| 李哲,孙鹏劼.基于知识图谱和大模型的博物馆智能问答系统研究[J].情报工程,2025,(6):003-013 |
| 基于知识图谱和大模型的博物馆智能问答系统研究 |
| Research on the Intelligent Q&A System for Museums Based on Knowledge Graph and Large Models |
| |
| DOI： |
| 中文关键词: 博物馆；大模型；知识提取；知识图谱；问答系统 |
| 英文关键词: Museum; Large Model; Knowledge Extraction; Knowledge Graph; Question-Answering System |
| 基金项目:北京市社会科学基金项目“北京智慧型博物馆城市建设路径研究”（24LSB008）。 |
| 作者 | 单位 | | 李哲 | 北京市科学技术研究院 北京 100089 | | 孙鹏劼 | 北京自动测试技术研究所有限公司 北京 100094 |
|
| 摘要点击次数: 51 |
| 全文下载次数: 52 |
| 中文摘要: |
| [目的 /意义 ] 随着博物馆的建设规模和参观人数日益增长，对博物馆的服务形式和服务能力也提出了更高的要求。以某自然博物馆为例探索一种基于知识图谱和大模型的问答系统构建新方法，整合大模型和知识图谱各自具备的优势，提升博物馆公众服务能力。[方法/过程]在知识图谱构建过程中，为了提升该博物馆知识图谱的构建效率、降低构建成本，对 13B 大模型进行微调训练，利用微调后的模型抽取构建图谱所需的实体、关系和属性等元素。问答系统将大模型与知识图谱相结合，利用大模型识别用户意图并提取实体和关系在知识图谱中进行精准匹配，用匹配结果设计提示词交由大模型推理并返回最终结果。[ 结果 / 结论 ] 知识抽取平均准确率达到了 85% 以上，整体召回率为 88%，满足知识图谱构建和更新的工作需要。在问答系统单轮问答准确率为 91.00%，基于上下文的多轮对话准确率为 88.93%，可以满足博物馆使用需求。 |
| 英文摘要: |
| [Objective/Significance] With the increasing scale of museum construction and growing number of visitors, higher demands are being placed on museum service formats and capabilities. This paper takes a natural history museum as an example to explore a novel method for constructing a question-answering system based on knowledge graphs and large models, integrating the respective advantages of large models and knowledge graphs to enhance public service capabilities. [Methods/Processes] During the knowledge graph construction process, to improve efficiency and reduce costs, a 13B model is fine-tuned and utilized to extract required entities, relationships, and attributes for graph construction. In the process of constructing the question-answering system, the large language model is integrated with the knowledge graph.By leveraging the large model to identify user intent and extract entities/relationships for precise matching in the knowledge graph, the matched results are then used to design prompts for large model reasoning. [Results/Conclusions] The average accuracy of knowledge extraction exceeds 85% with an overall recall rate of 88%, meeting the requirements for knowledge graph construction and updates. Q&A System achieving 91.00% accuracy in single-round dialogue and 88.93% accuracy in context-based multi-round dialogues,effectively meeting museum operational needs. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
