Large model technology-based ultrahigh-speed optical module digital manufacturing scene knowledge reasoning method, device, equipment and medium

Abstract

The invention discloses an ultra-high-speed optical module digital manufacturing scene knowledge reasoning method, device and equipment based on a large model technology and a medium, and belongs to the technical field of artificial intelligence, and the method comprises the steps: constructing a knowledge graph of an ultra-high-speed optical module digital manufacturing scene; converting the multi-modal data into vector representation; a natural language input by a user is converted into a high-dimensional vector, and related knowledge points and reference data are matched; performing labeling and data enhancement on historical reasoning question and answer pairs and contextual dialogues, determining a contextual relationship by referring to a knowledge graph, and forming a fine tuning data set; performing fine tuning on the large language model by utilizing the fine tuning data set; and performing deep understanding on the matched related knowledge points and reference data by the fine-tuned large language model to generate a reasoning question and answer result. According to the method, professional knowledge and multi-modal data in the manufacturing process of the ultra-high-speed optical module are considered, the natural language processing technology and the multi-modal data understanding capacity are combined, and the understanding accuracy of the system on complex queries and the answering accuracy are improved.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and more specifically, relates to a method, device, equipment and medium for reasoning knowledge of ultra-high-speed optical module digital manufacturing scenarios based on large model technology. Background Art

[0002] In recent years, my country's optical module manufacturing industry has continued to flourish and has become one of the world's largest optical module producers. China's optical module industry has made remarkable achievements in the fields of commercial optical modules, data center optical modules, and fiber-optic communications. With the changes in the manufacturing and operating environment of optical modules, the digitalization of optical module manufacturing systems has been one of the important directions for the development of the optical module industry in recent years. In the ultra-high-speed optical module digital manufacturing scenario, knowledge reasoning based on large model technology plays a key role. This method not only adapts to the rapid development needs of my country's optical module manufacturing industry, but also greatly improves the efficiency of information query and decision-making support capabilities, helping China become a leader in the world's optical module manufacturing field.

[0003] Big model technology has powerful knowledge integration and understanding capabilities. Through large-scale data pre-training, the system can fully grasp all types of information involved in the optical module manufacturing process, from physical structure to control system, from production process to equipment parameters. The natural language processing capability of the big model enables it to accurately understand the questions and queries raised by users in natural language, reducing the dependence on specific keywords or query syntax. In addition, the big model also has semantic reasoning capabilities, which can integrate multimodal data into the answers to users and provide more comprehensive and accurate reference information.

[0004] At present, large model technology has made significant progress in multiple industries and has been widely used. However, there are still many challenges in deploying knowledge reasoning systems in ultra-high-speed optical module digital manufacturing scenarios. On the one hand, most existing knowledge reasoning systems are general and can answer a wide range of questions, but for questions involving specific fields or expertise, because these systems lack a deep understanding of specific fields and support from expertise, the accuracy and depth of the answers may be insufficient. On the other hand, in practical applications, most of these systems can only process text data, while ignoring the rich multimodal information in ultra-high-speed optical module digital manufacturing scenarios, such as images and sensor data, which limits their ability to understand and respond to complex queries. In addition, existing systems often lack a deep understanding of the context when matching user queries with knowledge base information. Summary of the invention

[0005] In response to the above defects or improvement needs of the prior art, the present invention proposes a knowledge reasoning method, device, equipment and medium for digital manufacturing scenarios of ultra-high-speed optical modules based on big model technology, which fully considers the professional knowledge and multimodal data processing requirements involved in the manufacturing process of ultra-high-speed optical modules. By introducing a domain-specific knowledge base and deeply integrating it with the big model, and combining natural language processing technology with multimodal data understanding capabilities, the system's understanding of complex queries and the accuracy of its answers are improved.

