基于关系感知的知识子图构建方法研究  

Research on the Construction Method of Knowledge Subgraph Based on Relational Perception

作　　者：骆丹 谢维信[1] 陶志强 窦昌靖 LUO Dan;XIE Weixin;TAO Zhiqiang;DOU Changjing(Shenzhen University,Guangdong,Shenzhen 518060,China;Guangdong Southern Planning&Designing Institute of TelecomCo.,Ltd.,Shenzhen,Guangdong 518000,China;Shenzhen Power Supply Bureau Co.,Ltd.,Shenzhen,Guangdong 518000,China)

机构地区：[1]深圳大学,广东深圳518060 [2]广东南方电信规划咨询设计院有限公司,广东深圳518000 [3]深圳供电局有限公司,广东深圳518000

出　　处：《移动信息》2024年第9期210-213,共4页Mobile Information

摘　　要：随着现代用户对知识需求的不断扩大,各种大规模的知识库相继涌现,复杂的知识库问答引起了研究者的广泛关注.复杂问题涉及多个实体和复杂的逻辑关系,导致答案搜索空间扩大.因此,在训练模型之前需要在数据预处理阶段构建知识子图.文中改进了传统的知识子图构建方法,提出了一种基于关系感知的知识子图构建方法,在子图检索和子图剪枝中利用关系与问题的相似性进行指导,优化子图规模的同时保证答案存在率和覆盖率.With the continuous expansion of modern users' demand for knowledge,various large-scale knowledge bases have emerged one after another,and complex knowledge base question answering has attracted extensive attention from researchers.Complex problems involve multiple entities and complex logical relationships,resulting in an expansion of the answer search space.Therefore,knowledge subgraphs need to be constructed in the data preprocessing stage before training the model.This paper improves the traditional knowledge subgraph construction method,and proposes a knowledge subgraph construction method based on relationship awareness.In subgraph retrieval and subgraph pruning,the similarity between relationships and problems is used to guide,and the size of subgraphs is optimized to ensure the existence rate and coverage of answers.

关 键 词：知识库问答 信息检索 知识子图 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]