Research On Incorporating External Knowledge Into Dialogue Generation

In the era of big data,artificial intelligence technology has achieved rapid development.As an important part of artificial intelligence,human-machine dialogue is expected to become one of the main human-machine interaction modes in the era of the Internet of Things.Dialogue generation is a key part of human-machine dialogue,which can generate machine response according to the content of dialogue and convert this response into natural language feedback to users.The quality of dialogue generation has a direct impact on user experience and reflects the intelligence of human-machine dialogue system to a large extent.Current research on dialogue generation is mainly based on the sequence-to-sequence model.This model is able to automatically learn the semantic information of the conversation from a large amount of dialogue corpus and then generate responses.However,the standard sequence-to-sequence model tends to generate generic responses,which contain less information.These responses are not conducive to the continuity of the dialogue and reduce the user experience of the dialogue system.Knowledge can enrich the expression of language.Incorporating knowledge into the dialogue generation model is able to provide additional information and guide the model to generate more diverse responses,alleviating the problem of generating generic responses.In view of the above problems,this paper will explore how to effectively incorporate external knowledge into the dialogue generation model to generate responses containing richer information,so as to improve the quality of the responses.The main research work of this paper is divided into the following two parts:(1)This paper proposes a dialogue generation model based on knowledge routing.For different dialogue contexts,the knowledge required to generate responses is different.To solve this problem,a knowledge routing module is designed in this model for knowledge selection.The knowledge routing module selects the knowledge in line with the current context by mining the correlation between each piece of knowledge and the connection between the knowledge and the dialogue context utterance,so as to guide the model to generate higher quality responses.(2)This paper proposes a dialogue generation model based on soft fusion decoding.In actual dialogue,a statement containing knowledge information can convey a wealth of information.However,it is not that the more knowledge incorporated into every dialogue sentence,the better.In order to flexibly integrate knowledge into the dialogue,a soft fusion decoder is designed in this model.This decoder can automatically control the proportion of knowledge incorporating according to the dialogue situation,thus increasing or decreasing the influence of knowledge on generating responses.The experimental results show that our model can flexibly use knowledge to generate response sentences that are more consistent with the dialogue situation,which verifies the effectiveness of the dialogue generation model based on soft fusion decoding.