[0006] To achieve the above object, according to one aspect of the present invention, a method for reasoning knowledge of ultra-high-speed optical module digital manufacturing scenarios based on large model technology is provided, comprising:

[0007] Integrate equipment, process and time information in the digital manufacturing scenario of ultra-high-speed optical modules to build a knowledge graph;

[0008] Based on the knowledge graph, multimodal data including text, images and sensor data are converted into vector representations and stored in a multimodal vector database;

[0009] Convert the natural language input by the user into a high-dimensional vector, and match the high-dimensional vector input by the user with relevant knowledge points and reference materials in the multimodal vector database;

[0010] Label and enhance the historical reasoning question-answer pairs and contextual dialogue data, and refer to the knowledge graph to clarify the contextual relationship, thereby forming a fine-tuning dataset;

[0011] Use the fine-tuning dataset to fine-tune the large language model, handle multi-round dialogue and context dependency issues, and evaluate the model performance of the large language model;

[0012] Based on the fine-tuned large language model, we conduct in-depth understanding of the matched relevant knowledge points and reference materials to generate reasoning question and answer results.

[0013] In some optional implementation schemes, the integration includes equipment, process and time information in the ultra-high-speed optical module digital manufacturing scene, and constructs a knowledge graph, including:

[0014] Clarify the goals and domain modeling of the knowledge graph, and identify the key entities and relationships involved in the digital manufacturing scenarios of ultra-high-speed optical modules;

[0015] Identify entities in multiple data sources and extract the relationships between entities, use the graph database to generate a preliminary knowledge graph, verify and optimize the preliminary knowledge graph, and obtain the final knowledge graph.

[0016] In some optional implementations, the method of converting multimodal data including text, image and sensor data into vector representation based on the knowledge graph and storing the vector representation in a multimodal vector database includes:

[0017] Preprocess different types of data, and perform feature extraction and vectorization on the preprocessed data through pre-trained models;

[0018] Use alignment technology and attention mechanism to fuse the vectorized multimodal data, store the fused vectors in the vector database, and set the retrieval index;

[0019] The knowledge graph provides a framework for semantic annotation and relationship definition, mapping multimodal data to specific equipment and process nodes to clarify the contextual meaning of the data.

[0020] In some optional implementation schemes, converting the natural language input by the user into a high-dimensional vector includes:

[0021] Normalize the natural language input by the user, and use natural language processing technology to perform word segmentation and part-of-speech tagging on the normalized natural language;

[0022] The natural language after word segmentation and part-of-speech tagging is used as the input of the pre-trained language model to obtain a high-dimensional vector corresponding to the input natural language.

[0023] In some optional implementation schemes, matching relevant knowledge points and reference materials to the high-dimensional vector input by the user in the multimodal vector database includes:

[0024] Select a similarity measure and a retrieval algorithm to match the high-dimensional vector input by the user with the vectors in the multimodal vector database.

[0025] In some optional implementation schemes, the historical reasoning question-answer pairs and contextual dialogue data are annotated and data enhanced, and the contextual relationship is clarified by referring to the knowledge graph to form a fine-tuning dataset, including:

[0026] Collect large-scale reasoning question-answer pairs, domain-related documents, and contextual conversation data, and preprocess the collected data;

[0027] The hierarchical structure of the knowledge graph is used to annotate the contextual dependencies between the preprocessed data to clarify the relationship between questions, answers and context, and then a fine-tuning dataset is obtained, which is then expanded using text enhancement technology.

[0028] In some optional implementation schemes, fine-tuning the large language model using the fine-tuning dataset, processing multi-round dialogue and context dependency issues, and evaluating the model performance of the large language model include:

[0029] Select a pre-trained large language model as the base model, determine a fine-tuning strategy, use labeled reasoning questions and answers to adjust the base model, and handle multi-round dialogues and context dependency issues;

[0030] Determine the loss function, determine the learning rate and training rounds, optimize the text generation quality, and use the fine-tuning dataset for training.

[0031] In some optional implementation schemes, the fine-tuned large language model performs in-depth understanding of the matched relevant knowledge points and reference materials to generate reasoning question-answering results, including:

[0032] The matched relevant knowledge points and reference materials are input into the large language model for further screening and understanding, and the content that best meets the user's intention is identified;

[0033] Combining user input, contextual information, and matched relevant knowledge points and reference materials, the large language model determines the user's true intention and semantically integrates and reorganizes the matched relevant knowledge points and reference materials.

