Atinide alloy knowledge question-answering system based on large language model and storage medium

Abstract

The invention discloses an actinide alloy knowledge question-answering system based on a large language model and a storage medium, and belongs to the technical field of actinide alloy knowledge question-answering systems and artificial intelligence. The question-answering system comprises a knowledge base construction module, a user input enhancement module, a retrieval knowledge arrangement module and a reply module; the knowledge base construction module is used for constructing a knowledge base, collecting actinide alloy knowledge through network retrieval and carrying out structured processing on the actinide alloy knowledge; the user input enhancement module is used for performing semantic optimization on a user input problem; the retrieval knowledge arrangement module is used for finely sorting the retrieval and collection results; and the reply module is used for understanding the query intention of the user, recalling related professional contents from the knowledge base and generating answers. According to the system, non-professionals can quickly obtain required actinide alloy knowledge, interdisciplinary communication and cooperation are promoted, and popularization and innovative development of the actinide alloy technology are promoted.

Description

Technical Field

[0001] The present invention belongs to the field of actinide alloy knowledge question-answering system and artificial intelligence technology, and in particular relates to an actinide alloy knowledge question-answering system based on a large language model and a storage medium. Background Art

[0002] The actinide alloy industry, due to its unique characteristics, presents high barriers to entry. Knowledge of actinide alloys involves complex physical and chemical processes, equipment operating procedures, and production management expertise. The accumulation and transfer of this knowledge often requires long-term production experience. For outsiders to gain a deep understanding of actinide alloys and related knowledge, it often requires a significant investment of time and effort. This not only limits interdisciplinary communication and collaboration, but also hinders the rapid application and diffusion of new technologies in the actinide alloy field. Furthermore, in their actual operations, actinide alloy workers often need to consult extensive professional books and literature to obtain relevant knowledge. This is not only time-consuming and labor-intensive, but can also hinder production efficiency due to untimely or inaccurate information. Rapid access to accurate professional knowledge is particularly important when encountering unexpected production issues. Therefore, establishing a professional knowledge question-and-answer system for the actinide alloy industry can not only effectively reduce industry barriers and promote interdisciplinary communication, but also provide timely and accurate professional knowledge support to actinide alloy workers, thereby significantly improving the production efficiency of actinide alloy production lines. Summary of the Invention

[0003] In view of this, the present invention discloses an actinide alloy knowledge question-answering system and a storage medium based on a large language model.

[0004] The present invention adopts the following technical solutions:

[0005] A large language model-based actinide alloy knowledge question-answering system, comprising a knowledge base construction module, a user input enhancement module, a search knowledge organization module, and a reply module;

[0006] The knowledge base construction module is used to construct a knowledge base, collect actinide alloy knowledge through network retrieval, and structure the actinide alloy knowledge; the user input enhancement module is used to semantically optimize user input questions; the retrieval knowledge collation module is used to fine-tune the results of the retrieval and collection; the response module is used to understand the user's query intention, recall relevant professional content from the knowledge base, and generate answers.

[0007] Furthermore, the knowledge base construction module includes: obtaining literature related to actinide alloys from academic databases, extracting the text content of the literature, and then using GPT-4 API to clean and structure the literature.

[0008] Furthermore, the knowledge base construction module also includes: converting continuous text into question-answer pairs based on the GPT-4 API to construct a supervised fine-tuning dataset.

[0009] Furthermore, the semantic optimization includes: based on the DeepSeek-LLM-7B-Chat model, using LoRA fine-tuning technology to perform domain adaptation on the user input questions.

[0010] Furthermore, the semantic optimization also includes: using the DeepSeek-LLM-7B-Chat model fine-tuned by LoRA to allow the user input questions to generate richer and more accurate query expressions.

[0011] Furthermore, the retrieval knowledge sorting module includes: performing knowledge retrieval through the M3-Embedding model, and using the Cross-Encoder structure to finely sort the results of the retrieval collection.

[0012] Furthermore, the retrieval knowledge organization module also includes: converting the user input question into a vector representation through the M3-Embedding model, and performing similarity calculation with the text vector in the knowledge base, recalling candidate documents based on the results of the similarity calculation, and then refining the candidate documents to screen out a subset of documents that match the user's query intention.

[0013] Furthermore, the reply module includes an improved RAG framework, which is obtained by adding dynamic query expansion and multi-source knowledge structured reorganization functions on the basis of the standard RAG process.

[0014] Furthermore, the reply module further includes: inputting the retrieved documents into the improved RAG framework, and structurally reorganizing the retrieved knowledge fragments in combination with the semantic optimization provided by the user input enhancement module.

[0015] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned actinide alloy knowledge question-answering system based on a large language model.

[0016] Beneficial effects of the present invention:

[0017] Significantly improve knowledge retrieval and question-answering efficiency: Through the user input enhancement module and the retrieval knowledge organization module, the present invention enables the system to quickly and accurately retrieve the knowledge content most relevant to the user's question from a large-scale knowledge base and generate high-quality answers, greatly shortening the time for actinide alloy workers to acquire professional knowledge and improving production efficiency.

