ElderQA-GPT: A Large Language Model for Online Q&A on Geriatric Diseases Based on BGE Semantic Vector Knowledge Base and LangChain Architecture

Scientific guidance of disease Q&A for the elderly is of distinct significance for improving the health level of the elderly, which is the immediate need of the current aging society. The traditional question answering system for geriatric diseases based on knowledge graph relies heavily on manual annotation and has some problems such as difficulty in knowledge fusion. The rapid development of large models provides a new opportunity for geriatric disease Q&A system, but it also has problems such as interpretability, applicability and multi-round Q&A illusion. In this paper, we propose a large language model ElderQA-GPT for online Q&A of geriatric diseases based on BGE semantic vector knowledge base and LangChain architecture. First, in order to improve the interpretability of Q&A dialogues, this paper adopts the SELF-QA framework and combines the current professional websites of geriatric diseases (including Dr. Ding Xiang, Seeking Medical Care, and the Chinese government website) to construct a vector database of geriatric diseases based on the BGE semantic vector model; secondly, in order to improve the accuracy and applicability of the system, the lightweight open source ChatGLM3 model is privately deployed based on the LangChain architecture, which enhances the retrieval capability of the localized inference knowledge base; lastly, in order to better accumulate and comprehend the information in the complex contexts and multiple rounds of dialogues, and to maintain the semantic consistency, a LoRA fine-tuning technique is used to enhance the ElderQA-GPT multi-round dialogue capability. Two sets of quantitative experiments were conducted to explore two dimensions: the interpretability of model responses and the resolution of the illusion problem in multi-round Q&A sessions. Additionally, an ablation study was carried out by omitting the knowledge base to comparatively evaluate the impact on the system's performance. The validation results demonstrate that ElderQA-GPT exhibits improved answer interpretability and applicability, effectively maintaining semantic consistency across multi-round dialogues without introducing false information. The findings of this study significantly advance the field of online Q&A for geriatric diseases by addressing critical challenges such as interpretability, accuracy, and multi-round dialogue. Our proposed model ElderQA-GPT not only improves public knowledge and understanding of geriatric diseases but also optimizes medical resource allocation and enhances health management for the elderly population. By providing interpretable and accurate responses across multi-round dialogues, ElderQA-GPT ensures that users receive reliable medical information tailored to their needs. These advancements have profound implications for healthcare delivery and society, paving the way for more efficient and effective healthcare services for elderly individuals.