[0034] According to another aspect of the present invention, there is provided an ultra-high-speed optical module digital manufacturing scene knowledge reasoning device based on large model technology, comprising:

[0035] A knowledge graph construction module is used to integrate the equipment, process and time information in the digital manufacturing scenario of ultra-high-speed optical modules to build a knowledge graph;

[0036] A multimodal data processing module, used to convert multimodal data including text, image and sensor data into vector representation based on the knowledge graph and store it in a multimodal vector database;

[0037] A user input vectorization module, which is used to convert the natural language input by the user into a high-dimensional vector;

[0038] The vector retrieval module is used to match relevant knowledge points and reference materials to the high-dimensional vectors input by the user in the multimodal vector database;

[0039] The fine-tuning dataset production module is used to annotate and enhance historical reasoning question-answer pairs and contextual dialogue data, and to refer to the knowledge graph to clarify the contextual relationship, thereby forming a fine-tuning dataset;

[0040] Large language model fine-tuning module, which is used to fine-tune the large language model using the fine-tuning dataset, handle multi-round dialogue and context dependency issues, and evaluate the model performance of the large language model;

[0041] The knowledge reasoning output module is used to deeply understand the matched relevant knowledge points and reference materials based on the fine-tuned large language model and generate reasoning question and answer results.

[0042] According to another aspect of the present invention, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the above methods when executing the computer program.

[0043] According to another aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the steps of any one of the above methods are implemented.

[0044] In general, the above technical solutions conceived by the present invention can achieve the following beneficial effects compared with the prior art:

[0045] The present invention provides a knowledge reasoning method for ultra-high-speed optical module digital manufacturing scenarios based on large model technology. First, by integrating multimodal data, a knowledge graph in this field is constructed, and a multimodal vector database is created using vector conversion technology to achieve efficient retrieval of user queries. Then, through a fine-tuned large language model, it is possible to semantically understand and reason about complex manufacturing processes and generate accurate reasoning question and answer results. Finally, through modular design, efficient integration and optimization of knowledge reasoning functions are achieved. Aiming at the professional knowledge and multimodal data processing requirements involved in the manufacturing process of ultra-high-speed optical modules, and the problems of complex information and low retrieval efficiency in the manufacturing process of ultra-high-speed optical modules, the present invention proposes a solution based on the combination of large models and vector retrieval technology, which significantly improves the system's reasoning question and answer accuracy and response speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a schematic diagram of an overall architecture provided by an embodiment of the present invention;

[0047] Figure 2 It is a schematic diagram of a knowledge graph construction process provided by an embodiment of the present invention;

[0048] Figure 3 It is a schematic diagram of a multimodal vector database construction process provided by an embodiment of the present invention;

[0049] Figure 4 It is a schematic diagram of a user input processing and vector retrieval process provided by an embodiment of the present invention;

[0050] Figure 5 It is a schematic diagram of a process of preparing a fine-tuning dataset and fine-tuning a large language model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0052] like Figure 1 As shown, an embodiment of the present invention provides an ultra-high-speed optical module digital manufacturing scene knowledge reasoning method based on large model technology, including:

[0053] Step S1: Knowledge graph construction. The system integrates various information in the ultra-high-speed optical module digital manufacturing scenario, such as equipment, process, and time;

[0054] Step S2: Multimodal data processing. Using natural language processing and deep learning techniques, text, images, and sensor data are converted into vector representations and stored in a multimodal vector database.

[0055] Step S3: Vectorization of user input. Normalize the user's natural language input and perform word segmentation and part-of-speech tagging to convert the text into a high-dimensional vector representation;

[0056] Step S4: Vector search: Quickly match the user's vectorized input with relevant knowledge points and reference materials.

[0057] Step S5: Fine-tune the dataset. Collect large-scale reasoning question-answer pairs and contextual dialogue data for annotation and data enhancement to clarify the contextual relationship.

