基于自适应扩展机制的领域智能问答系统  

Domain Intelligent Q&A System Based on Adaptive Extension Mechanism

作　　者：乔奋凤 朱欣娟[1] 高岭 QIAO Fen-feng;ZHU Xin-juan;GAO Ling(School of Computer Science,Xi’an Polytechnic University,Xi’an 710600,China)

机构地区：[1]西安工程大学计算机科学学院,陕西西安710600

出　　处：《计算机技术与发展》2021年第12期13-20,共8页Computer Technology and Development

基　　金：国家重点研发计划项目(2019YFC1521405);陕西省重点研发计划(2019ZDLSF07-01);2020年西安工程大学研究生创新基金重点项目(chx2020027)。

摘　　要：针对目前智能问答系统采用单层网络模型理解用户意图,未能准确关注用户语句中的细节特征的问题,提出了一种基于关键词分离的双层网络模型用户意图识别方法。第一层使用双向长短时记忆网络和条件随机场模型对用户语句中的关键词及问题句式进行识别,第二层将识别出的关键词作为细节特征,采用融合注意力机制的双层双向长短时记忆网络进行问题类型的识别,两层识别的结果为用户意图。实验证明,该方法的准确率和召回率平均提升了6%。针对用户数据较少时智能问答系统仍要扩展的需求,提出基于自适应扩展的智能问答系统优化方法。该方法使用基于句法结构的层次聚类算法对未识别的用户问题进行聚类,定期更新问题类型库。实验证明,基于句法结构的层次聚类算法正确率可达76%。In order to solve the problem that the intelligent question answering system uses the single-layer network model to understand the user’s intention,but fails to pay attention to the details of the user’s statements,we propose a hierarchical user intention recognition approach based on keyword separation.In the first layer,a bidirectional long short memory network and conditional random field model are used to pick the keywords out and label the sentence patterns.In the second layer,the previously selected keywords are used as input detail features,and a two-layer bidirectional long short memory network with attention mechanism is used to identify the question types.Experiment shows that the accuracy and recall of the proposed approach are improved by 6%on average.According to the needs that intelligent question answering system still needs to be expanded when the user data is small,an optimization method of intelligent question answering system based on adaptive expansion is proposed.The unrecognized user questions can be clustered by using the hierarchical clustering algorithm based on syntactic structure,and the question type library can be updated regularly.Experiments show that the accuracy of the hierarchical clustering algorithm reaches 76%.

关 键 词：关键词分离 智能问答系统 用户意图识别 自适应扩展 句法结构 层次聚类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]