction fusion bias vector to obtain an optimized user consultation question-industry technology query matching content semantically significant interaction fusion representation vector; input the optimized user consultation question-industry technology query matching content semantically significant interaction fusion representation vector into an intelligent answering robot based on a large language model to obtain a question answer text.

[0090] Here, the optimized user consultation question-industry technology query matching content semantic significant interaction fusion representation vector It is expressed as: ;exist ,and In the case of: ; ;in, Represents the semantically significant interaction fusion representation vector of user consultation question-industry technology query matching content, Represents the user consultation question-industry technology query matching content semantic significant interaction fusion correlation response matrix, Represents the user consultation question-industry technology query matching content semantic significant interaction fusion distance response matrix, represents the transposed vector of a vector, It means point multiplication by position. represents a real number, represents the length of the vector, represents the semantic feature vector of the matching content between the first user's consultation question and the industry technology query, represents the semantically significant interaction fusion bias vector of user consultation question-industry technology query matching content, Indicates adding by position point. Representation optimized user consultation question - industry technology query matching content semantically significant interaction fusion representation vector.

[0091] That is, when the semantic coding features of the user consultation questions and the semantic coding features of the industrial technology query matching content respectively represent the semantic embedded coding features of the user consultation questions and the industrial technology query matching content, when performing significant guided interaction between features with the joint implicit features as conditional information, the semantic feature differences of the embedded encoded semantic features caused by the semantic differences of the source texts will also lead to local association gaps in the joint implicit conditions, resulting in a certain area of ​​simple repetition of the significant guided interaction between features of the encoded semantic features, affecting the interactive semantic logical dependency and reducing the text quality of the generated results based on the interactive semantic distribution.

[0092] Therefore, by using the user consultation question - industry technology query matching content semantically significant interaction fusion representation vector as the statistical reference-free distribution response framework of the user consultation question - industry technology query matching content semantically significant interaction fusion representation vector based on the self-association matrix and the self-distance matrix, the simple repetition of feature distribution is avoided based on the retrieval-enhanced reverse response construction of the user consultation question - industry technology query matching content semantically significant interaction fusion representation vector, and by ensuring the intrinsic mapping logic of the user consultation question - industry technology query matching content semantically significant interaction fusion representation vector based on the retrieval-response context, the surface combinatorial mapping of the user consultation question - industry technology query matching content semantically significant interaction fusion representation vector is avoided, so as to achieve the logical dependency mapping of the user consultation question - industry technology query matching content semantically significant interaction fusion representation vector to the generated target domain while maintaining the intuitive response relevance, thereby improving the text quality of the question answer text obtained by inputting the user consultation question - industry technology query matching content semantically significant interaction fusion representation vector into an intelligent answer robot based on a large language model. In this way, the big model and knowledge graph technology can be integrated to conduct semantic interactive analysis and automatic answer generation for industry technology-related documents and user consultation questions, thereby providing more personalized and comprehensive question answer texts related to industry technology analysis for different user consultation questions.

[0093] In summary, the analysis method based on the fusion of large model and knowledge graph technology in the embodiment of the present application is explained, which first extracts the text content of original documents related to industrial technology from patent databases, academic literature libraries and policy documents, and uses this to construct an industrial technology knowledge graph. Then, the relevant information nodes are located through the industrial technology knowledge graph to extract the matching content of industrial technology queries, and a large language model is used to perform significant semantic interaction analysis, and then an intelligent answering robot based on a large language model is used to generate question answering texts that match user consultations, thereby performing semantic interaction analysis and automatic answer generation on industrial technology-related documents and user consultation questions, providing more personalized and comprehensive question answering texts related to industrial technology analysis for different user consultation questions, and improving the performance of the question-answering system.