[0058] Step S6: Fine-tune the large language model. Use the fine-tuning dataset to fine-tune the large language model, handle multi-round dialogue and context dependency issues, and evaluate the model performance;

[0059] Step S7: Knowledge reasoning output: Based on the fine-tuned large language model, combined with reasoning algorithms and context analysis, the user's query is deeply understood and accurate reasoning question and answer results are generated.

[0060] Furthermore, if Figure 2As shown in the figure, building a knowledge graph in the field of ultra-high-speed optical module manufacturing is a systematic process involving multiple steps such as demand analysis, data collection, preprocessing, entity recognition and relationship extraction, graph construction, completion and reasoning, verification and optimization, application integration, and system deployment and maintenance. First, by clarifying the goals and domain modeling of the knowledge graph, key entities and relationships are determined. Then, multiple data sources are collected and processed, entities are identified and relationships are extracted, and a preliminary knowledge graph is generated using a graph database to complete the graph and perform semantic reasoning. Finally, the accuracy of the graph is improved through verification and optimization, and it is integrated into the knowledge reasoning system to achieve real-time updates and system maintenance.

[0061] Knowledge graphs play a key role in multimodal data processing and fine-tuning dataset production. In the multimodal data processing stage, knowledge graphs accurately map multimodal information such as sensor data and text to specific equipment and process nodes by providing a framework for semantic annotation and relationship definition. For example, temperature sensor data is associated with the "heat treatment process" node, thereby clarifying the contextual meaning of the data. In the fine-tuning dataset production stage, knowledge graphs provide structured references for the generation and annotation of question-answer pairs and contextual dialogue data, generate large-scale question-answer data using entity relationships in the graph, and annotate contextual dependencies between data through its hierarchical structure to ensure the high relevance of the dataset to the scenario.

[0062] Furthermore, if Figure 3 As shown in the figure, converting multimodal data (text, image, sensor data) into vector representation and storing it in a multimodal vector database is an important step in the knowledge reasoning system. First, different types of data are preprocessed, such as text segmentation, image normalization, and sensor data filtering. Then, the data is feature extracted and vectorized through a pre-trained model, such as using BERT to extract text vectors and CNN to extract image features. Then, the multimodal data is fused using alignment technology and attention mechanism, the processed vectors are stored in the vector database, and fast query is achieved through efficient retrieval indexing. Finally, the system is integrated and optimized, deployed to the production environment, and the model and database are updated regularly to ensure the continuous performance and timeliness of the system.

[0063] User input processing includes normalizing the user's natural language input, removing noise and redundant information, and correcting spelling errors to ensure semantic accuracy; using natural language processing technology to segment and tag the input, and performing language identification when necessary to support multi-language processing; using pre-trained language models to convert input text, images and other multimodal information into high-dimensional vector representations for similarity calculations in vector space.

[0064] Furthermore, if Figure 4As shown in the figure, the key to vector retrieval is to clarify the retrieval target, select appropriate similarity measurement methods (such as cosine similarity, Euclidean distance) and retrieval algorithms (such as FAISS, HNSW) to efficiently match user input vectors with vectors in the database. To improve retrieval efficiency, it is necessary to build an optimized index structure, such as KD tree or LSH, and perform block and quantization processing on large-scale data. This method involves multimodal data, and uses a multimodal alignment model to convert data of different modalities into vector representations of the same space to achieve cross-modal retrieval.

[0065] Furthermore, if Figure 5 As shown in the figure, the process of making the fine-tuning dataset includes data collection, data cleaning, data labeling, and data enhancement. First, collect large-scale reasoning question-answer pairs, domain-related documents, and contextual dialogue data to ensure that various questions that users may ask are covered. Data sources include existing FAQs, technical documents, customer service, and equipment maintenance records. Then, clean the data, remove duplicate, noisy, and irrelevant data, and standardize the text, such as word segmentation, punctuation processing, and word form restoration. Next, annotate the data to clarify the relationship between questions, answers, and context, especially data involving reasoning and multi-round dialogues. Finally, use text enhancement technology to expand the dataset to increase the diversity and robustness of model training.

