文章摘要
李琳娜1, 2 丁 楷 3 韩红旗 1, 2 王 力 1, 2 李艾丹 1.基于知识图谱的中文科技文献问答系统构建研究[J].中国科技资源导刊,2024,(4):51~62
基于知识图谱的中文科技文献问答系统构建研究
Research on the Construction of Question Answering System of Chinese Scientific and Technical Literature Based on Knowledge Graph
投稿时间:2023-02-23  
DOI:
中文关键词: 中文科技文献问答系统;知识图谱;问题分类体系;集成学习
英文关键词: Q&A system of Chinese scientific and technical literature, knowledge graph, question categorical method, ensemble learning
基金项目:中国科学技术信息研究所重点工作项目“智能情报融合创新体系建设研究与应用”(ZD2023-11);国家重点研发计划 项目“颠覆性技术识别理论、方法与专家预判系统”(2019YFA0707201)
作者单位
李琳娜1, 2 丁 楷 3 韩红旗 1, 2 王 力 1, 2 李艾丹 1 (1. 中国科学技术信息研究所,北京 100038
2. 富媒体数字出版内容组织与知识服务重点实验室,北京 100038
3. 中国航天科工集团六院情报信息研究中心,内蒙古呼和浩特 010000) 
摘要点击次数: 35
全文下载次数: 92
中文摘要:
      科技文献问答系统能以自然语言对话的方式为用户提供高水平的知识服务。针对语义解析型知识图谱问答系统存在跨领域适应性弱及现有基于深度学习、大模型的问答系统存在结果可解释性差且难以溯源的问题,提出基于句式特点的中文问题分类方法,并设计基于Pipeline方法的中文科技文献问答系统框架。实验结果表明,基于句式特点的问题分类具有不依赖于特定领域的特点且在效果上与基于意图的问题分类基本相当,基于Pipeline的问题解析方法能有效地将问题转化为知识图谱查询语句,从而满足用户对自动问答结果可解释、可溯源的基本需求。
英文摘要:
      The Q&A system of scientific and technical literature can provide high-level knowledge services for researchers with natural language. But the current semantic parsing-based knowledge graph Q&A system has poor cross-domain adaptability and Q&A systems based on deep learning or large language model suffer from poor interpretability and traceability of results. Aiming to address these issues, this article proposed a Chinese question categorical method based on sentence patterns and designed a Pipeline based framework for the Q&A system of Chinese scientific and technical literature. The experimental results show that question classification based on sentence patterns does not rely on specific domains and its effectiveness is basically comparable to the question classification based on intentions. The Pipeline-based question parsing method can effectively transform questions into knowledge graph query statements and effectively meets users’ need for Q&A answers with interpretability and traceability of results.
查看全文   查看/发表评论  下载PDF阅读器
关闭

分享按钮