[0094] Figure 5 : is a system block diagram of an analysis system integrating a large model and knowledge graph technology according to an embodiment of the present application. Figure 5 As shown, according to the analysis system 100 integrating the big model and knowledge graph technology of the embodiment of the present application, it includes: an industrial technology-related original document acquisition module 110, which is used to acquire industrial technology-related original documents from patent databases, academic literature libraries and policy documents; a text content extraction module 120, which is used to extract text content from the industrial technology-related original documents; an industrial technology knowledge graph construction module 130, which is used to construct an industrial technology knowledge graph based on the text content, and the industrial technology knowledge graph includes nodes and edges between nodes; a user consultation question acquisition module 140, which is used to acquire user consultation questions; an industrial technology query matching content extraction module 150, which is used to extract information from the industrial technology knowledge graph based on the user consultation questions. Industrial technology query matching content; a text content encoding module 160, used to encode the user consultation question and the industrial technology query matching content to obtain the semantic encoding features of the user consultation question and the semantic encoding features of the industrial technology query matching content; a significant interaction analysis module 170, used to perform significant interaction analysis between semantic features on the semantic encoding features of the user consultation question and the semantic encoding features of the industrial technology query matching content to obtain a user consultation question-industry technology query matching content semantic significant interaction fusion representation; a question answer text generation module 180, used to generate a question answer text based on the user consultation question-industry technology query matching content semantic significant interaction fusion representation using an intelligent answer robot based on a large language model.

[0095] Here, those skilled in the art can understand that the specific operations of each step in the above analysis system integrating the large model and the knowledge graph technology have been referred to above. Figures 1 to 4 The description of the analysis method of integrating the big model and the knowledge graph technology has been introduced in detail, and therefore, its repeated description will be omitted.

[0096] As described above, the analysis system 100 integrating the large model and the knowledge graph technology according to the embodiment of the present application can be implemented in various terminal devices. In one example, the analysis system 100 integrating the large model and the knowledge graph technology can be integrated into the terminal device as a software module and / or a hardware module. For example, the analysis system 100 integrating the large model and the knowledge graph technology can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the analysis system 100 integrating the large model and the knowledge graph technology can also be one of the many hardware modules of the terminal device.

[0097] Alternatively, in another example, the analysis system 100 integrating the big model and the knowledge graph technology and the terminal device may be separate devices, and the analysis system 100 integrating the big model and the knowledge graph technology may be connected to the terminal device via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0098] In summary, the analysis system based on the fusion of large model and knowledge graph technology in the embodiment of the present application is explained, which first extracts the text content of original documents related to industrial technology from patent databases, academic literature libraries and policy documents, and uses this to construct an industrial technology knowledge graph. Then, the relevant information nodes are located through the industrial technology knowledge graph to extract the matching content of industrial technology queries, and a large language model is used to perform significant semantic interaction analysis, and then an intelligent answering robot based on a large language model is used to generate question answering texts that match user consultations, thereby performing semantic interaction analysis and automatic answer generation on industrial technology-related documents and user consultation questions, providing more personalized and comprehensive question answering texts related to industrial technology analysis for different user consultation questions, and improving the performance of the question-answering system.