[0066] The large language model fine-tuning process includes the following steps: First, select a pre-trained large language model (such as ChatGLM2-6B) as the base model. Then, determine the fine-tuning strategy (such as supervised fine-tuning, contextual fine-tuning, and hybrid fine-tuning), and use the labeled reasoning question and answer dataset to adjust the model to ensure that the answers it generates meet the needs of specific fields and handle multi-round dialogues and context dependency issues. Next, determine the loss function, determine the appropriate learning rate, training rounds and other hyperparameters, optimize the quality of text generation, and combine multi-task learning for joint training when necessary. Finally, evaluate the model performance regularly to prevent overfitting.

[0067] In particular, before the knowledge reasoning results are output, the big model is combined with vector retrieval technology. First, the user query is vectorized and the relevant content is quickly screened using vector retrieval. Next, the candidate content obtained by the vector retrieval is input into the big language model, and the big model further screens and understands these candidate contents to identify the content that best meets the user's intention. Finally, combining user input, contextual information, and vector retrieval candidate content, the big model can determine the user's true intention and semantically integrate and reorganize the candidate content. This combination can make up for some of the shortcomings when used alone, both improving the retrieval speed and enhancing the accuracy and depth of the answer.

[0068] The embodiment of the present invention further provides an ultra-high-speed optical module digital manufacturing scene knowledge reasoning device based on large model technology, comprising:

[0069] A knowledge graph construction module is used to integrate the equipment, process and time information in the digital manufacturing scenario of ultra-high-speed optical modules to build a knowledge graph;

[0070] A multimodal data processing module, used to convert multimodal data including text, image and sensor data into vector representation based on the knowledge graph and store it in a multimodal vector database;

[0071] A user input vectorization module, which is used to convert the natural language input by the user into a high-dimensional vector;

[0072] The vector retrieval module is used to match relevant knowledge points and reference materials to the high-dimensional vectors input by the user in the multimodal vector database;

[0073] The fine-tuning dataset production module is used to annotate and enhance historical reasoning question-answer pairs and contextual dialogue data, and to refer to the knowledge graph to clarify the contextual relationship, thereby forming a fine-tuning dataset;

[0074] Large language model fine-tuning module, which is used to fine-tune the large language model using the fine-tuning dataset, handle multi-round dialogue and context dependency issues, and evaluate the model performance of the large language model;

[0075] The knowledge reasoning output module is used to deeply understand the matched relevant knowledge points and reference materials based on the fine-tuned large language model and generate reasoning question and answer results.

[0076] The specific implementation of each module can refer to the description of the above method embodiment, and the embodiment of the present invention will not be repeated.

[0077] An embodiment of the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any of the above methods when executing the computer program.

[0078] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any of the above methods are implemented.

[0079] It should be pointed out that, according to the needs of implementation, the various steps / components described in this application can be split into more steps / components, and two or more steps / components or partial operations of steps / components can be combined into new steps / components to achieve the purpose of the present invention.

[0080] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for reasoning knowledge of ultra-high-speed optical module digital manufacturing scenarios based on large model technology, characterized in that: include: Integrate equipment, process and time information in the digital manufacturing scenario of ultra-high-speed optical modules to build a knowledge graph; Based on the knowledge graph, multimodal data including text, images and sensor data are converted into vector representations and stored in a multimodal vector database; Convert the natural language input by the user into a high-dimensional vector, and match the high-dimensional vector input by the user with relevant knowledge points and reference materials in the multimodal vector database; Label and enhance the historical reasoning question-answer pairs and contextual dialogue data, and refer to the knowledge graph to clarify the contextual relationship, thereby forming a fine-tuning dataset; Use the fine-tuning dataset to fine-tune the large language model, handle multi-round dialogue and context dependency issues, and evaluate the model performance of the large language model; Based on the fine-tuned large language model, we conduct in-depth understanding of the matched relevant knowledge points and reference materials to generate reasoning question and answer results.

2. The method according to claim 1, characterized in that The integration includes the equipment, process and time information in the ultra-high-speed optical module digital manufacturing scenario, and constructs a knowledge graph, including: Clarify the goals and domain modeling of the knowledge graph, and identify the key entities and relationships involved in the digital manufacturing scenarios of ultra-high-speed optical modules; Identify entities in multiple data sources and extract the relationships between entities, use the graph database to generate a preliminary knowledge graph, verify and optimize the preliminary knowledge graph, and obtain the final knowledge graph.