[0018] Improving the accuracy and professionalism of question-answering: The improved RAG framework combines user input enhancement with knowledge integration mechanisms, enabling a more accurate understanding of complex process issues and generating answers that include precise technical details and in-depth analysis. This effectively avoids the knowledge fragmentation and inaccurate answers common in traditional question-answering systems, providing reliable decision support for actinide alloy production.

[0019] Lowering the threshold for acquiring industry knowledge: The present invention constructs an actinide alloy knowledge question-and-answer system based on a large language model, which breaks the knowledge barrier of the actinide alloy industry, allowing non-professionals to quickly acquire the required actinide alloy knowledge, promoting interdisciplinary exchanges and cooperation, and promoting the popularization and innovative development of actinide alloy technology.

[0020] Good scalability and application prospects: The technical solution of the present invention is not only applicable to the field of actinide alloys, but can also be expanded to upstream and downstream links such as steelmaking and rolling, building an intelligent knowledge service system covering the entire steel production process, and providing a new technical path and reference example for the construction of intelligent knowledge question-and-answer systems in other industrial fields. It has broad application prospects and important social and economic benefits. DETAILED DESCRIPTION

[0021] In order to better understand the technical solution of the present invention, the following is a detailed description in conjunction with the embodiments of the present invention.

[0022] It should be understood that the embodiments described are only a portion of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative work are within the scope of protection of the present invention.

[0023] Example 1

[0024] A large language model-based actinide alloy knowledge question-answering system, comprising a knowledge base construction module, a user input enhancement module, a search knowledge organization module, and a reply module;

[0025] The knowledge base construction module is used to construct a knowledge base, collect actinide alloy knowledge through network retrieval, and structure the actinide alloy knowledge; the user input enhancement module is used to semantically optimize user input questions; the retrieval knowledge collation module is used to fine-tune the results of the retrieval and collection; the response module is used to understand the user's query intention, recall relevant professional content from the knowledge base, and generate answers.

[0026] Furthermore, the knowledge base construction module includes: obtaining literature related to actinide alloys from academic databases, extracting the text content of the literature, and then using the GPT-4 API to clean and structure the literature to form a knowledge base containing multiple texts.

[0027] Furthermore, the knowledge base construction module also includes: converting continuous text into question-answer pairs based on the GPT-4 API to construct a supervised fine-tuning dataset, laying the foundation for subsequent model training.

[0028] Furthermore, the semantic optimization includes: based on the DeepSeek-LLM-7B-Chat model, using LoRA fine-tuning technology to adapt the input questions to the domain. Through multiple rounds of semantic optimization of user input questions, the search recall rate is improved, and the system's ability to understand complex process problems is enhanced.

[0029] Furthermore, the semantic optimization also includes: using the DeepSeek-LLM-7B-Chat model fine-tuned by LoRA to generate richer and more accurate query expressions when users input questions, thereby enhancing the matching between questions and relevant content in the knowledge base.

[0030] Furthermore, the retrieval knowledge sorting module includes: performing knowledge retrieval through the M3-Embedding model and using the Cross-Encoder structure to perform refined sorting on the retrieval results, thereby improving the accuracy and relevance of knowledge retrieval.

[0031] Furthermore, the retrieval knowledge organization module also includes: converting the user input question into a vector representation through the M3-Embedding model, and performing similarity calculation with the text vector in the knowledge base, recalling candidate documents based on the results of the similarity calculation, and then refining the candidate documents to screen out a subset of documents that match the user's query intention.

[0032] Furthermore, the response module includes an improved RAG framework, which is achieved by adding dynamic query expansion and multi-source knowledge structured reorganization capabilities to the standard RAG process. This enables the system to more accurately understand the user's query intent, recall the most relevant professional content from the knowledge base, and generate high-quality answers.

[0033] Furthermore, the answer module also includes: inputting the retrieved documents into the improved RAG framework, combining the semantic optimization provided by the user input enhancement module, and structurally reorganizing the retrieved knowledge fragments to generate accurate, complete, and professional answers, which are ultimately presented to the user.

[0034] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned actinide alloy knowledge question-answering system based on a large language model.

[0035] Example 2

[0036] This paper proposes an actinide alloy knowledge question-answering system based on a large language model. Its technical solution mainly includes the following core components:

[0037] Knowledge Base Construction Module: Automated data collection tools were used to obtain literature related to actinide alloys from authoritative academic resources such as China National Knowledge Infrastructure (CNKI). Optical Character Recognition (OCR) technology combined with the GPT-4 API was used to clean and structure the literature, resulting in a high-quality knowledge base containing 9,292 texts. Furthermore, 15,648 question-answer pairs were generated using the GPT-4 API, providing rich data support for domain model training.

