| 摘 要: 意图识别旨在确定用户输入问句中所表达的意图或目的,是实现问答系统的关键步骤。针对建筑施工安全领域智能问答中问句术语繁多,导致意图识别性能不佳的问题,采用 K-BERT模型进行解决。该模型将知识图谱中蕴含的建筑施工安全事故知识三元组,插入问句中形成文本树,并将文本树转换为嵌入式表示和实体可视矩阵,然后进行编码,最终实现对问句的意图识别。在包含6个意图类别的自建语料上进行实验,K-BERT模型意图识别的精确率为99.32%,召回率为99.25%,F1值为99.28%,显著优于 TextCNN、BERT和BERT-TextCNN等基准模型。 |
| 关键词: 建筑施工安全事故 意图识别 知识图谱 K-BERT模型 |
|
中图分类号: TP391
文献标识码: A
|
| 基金项目: 山东省重点研发计划(重大科技创新工程)项目(2021CXGC011204) |
|
| Research on the Application of K-BERT in Intent Recognition for Construction Safety Accident Q&A |
|
ZHANG Wenwen, GAO Yanfang, HOU Guangxin, JI Shengxu
|
(School of Management Engineering, Shandong Jianzhu University, Jinan 250101, China)
dongshi08012022@163.com; gaoyanfang@sdjzu.edu.cn; 205586533@qq.com; ji_sheng_xu@163.com
|
| Abstract: Intent recognition aims to identify the intention or purpose expressed in user queries, serving as a critical step in building question-answering systems. To address the challenge of poor intent recognition performance caused by extensive domain-specific terminology in construction safety Q&A, this study adopts the K-BERT model.The model incorporates knowledge triplets from a construction safety accident knowledge graph into queries to form text tree. This text is then transformed into embedded representations and entity visibility matrices for encoding,ultimately achieving intent recognition. Experiments on a sel-f built corpus containing six intent categories show that K-BERT achieves a precision of 99.32%, recall of 99.25%, and F1-score of 99.28%, significantly outperforming baseline models such as TextCNN, BERT and BERT-TextCNN. |
| Keywords: construction safety accidents intent recognition knowledge graph K-BERT modal |
