基于ChatGLM和提示微调的旅游知识图谱构建  

Tourism Knowledge Graph Construction Based on ChatGLM and Prompt-tuning

作　　者：徐春[1] 苏明钰 孙彬[1] XU Chun;SU Ming-yu;SUN Bin(School of Information Management,Xinjiang University of Finance and Economics,Urumqi 830012,China)

机构地区：[1]新疆财经大学信息管理学院,乌鲁木齐830012

出　　处：《科学技术与工程》2024年第31期13484-13492,共9页Science Technology and Engineering

基　　金：国家自然科学基金(62266041);新疆自然科学基金(2023D01A73)。

摘　　要：为缓解旅游领域知识分散、信息碎片化的问题,提出一种基于ChatGLM(chat generative language model)和提示微调的实体关系抽取模型ChatGLM-ppt(ChatGLM with prompt and p-tuning)。该模型借助ChatGLM以对话形式完成实体关系抽取任务,并通过P-Tuning v2微调和添加提示模板的方法应对实体关系抽取中错误传播、实体冗余和关系重叠等问题。实验建立在自建的旅游领域数据集上,结果表明:在旅游领域实体关系抽取问题上ChatGLM-ppt模型F 1为92.19%,在处理重叠关系问题中F 1均大于90%,优于目前主流的实体关系抽取模型,证明该模型可有效提高实体关系抽取的准确率。进一步运用Neo4j图数据库构建旅游知识图谱,整合分散的旅游信息资源,对促进旅游业的数字化转型和智能化发展具有一定的参考意义。In order to alleviate the problem of knowledge dispersion and information fragmentation in the tourism domain,an entity relation extraction model ChatGLM-ppt(ChatGLM with prompt and p-tuning)based on ChatGLM(chat generative language model)and prompt fine-tuning was proposed.The model accomplished the entity and relation extraction task in the form of dialog with the help of ChatGLM,and coped with the problems of error propagation,entity redundancy and relationship overlapping in entity and relation extraction by means of P-Tuning v2 fine-tuning and adding prompt templates.The experiments were built on the self-constructed tourism domain dataset,and the results show that the F 1 of the ChatGLM-ppt model is 92.19%in the entity and relation extraction problem in the tourism domain,and the F 1 are all greater than 90%in dealing with overlapping relationships,which is better than that of the current mainstream entity and relation extraction models,proving that the model can effectively improve the accuracy of entity and relation extraction.Further use of Neo4j graph database to construct tourism knowledge graph and integrate dispersed tourism information resources has certain reference significance for promoting the digital transformation and intelligent development of tourism.

关 键 词：实体关系抽取 关系重叠 大语言模型 知识图谱 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]