Research On Multi-turns Retrieval Dialogue Method Based On Semantic Enhancement

In recent years,the field of artificial intelligence has developed rapidly.The humanmachine dialogue system can have a smooth dialogue with human,freeing people from repetitive work.It has great significance to the transformation of human-computer interaction methods.It is important research task in the field of artificial intelligence and natural language process.Among human-machine dialogue systems,the retrieval-based dialog system is an important implementation of the dialogue system.Unlike the generated-based dialogue system,the retrieval-based dialog system simulates the way of search engine works and finds the required response from the candidate answer set.According to the rounds of dialogue history,retrieval-based dialog system can be divided into single-round retrieval-based dialog system and multi-round retrieval-based dialog system.The goal of the multi-round retrieval-based dialog system is to select the most suitable answer from the candidate answer pool based on the dialogue history information and the current question to form a smooth and natural dialogue.The multi-round retrievalbased dialog system is an important research task for humans in exploring machines or systems which is similar to human intelligence.And it is widely used in the fields of intelligent customer service and personal assistants.Therefore,the multi-round retrievalbased dialog system has great research significance and application value.In the early days,the methods of retrieval-based dialog system are mainly rule-based methods,statistics-based methods and deep learning-based methods.However,these methods mainly focus on extracting the information between the conversation history and the candidate answers to solve a single round or a simple multi-round dialogue.When faced with situations that are closer to real conversations and more complex conversation scenarios,such as multiple rounds of conversations,multiple participants,and multiple topic conversations entangled together,these methods focus on deepening the conversation history and the interactive depth of candidate responses These methods cannot take advantage of the characteristics of multiple rounds of dialogue,and cannot extract effective features according to different topic dialogues.The semantic information in the dialogue history is insufficiently extracted,and the accuracy rate is low.In order to solve the above problems,this paper proposes a multi-round retrieval-based dialog system based on semantic enhancement.The main research contents and innovations are as follows:(1)A retrieval-based dialog system method based on conversational matching network is proposed.Inspired by the idea of dialogue disentanglement,the topic information in the sentence is modeled through the conversational matching network,and the knowledge at the dialogue level is extracted into the screening process of candidate answers.(2)A retrieval-based dialog system method with advanced context modeling is proposed.By extracting the indication information in the dialogue from multiple aspects,including hard indication information,soft indication information and dialogue context disentanglement,these indications are incorporated into the semantics vector,and the semantic information of the dialogue system is enhanced.In short,this paper aims at the problem of insufficient semantic information extraction in multi-round retrieval-based dialog systems,and it proposes a method of fusing conversational matching networks and advanced context modeling to enhance the semantics of retrieval dialogue systems.On the one hand,the conversational matching network of the topic information,enhancing the transmission of semantic information between sentences of the same topic,reducing redundant information,and indirectly shortening the history of the conversation;on the other hand,by enhancing the context semantics,the instruction information in the sentence is extracted and integrated into the sentence vector,the semantic information of the dialogue system is enhanced.Experiments show that the conversational matching network can extract topic information in sentences,and advanced context modeling can capture the indication information in sentences,which comprehensively improves the accuracy of the dialogue system.