Research on Construction and Optimization of DeepSeek E-Commerce Model Based on RAG Local E-Commerce Knowledge Base

本文面向电商业客户对话及文案写作需求，提出基于RAG增强型检索技术的电商本地知识库与轻量化本地部署大语言模型结合，建立一种电商大语言模型智能助理。采用了Nomic-embed模型和数据处理方法，建立本地电商知识库的向量数据，通过设置采样和检索参数，结合DeepSeek-R1大语言模型实现对话与文案内容的推理。通过Dify框架和Ollama参数的试验，提出了基于大语言模型参数的模型部署调优方法。新搭建的本地电商大语言模型助理可有效解决在线大语言模型工具响应速度慢、信息传输不安全及推理内容不特定、不精确等问题，可为电商业提升工作效率提供新的途径。This paper addresses the needs of e-commerce businesses in customer dialogue and copywriting by proposing an intelligent e-commerce assistant that integrates a RAG-enhanced retrieval technology-based local knowledge base with a lightweight, locally deployed large language model. The Nomic-embed model and data processing methods are employed to construct vector data for the local e-commerce knowledge base. By configuring sampling and retrieval parameters and combining them with the DeepSeek-R1 large language model, the system enables reasoning for dialogue and content generation. Through experiments with the Dify framework and Ollama parameters, a method for optimizing model deployment based on large language model parameters is proposed. The newly developed local e-commerce large language model assistant effectively addresses issues such as slow response times, insecure data transmission, and non-specific or imprecise reasoning content commonly found in online large language model tools. This approach offers a new pathway for improving operational efficiency in the e-commerce industry.