Claims

1. An analysis method integrating a large model and knowledge graph technology, characterized in that: include: Obtain original documents related to industrial technology from patent databases, academic literature libraries and policy documents; Extracting text content from original documents related to the industrial technology; Based on the text content, construct an industrial technology knowledge graph, wherein the industrial technology knowledge graph includes nodes and edges between nodes; Obtain user consultation questions; Based on the user's inquiry question, extracting industrial technology query matching content from the industrial technology knowledge graph; Encoding the user consultation question and the industry technology query matching content to obtain semantic coding features of the user consultation question and semantic coding features of the industry technology query matching content; The semantic coding features of the user consultation questions and the semantic coding features of the industrial technology query matching content are subjected to significant interaction analysis between semantic features to obtain a fusion representation of the user consultation questions and industrial technology query matching content semantic significant interaction, including: performing joint implicit feature activation processing on the semantic coding features of the user consultation questions and the semantic coding features of the industrial technology query matching content to obtain semantic conditional features of the user consultation questions and industrial technology query matching content; taking the semantic conditional features of the user consultation questions and industrial technology query matching content as conditional information, performing feature interaction fusion on the semantic coding features of the user consultation questions and the semantic coding features of the industrial technology query matching content to obtain a fusion representation of the user consultation questions and industrial technology query matching content semantic significant interaction; Based on the semantically significant interactive fusion representation of the user consultation question-industry technology query matching content, an intelligent answering robot based on a large language model is used to generate a question answering text; The user consultation question semantic coding features and the industry technology query matching content semantic coding features are jointly implicitly activated to obtain the user consultation question-industry technology query matching content semantic condition features, including: Inputting the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content into a joint implicit feature capture network to obtain a user consultation question-industry technology query matching content semantic joint implicit feature vector; Performing feature activation based on the Sigmoid function on the semantic joint implicit feature vector of the user consultation question-industry technology query matching content to obtain a semantic condition feature vector of the user consultation question-industry technology query matching content as the semantic condition feature of the user consultation question-industry technology query matching content; The user consultation question semantic coding features and the industry technology query matching content semantic coding features are input into a joint implicit feature capture network to obtain a user consultation question-industry technology query matching content semantic joint implicit feature vector, including: After adding the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content by position point, the obtained user consultation question-industrial technology query matching content semantic sum vector is multiplied by the weight matrix and then added by position with the bias vector to obtain the user consultation question-industrial technology query matching content semantic joint interaction vector; Using a tanh function to process the user consultation question-industry technology query matching content semantic joint interaction vector to obtain the user consultation question-industry technology query matching content semantic joint implicit feature vector; Among them, taking the semantic conditional features of the user consultation question-industry technology query matching content as conditional information, performing feature interactive fusion on the semantic coding features of the user consultation question and the semantic coding features of the industry technology query matching content to obtain the semantic significant interactive fusion representation of the user consultation question-industry technology query matching content, including: Calculate the user consultation question semantic contribution of the user consultation question semantic coding feature relative to the user consultation question-industry technology query matching content semantic condition feature vector; Calculating the semantic contribution of the semantic coding feature of the industrial technology query matching content to the semantic condition feature vector of the user consultation question-industrial technology query matching content; Normalizing the semantic contribution of the user consultation question and the semantic contribution of the industrial technology query matching content, and using the normalized semantic contribution of the user consultation question and the normalized semantic contribution of the industrial technology query matching content to weighted modulate the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content to obtain a modulated user consultation question semantic coding vector and a modulated industrial technology query matching content semantic coding vector; Using the modulated user consultation question semantic coding vector as the query vector, the modulated industry technology query matching content semantic coding vector as the key vector and the user consultation question-industry technology query matching content semantic condition feature vector as the value vector, the modulated user consultation question semantic coding vector, the modulated industry technology query matching content semantic coding vector and the user consultation question-industry technology query matching content semantic condition feature vector are input into a feature-to-feature significant guided interaction module based on a converter structure to obtain a user consultation question-industry technology query matching content semantic significant interaction fusion representation vector as the user consultation question-industry technology query matching content semantic significant interaction fusion representation.

2. The analysis method according to claim 1 that integrates the big model and the knowledge graph technology is characterized in that: Encoding the user consultation question and the industrial technology query matching content to obtain semantic coding features of the user consultation question and semantic coding features of the industrial technology query matching content, including: semantically embedding coding the user consultation question and the industrial technology query matching content to obtain a user consultation question semantic coding vector as the user consultation question semantic coding feature and an industrial technology query matching content semantic coding vector as the industrial technology query matching content semantic coding feature.

