文献类型：会议论文

英文题名：Research on Intelligent Relic Interpretation System Based on a Large Language Model

作者：Guo, Jingjing[1,2]; Han, Xi[1,2]

第一作者：Guo, Jingjing

机构：[1] Beijing Union University, Beijing Key Laboratory of Information Service Engineering, Beijing, China; [2] College of Robotics, Beijing Union University, Beijing, China

第一机构：北京联合大学北京市信息服务工程重点实验室

通讯机构：[1]Beijing Union University, Beijing Key Laboratory of Information Service Engineering, Beijing, China|[11417103]北京联合大学北京市信息服务工程重点实验室;[11417]北京联合大学;

会议论文集：Proceedings - 2024 IEEE 24th International Conference on Software Quality, Reliability and Security Companion, QRS-C 2024

会议日期：July 1, 2024 - July 5, 2024

会议地点：Cambridge, United kingdom

语种：英文

外文关键词：Biographies - Database systems - Economic and social effects - Linguistics

摘要：Nowadays, docents are the main force for inheriting and displaying human cultural heritage. However, the experience of visitors largely depends on their knowledge depth and updates. Large language models (LLMs) have demonstrated promising results across various fields due to their robust linguistic intelligence acquired through extensive data training. Nevertheless, training LLMs remains challenging due to their high hardware requirements. How can we integrate a LLM into a relic interpretation system without the need for extensive training? To this end, a solution is proposed by combining a LLM with a private domain database to create an intelligent relic interpretation system. This system encompasses the following functionalities:(1) Speech recognition and synthesis: Users input speech, and the system utilizes speech recognition technology to convert the speech into text. This text is then processed by the LLM, and the model's responses are synthesized into speech output; (2) Professional knowledge explanation: Leveraging a LLM and a private domain database to achieve more accurate and professional responses, compared to simply fine-tuning LLMs, this allows for the application of LLMs in vertical fields without the need for additional training; (3) Private domain database renewal: Two methods for updating the database are used: extracting relevant data from intranet sources and uploading data by professionals. The performance of the system is evaluated through comparison and ablation experiments. The results demonstrate that the system effectively addresses the limitations of traditional relic interpretation methods and introduces novel perspectives for relic interpretation. ? 2024 IEEE.