3. The method according to claim 2, characterized in that The method of converting multimodal data including text, image and sensor data into vector representation based on the knowledge graph and storing the vector representation in a multimodal vector database includes: Preprocess different types of data, and perform feature extraction and vectorization on the preprocessed data through pre-trained models; Use alignment technology and attention mechanism to fuse the vectorized multimodal data, store the fused vectors in the vector database, and set the retrieval index; The knowledge graph provides a framework for semantic annotation and relationship definition, mapping multimodal data to specific equipment and process nodes to clarify the contextual meaning of the data.

4. The method according to claim 3, characterized in that The step of converting the natural language input by the user into a high-dimensional vector includes: Normalize the natural language input by the user, and use natural language processing technology to perform word segmentation and part-of-speech tagging on the normalized natural language; The natural language after word segmentation and part-of-speech tagging is used as the input of the pre-trained language model to obtain a high-dimensional vector corresponding to the input natural language.

5. The method according to claim 4, characterized in that In the multimodal vector database, matching relevant knowledge points and reference materials to the high-dimensional vector input by the user includes: Select a similarity measure and a retrieval algorithm to match the high-dimensional vector input by the user with the vectors in the multimodal vector database.

6. The method according to claim 5, characterized in that The historical reasoning question-answer pairs and contextual dialogue data are annotated and enhanced, and the contextual relationship is clarified by referring to the knowledge graph to form a fine-tuning dataset, including: Collect large-scale reasoning question-answer pairs, domain-related documents, and contextual conversation data, and preprocess the collected data; The hierarchical structure of the knowledge graph is used to annotate the contextual dependencies between the preprocessed data to clarify the relationship between questions, answers and context, and then a fine-tuning dataset is obtained, which is then expanded using text enhancement technology.

7. The method according to claim 6, characterized in that The fine-tuning dataset is used to fine-tune the large language model, handle multi-round dialogue and context dependency issues, and evaluate the model performance of the large language model, including: Select a pre-trained large language model as the base model, determine a fine-tuning strategy, use labeled reasoning questions and answers to adjust the base model, and handle multi-round dialogues and context dependency issues; Determine the loss function, determine the learning rate and training rounds, optimize the text generation quality, and use the fine-tuning dataset for training.

8. The method according to claim 7, characterized in that The fine-tuned large language model performs in-depth understanding of the matched relevant knowledge points and reference materials to generate reasoning question-answering results, including: The matched relevant knowledge points and reference materials are input into the large language model for further screening and understanding, and the content that best meets the user's intention is identified; Combining user input, contextual information, and matched relevant knowledge points and reference materials, the large language model determines the user's true intention and semantically integrates and reorganizes the matched relevant knowledge points and reference materials.

9. An ultra-high-speed optical module digital manufacturing scene knowledge reasoning device based on large model technology, characterized in that: include: A knowledge graph construction module is used to integrate the equipment, process and time information in the digital manufacturing scenario of ultra-high-speed optical modules to build a knowledge graph; A multimodal data processing module, used to convert multimodal data including text, image and sensor data into vector representation based on the knowledge graph and store it in a multimodal vector database; A user input vectorization module, which is used to convert the natural language input by the user into a high-dimensional vector; The vector retrieval module is used to match relevant knowledge points and reference materials to the high-dimensional vectors input by the user in the multimodal vector database; The fine-tuning dataset production module is used to annotate and enhance historical reasoning question-answer pairs and contextual dialogue data, and to refer to the knowledge graph to clarify the contextual relationship, thereby forming a fine-tuning dataset; Large language model fine-tuning module, which is used to fine-tune the large language model using the fine-tuning dataset, handle multi-round dialogue and context dependency issues, and evaluate the model performance of the large language model; The knowledge reasoning output module is used to deeply understand the matched relevant knowledge points and reference materials based on the fine-tuned large language model and generate reasoning question and answer results.

10. A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method according to any one of claims 1 to 8.

Images