3. The analysis method integrating large model and knowledge graph technology according to claim 2 is characterized in that: Calculating the user consultation question semantic contribution of the user consultation question semantic coding feature relative to the user consultation question-industry technology query matching content semantic condition feature vector, including: Calculate the positional division of the user consultation question semantic coding feature and the user consultation question-industry technology query matching content semantic condition feature vector to obtain the user consultation question semantic contribution vector; Calculating the logarithmic function value with base 2 of the absolute value of each eigenvalue of the user consultation question semantic contribution vector to obtain the user consultation question semantic contribution logarithmic vector; Calculate the positional dot product of the semantic coding feature of the user consultation question and the logarithmic vector of the semantic contribution of the user consultation question, and perform positional multiplication on the obtained dot product vector to obtain the semantic contribution value of the user consultation question; An exponential function with the natural constant e as the base and the semantic contribution value of the user consultation question as the exponent is calculated to obtain the semantic contribution degree of the user consultation question.

4. The analysis method integrating large model and knowledge graph technology according to claim 3 is characterized in that: The semantic contribution of the user consultation question and the semantic contribution of the industrial technology query matching content are normalized, and the normalized semantic contribution of the user consultation question and the normalized semantic contribution of the industrial technology query matching content are used to weighted modulate the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content to obtain a modulated user consultation question semantic coding vector and a modulated industrial technology query matching content semantic coding vector, including: Calculate the sum of the semantic contribution of the user consultation question and the semantic contribution of the industry technology query matching content to obtain the total sum of the semantic contribution of the user consultation question-industry technology query matching content; Respectively dividing the user consultation question semantic contribution and the industry technology query matching content semantic contribution by the total sum of the user consultation question-industry technology query matching content semantic contribution to obtain the normalized user consultation question semantic contribution and the normalized industry technology query matching content semantic contribution; Multiplying the user consultation question semantic coding feature by the normalized user consultation question semantic contribution by position point to obtain the modulated user consultation question semantic coding vector; The semantic coding feature of the industrial technology query matching content and the normalized semantic contribution of the industrial technology query matching content are multiplied by position point to obtain the modulated industrial technology query matching content semantic coding vector.

5. The analysis method integrating large model and knowledge graph technology according to claim 4 is characterized in that: The modulated user consultation question semantic coding vector is used as a query vector, the modulated industry technology query matching content semantic coding vector is used as a key vector, and the user consultation question-industry technology query matching content semantic condition feature vector is used as a value vector. The modulated user consultation question semantic coding vector, the modulated industry technology query matching content semantic coding vector, and the user consultation question-industry technology query matching content semantic condition feature vector are input into a feature-to-feature significant guided interaction module based on a converter structure to obtain a user consultation question-industry technology query matching content semantic significant interaction fusion representation vector as the user consultation question-industry technology query matching content semantic significant interaction fusion representation, including: Perform vector multiplication of the modulated user consultation question semantic coding vector and the transposed vector of the modulated industry technology query matching content semantic coding vector, and divide the obtained modulated user consultation question-industry technology query matching content semantic association matrix by the square root of the length of the modulated industry technology query matching content semantic coding vector by position to obtain a modulated user consultation question-industry technology query matching content semantic association scaling matrix; The modulated user consultation question-industry technology query matching content semantic association scaling matrix is ​​activated using a softmax function, and the obtained modulated user consultation question-industry technology query matching content semantic association scaling activation matrix is ​​multiplied by the user consultation question-industry technology query matching content semantic conditional feature vector to obtain the user consultation question-industry technology query matching content semantic significant interaction fusion representation vector.

6. The analysis method integrating large model and knowledge graph technology according to claim 5 is characterized in that: Based on the semantically significant interactive fusion representation of the user consultation question-industry technology query matching content, an intelligent answering robot based on a large language model is used to generate a question answering text, including: inputting the semantically significant interactive fusion representation vector of the user consultation question-industry technology query matching content into the intelligent answering robot based on a large language model to obtain the question answering text.