[0038] Data collection and preprocessing involved using automated crawler technology to retrieve literature related to actinide alloys from academic databases. Optical Character Recognition (OCR) was then used to extract text from these documents. This was then processed through the GPT-4 API's intelligent cleaning process to remove redundant and useless information, resulting in high-quality continuous text data. Furthermore, leveraging the GPT-4 API's contextual learning capabilities, the continuous text was converted into question-and-answer pairs to construct a supervised fine-tuning dataset, laying the foundation for subsequent model training.

[0039] User input enhancement module: A user input enhancement module was designed based on the DeepSeek-LLM-7B-Chat model. LoRA fine-tuning technology was used for domain adaptation. Through multiple rounds of semantic optimization of user input questions, the retrieval recall rate was significantly improved to 69%, enhancing the system's ability to understand complex process problems.

[0040] User input processing: When a user asks a question to the system, the user input enhancement module first performs semantic parsing and expansion of the question. Using the DeepSeek-LLM-7B-Chat model fine-tuned by LoRA, it generates richer and more accurate query expressions for the user input question, enhancing the match between the question and relevant content in the external knowledge base.

[0041] Retrieval knowledge organization module: Comparing the two vectorization models BGE-large-Embedding and M3-Embedding, the optimized M3-Embedding model is selected for knowledge retrieval, and the BGE-Reranker reranking model is introduced. The Cross-Encoder structure is used to fine-tune the sorting of the search results to improve the accuracy and relevance of knowledge retrieval.

[0042] Knowledge retrieval converts the enhanced user input question into a vector representation using the M3-Embedding model. Similarity is calculated with the text vectors in the knowledge base to quickly recall the top-K candidate documents. The BGE-Reranker reranking model then refines the candidate documents, selecting the subset of documents that best matches the user's query intent.

[0043] Reply module: This module proposes an improved RAG framework that integrates user input enhancement and knowledge integration mechanisms. It adds dynamic query expansion and multi-source knowledge structured reorganization functions on the basis of the standard RAG process, enabling the system to understand user query intentions more accurately, recall the most relevant professional content from the knowledge base, and generate high-quality answers.

[0044] Knowledge integration and answer generation: The retrieved relevant documents are input into the improved RAG framework. Combined with the rich semantic information provided by the user input enhancement module, the retrieved knowledge fragments are deeply processed and restructured through the powerful understanding and information integration capabilities of the large language model to generate accurate, complete and professional answers, which are finally presented to the user.

[0045] The embodiments of the present invention are described in detail above. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

Claims

1. An actinide alloy knowledge question-answering system based on a large language model, characterized by: The question-answering system includes a knowledge base construction module, a user input enhancement module, a search knowledge arrangement module and a reply module; The knowledge base construction module is used to construct a knowledge base, collect actinide alloy knowledge through network retrieval, and perform structured processing on the actinide alloy knowledge; The user input enhancement module is used to perform semantic optimization on user input questions; the retrieval knowledge collation module is used to perform refined sorting on the results of the retrieval collection; the response module is used to understand the user's query intention, and recall relevant professional content from the knowledge base to generate answers.

2. The question-answering system according to claim 1, characterized in that: The knowledge base construction module includes: obtaining literature related to actinide alloys from academic databases, extracting the text content of the literature, and then using the GPT-4 API to clean and structure the literature.

3. The question-answering system according to claim 2, characterized in that: The knowledge base construction module also includes: converting continuous text into question-answer pairs based on the GPT-4 API to construct a supervised fine-tuning dataset.

4. The question-answering system according to claim 3, characterized in that: The semantic optimization includes: based on the DeepSeek-LLM-7B-Chat model, using LoRA fine-tuning technology to perform domain adaptation on the user input questions.

5. The question-answering system according to claim 4, characterized in that: The semantic optimization also includes: using the DeepSeek-LLM-7B-Chat model fine-tuned by LoRA to allow the user input questions to generate richer and more accurate query expressions.

6. The question-answering system according to claim 5, characterized in that: The retrieval knowledge sorting module includes: performing knowledge retrieval through the M3-Embedding model, and using the Cross-Encoder structure to finely sort the results of the retrieval collection.

7. The question-answering system according to claim 6, characterized in that: The retrieval knowledge organization module also includes: converting the user input question into a vector representation through the M3-Embedding model, and performing similarity calculation with the text vector in the knowledge base, recalling candidate documents based on the results of the similarity calculation, and then fine-tuning the candidate documents to screen out a subset of documents that match the user's query intention.

8. The question-answering system according to claim 7, characterized in that: The reply module includes an improved RAG framework, which is obtained by adding dynamic query expansion and multi-source knowledge structured reorganization functions on the basis of a standard RAG process.

9. The question-answering system according to claim 8, characterized in that: The reply module further includes: inputting the retrieved documents into the improved RAG framework, and structurally reorganizing the retrieved knowledge fragments in combination with the semantic optimization provided by the user input enhancement module.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the actinide alloy knowledge question-answering system based on a large language model as described in any one of claims 1 to 9 is implemented.