Research And Design Of Intelligent Dialogue System For Media Vertical Domai

The human-machine dialogue system is the core technology in the field of artificial intelligence.It is a new way of harmonious human-computer interaction,and has always been a research hotspot in academia and industry.In recent years,due to the breakthrough of deep learning technology in natural language processing,man-machine dialogue technology has made great progress,and various smart speakers and voice assistants have emerged.At the same time,with the pursuit of fashion and personalization,the post-90s have become the main consumer of smart speakers and voice assistants,and the multimedia vertical field(music,movies and TV series)has become a high-frequency skill used by users.Therefore,creating indepth feature audio-visual scenes will bring a broad market to the dialogue system.As the result,this thesis studies and designs an intelligent dialogue system for media vertical domain,including media vertical task dialogue system and knowledge question answering system.Among them,the media task dialogue system aims to provide multimedia audio-visual services by understanding users’ intent,taking "Play Jay Chou’s Qilixiang" as an example,the system should be able to accurately identify its intent and play corresponding resources.As an extension of the media dialogue system,knowledge based Question Answering focuses on answering common-sense questions related to film,television,and celebrities,such as questions like "Who is the director of Infernal Affairs" and "How tall is Huang Xiaoming?"At the same time,in order to create a unique Question Answering,the system also supports some personalized chatting,such as "Is Gu Tianle handsome?","How about Li Xian being my boyfriend?" and so on.The main research contents of this article are as follows:(1)Research on the module of natural language understanding in the media-oriented task dialogue system.The most difficult challenge of media domain semantic understanding is to achieve high-precision understanding in the context of strong entity correlation.Therefore,this thesis designs modules including:entity recognition,entity linking,and intent recognition with knowledge label fusion to improve the accuracy of media domain semantic understanding.(2)Research on the module of natural language understanding in the media-oriented task dialogue system.The most difficult challenge of media domain semantic understanding is to achieve high-precision understanding in the context of strong entity correlation.Therefore,this thesis designs modules including:entity recognition,entity linking,and intent recognition with knowledge label fusion to improve the accuracy of media domain semantic understanding.In general,this thesis designs an intelligent dialogue system oriented to the media domain.Firstly,through the automatic construction of the data set,the dialogue system is started with low resources.Secondly,the system can make full use of media-related knowledge labels,and introduce strong prior knowledge in the media domain to improve the accuracy of semantic understanding.At the same time,the system also supports media-related knowledge question and answer and personalized question and answer,further improving the completeness of the dialogue system.Experiments show that the related algorithms proposed in this thesis have good performance,and the system can also effectively answer media vertical domain questions.