摘要
Pipeline结构是任务导向型对话系统设计中非常成熟的一种构建方案,能够完成领域内特定的任务型对话。由于实际环境的复杂性,任务流程外的领域相关常见问题经常被用户提出,而pipeline结构不能有效处理此类问题。提出一种基于多模型的融合方法,将使用检索方式实现的非任务型对话模型有效融合到pipeline结构中,同时确保不同模型能够高效地协同处理不同类别的应答需求。实验证明,该方法能够有效提高系统的意图识别和对话应答的准确率,确保对话系统能够应对更加复杂的实际对话环境。
The pipeline structure is a very mature framework in the task-oriented dialogue system design.This framework is capable of completing specific task-based dialogues.Due to the complexity of actual environment,domain-related questions out of preset task flow are frequently asked by the users.However,the pipeline structure cannot effectively handle such problems.This paper proposes a multi-model based fusion method.The method effectively integrates retrieval-based non-task-oriented dialogue model into pipeline framework,ensuring that different models can efficiently co-ordinate different types of response needs.Experiments show that this method effectively improve the accuracy of the system’s intent recognition and dialogue response,and ensure that the dialogue system can cope with a more complex actual conversation environment.
作者
李响
张磊
刘媛媛
LI Xiang;ZHANG Lei;LIU Yuanyuan(State Key Laboratory of Mathematical Engineering and Advanced Computing, Zhengzhou 450001, China;State Key Laboratory of Intelligent Technology and Systems, Tsinghua University, Beijing 100084, China;Knowledge Engineering Group, Department of Computer Science and Technology, Tsinghua University, Beijing 100084, China)
出处
《信息工程大学学报》
2019年第4期473-479,共7页
Journal of Information Engineering University
基金
国家重点研发计划资助项目(2018YFC0830200)
国家自然科学基金资助项目(61876096,61332007)。
关键词
任务导向
对话系统
信息检索
多模型
神经网络
task-oriented
dialogue system
information retrieval
multiple models
neural networks
作者简介
李响,(1989-),男,硕士生,主要研究方向为人工智能自然语言处理技术;张磊,(1990-),男,硕士生,主要研究方向为人工智能自然语言处理技术;刘媛媛,(1991-),女,硕士生,主要研究方向为人工智能自然语言处理技术。
