面向高考咨询问答系统的问句分类研究  

Research on the Question Classification of College Entrance Consultation in Question-answering System

作　　者：刘园园 李劲华 赵俊莉 LIU Yuan-yuan;LI Jin-hua;ZHAO Jun-li(School of Data Science and Software Engineering, Qingdao University, Qingdao 266071, China)

机构地区：[1]青岛大学数据科学与软件工程学院,青岛266071

出　　处：《青岛大学学报（自然科学版）》2021年第1期18-24,28,共8页Journal of Qingdao University(Natural Science Edition)

基　　金：国家自然科学基金(批准号:61702293)资助;山东省重点研发计划(批准号:2019JZZY020101)资助。

摘　　要：为建立一个高质量的问答系统,在建立高校信息知识图谱的基础上,提出一种在问答系统领域进行问句分类的方法,并构建了新的分类模型:基于改进的支持向量机模型、融合注意力机制的双向长短时记忆网络(BiLSTM-Attention)模型和BERT-BiLSTM相似度计算模型,并与BERT微调模型作比较。研究结果表明,本问句分类方法能获得较高的问句分类准确率,BERT模型具有更好的分类性能,而将Bi-LSTM和BERT进行融合对句子特征的提取能力更强。In order to establish a high-quality question answering system,a method for question classification in the field of question and answering system was proposed on the basis of the constructed knowledge map of colleges and universities.And new classification model is constructed based on an improved support vector machine model,the two-way long and short-term memory network model.And with attention mechanism and the BERT-LSTM similarity calculation model.These three models are compared with the BERT fine-tuning model.The results show that the question classification method in this article can obtain a higher question classification accuracy rate.BERT model has better classification performance,and the fusion of LSTM and BERT can have stronger ability to extract sentence features.

关 键 词：问答系统 问句分类 支持向量机 长短时记忆网络 BERT 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]