基于RAG的供应链智能问答模型

Supply Chain Intelligent Question and Answer Model Based on RAG

摘要: 随着人工智能技术的快速发展，问答模型已成为信息检索和知识获取的重要工具。在供应链管理相关领域，由于市场环境瞬息万变、专业术语复杂且相关流程繁多，传统的查询方式难以满足管理者和研究人员高效、准确获取信息的需求。相较于传统搜索引擎和原生开源LLMs，基于RAG的智能问答模型能够提供更高质量的答案，大幅提升知识检索的效率。因此本研究提出一种基于大模型、RAG技术的供应链智能问答模型，应用Embedding等方法，构建供应链领域相关知识文本块的向量数据库以及检索和生成模块，设计和优化Prompt提示词，提升大语言模型生成更精准和高质量的回答。结果表明，Ragas评测效果指标较好，忠实度分数，答案相关性分数，上下文相关性分数分别为0.73、0.8、0.83。模型应用测试回答准确度，专业度，可迁移性较原生LLMs具有显著优势，模型与数据集可以完全本地化，数据安全程度较高。基于RAG的供应链智能问答模型，验证了其能够充分利用供应链管理领域的大规模知识库，结合先进的自然语言处理技术和强化学习算法，实现对复杂供应链问题的深度理解和精准回答。

Abstract: With the rapid development of artificial intelligence technology, question and answer model has become an important tool for information retrieval and knowledge acquisition. In the field of supply chain management, due to the rapidly changing market environment, complex terminology and numerous related processes, traditional query methods are difficult to meet the needs of managers and researchers to obtain information efficiently and accurately. Compared with traditional search engines and native open source LLMs, RAG-based intelligent question and answer model can provide higher-quality answers and greatly improve the efficiency of knowledge retrieval. Therefore, this study proposes an intelligent question and answer model of supply chain based on large model and RAG technology. It applies the Embedding method to build a vector database of text blocks of relevant knowledge in the field of supply chain and search and generate modules, design and optimize Prompt words, and improve the large language model to generate more accurate and high-quality answers. The results show that the Ragas evaluation effect index is better, the scores of fidelity, answer relevance and context relevance are 0.73, 0.8 and 0.83, respectively. Compared with native LLMs, the model application test answer accuracy, professionalism, and portability have significant advantages. The model and data set can be fully localized, and the data security is higher. The supply chain intelligent question and answer model based on RAG proves that it can make full use of the large-scale knowledge base in the field of supply chain management, combine advanced natural language processing technology and reinforcement learning algorithm, and realize the deep understanding and accurate answer of complex supply chain questions.