7. An analysis system integrating large models and knowledge graph technology, characterized in that: include: The module for acquiring original documents related to industrial technology is used to acquire original documents related to industrial technology from patent databases, academic literature libraries and policy documents; A text content extraction module, used to extract text content from the original documents related to the industrial technology; An industrial technology knowledge graph construction module, used to construct an industrial technology knowledge graph based on the text content, wherein the industrial technology knowledge graph includes nodes and edges between nodes; User consultation question acquisition module, used to obtain user consultation questions; An industrial technology query matching content extraction module is used to extract industrial technology query matching content from the industrial technology knowledge graph based on the user's inquiry question; A text content encoding module, used for encoding the user consultation question and the industry technology query matching content to obtain semantic encoding features of the user consultation question and semantic encoding features of the industry technology query matching content; A significant interaction analysis module between semantic features, used to perform significant interaction analysis between semantic features on the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content to obtain a fusion representation of the significant interaction between the semantic features of the user consultation question and the industrial technology query matching content; A question answering text generation module is used to generate question answering text based on the semantically significant interactive fusion representation of the user consultation question-industry technology query matching content, using an intelligent answering robot based on a large language model; Among them, the significant interaction analysis module between semantic features includes: performing joint implicit feature activation processing on the semantic coding features of the user consultation questions and the semantic coding features of the industrial technology query matching content to obtain the semantic conditional features of the user consultation questions-industrial technology query matching content; taking the semantic conditional features of the user consultation questions-industrial technology query matching content as conditional information, performing feature interaction fusion on the semantic coding features of the user consultation questions and the semantic coding features of the industrial technology query matching content to obtain the semantic significant interaction fusion representation of the user consultation questions-industrial technology query matching content; The user consultation question semantic coding features and the industry technology query matching content semantic coding features are jointly implicitly activated to obtain the user consultation question-industry technology query matching content semantic condition features, including: Inputting the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content into a joint implicit feature capture network to obtain a user consultation question-industry technology query matching content semantic joint implicit feature vector; Performing feature activation based on the Sigmoid function on the semantic joint implicit feature vector of the user consultation question-industry technology query matching content to obtain a semantic condition feature vector of the user consultation question-industry technology query matching content as the semantic condition feature of the user consultation question-industry technology query matching content; The user consultation question semantic coding features and the industry technology query matching content semantic coding features are input into a joint implicit feature capture network to obtain a user consultation question-industry technology query matching content semantic joint implicit feature vector, including: After adding the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content by position point, the obtained user consultation question-industrial technology query matching content semantic sum vector is multiplied by the weight matrix and then added by position with the bias vector to obtain the user consultation question-industrial technology query matching content semantic joint interaction vector; Using a tanh function to process the user consultation question-industry technology query matching content semantic joint interaction vector to obtain the user consultation question-industry technology query matching content semantic joint implicit feature vector; Among them, taking the semantic conditional features of the user consultation question-industry technology query matching content as conditional information, performing feature interactive fusion on the semantic coding features of the user consultation question and the semantic coding features of the industry technology query matching content to obtain the semantic significant interactive fusion representation of the user consultation question-industry technology query matching content, including: Calculate the user consultation question semantic contribution of the user consultation question semantic coding feature relative to the user consultation question-industry technology query matching content semantic condition feature vector; Calculating the semantic contribution of the semantic coding feature of the industrial technology query matching content to the semantic condition feature vector of the user consultation question-industrial technology query matching content; Normalizing the semantic contribution of the user consultation question and the semantic contribution of the industrial technology query matching content, and using the normalized semantic contribution of the user consultation question and the normalized semantic contribution of the industrial technology query matching content to weighted modulate the semantic coding features of the user consultation question and the semantic coding features of the industrial technology query matching content to obtain a modulated user consultation question semantic coding vector and a modulated industrial technology query matching content semantic coding vector; Using the modulated user consultation question semantic coding vector as the query vector, the modulated industry technology query matching content semantic coding vector as the key vector and the user consultation question-industry technology query matching content semantic condition feature vector as the value vector, the modulated user consultation question semantic coding vector, the modulated industry technology query matching content semantic coding vector and the user consultation question-industry technology query matching content semantic condition feature vector are input into a feature-to-feature significant guided interaction module based on a converter structure to obtain a user consultation question-industry technology query matching content semantic significant interaction fusion representation vector as the user consultation question-industry technology query matching content semantic significant interaction fusion representation.

Images