| 摘 要: 火灾是常见的一种灾害,火灾应急知识的获取对个人安全和社会发展至关重要。文章提出一种面向火灾应急领域的知识图谱问答Pipeline(流水线模型)改进方法。首先对实体识别、实体链接及路径排序阶段的模型采用Task specific的思想进行独立训练,结合特征融合算法对知识三元组进行重新排序;其次在路径排序模型中引入Beam Search(集束搜索)算法;最后采用迁移学习策略,在通识领域的知识图谱问答语料场景下训练大参数量模型。实验证明,该方法应用于火灾应急领域语料上的准确率为89.0%,优于传统Pipeline方法,问答效果更好。 |
| 关键词: 火灾应急;知识图谱问答;信息检索;语义匹配 |
|
中图分类号: TP391.1
文献标识码: A
|
|
| Research on Q&A Method of Knowledge Graph in Fire Emergency Domain |
|
PAN Ru1, ZHA Jun1,2
|
(1.School of Electronic and Information Engineering, Anhui Jianzhu University, Heifei 230601, China;
2.Hefei Institute of Public Safety Research, Tsinghua University, Hefei 230601, China)
768314326@qq.com; zhajun@ahlzkj.com
|
| Abstract: Fire is one of the most common disasters, and the acquisition of fire emergency knowledge is crucial for personal safety and social development. This paper proposes a method of improving knowledge graph Q&A (Questioning-answering) Pipeline ( a pipeline model) for fire emergency domain. Firstly, the models for entity recognition, entity linking and path ranking are trained independently using the idea of Task specific, and the knowledge triplets is re-ranked by combining feature fusion algorithm. Secondly, Beam Search algorithm is introduced in the path ranking model. Finally, transfer learning strategy is adopted to train the large parametric model under the knowledge graph Q&A corpus of general domain. The experimental results show that the proposed method has an accuracy of 89.0% on the corpus of fire emergency field, which is better than the traditional Pipeline method, and has better Q&A performance. |
| Keywords: fire emergency; knowledge graph Q&A; information retrieval; semantic matching |
