Research On Dialogue Generation Algorithm Based On Background Knowledge Perception

Affected by the coverage of Internet technology and the proliferation of mobile devices,artificial intelligence has gradually penetrated into all aspects of business life built on the Internet.As an important product in the field of artificial intelligence,thanks to the support of data and the advancement of deep learning technology,the dialogue system has gradually developed into a more user-friendly.The dialogue system has gradually developed from a task type in a specific field to a chat type in an open field,and from a retrieval type under fixed rules to a dialogue generation type under the Seq2 Seq framework.Based on research status at home and abroad,the generation-based dialogue system can generate reasonable answers that are not available in the database without the constraints of rules,and the model has good transferability and flexibility.However,the current dialogue generation model still has problems such as over-reliance on specific data sets,single response,low information content,and chaotic logic in dialogue generation.Aiming at the crux of the current problem and the current needs of the dialogue system,this thesis explores the reply generation of the chat dialogue system from the dialogue context and knowledge graph.The main work is as follows:(1)Explicitly weighted context.To make full use of the historical information in the dialogue data for dialogue generation,this thesis introduces the point mutual information algorithm in the context utilization,by finding the dependency between the dialogue context and the current dialogue as the relevant value of the dialogue context,and further implements context fusion by means of concatenation and summation,finally realizes dialogue reply generation based on explicit weighted context encoding.The key of this method is to calculate the correlation between sentences,so that the model can focus on the context information that is highly relevant to the current dialogue.(2)Background knowledge perception.In order to solve the problem of low knowledge content and single answer in the dialogue reply,this thesis introduces an external knowledge graph,on the one hand,it makes up for the lack of background knowledge of the dialogue system in the dialogue with people,on the other hand,it can make full use of context information.In terms of specific implementation,first retrieve relevant knowledge based on the current dialogue,but there is information unrelated to the current dialogue in the retrieval knowledge,which inevitably affects the quality of reply generation.Therefore,this thesis selects the retrieved knowledge based on the context encoding that is highly relevant to the current dialogue,and further extends the relevant context to the selected knowledge in the form of sentences for reply generation.This thesis firstly completes the explicit weighting of the dialogue context through the point mutual information algorithm,and realizes the effective use of the dialogue context in the dialogue generation.In addition,through specific experimental analysis,the explicit weighted context encoding model constructed in this thesis can capture richer dialogue information and generate more diverse and fluent responses.Further,to make up for the lack of background knowledge in dialogue generation,this thesis starts from the needs of current chat dialogue systems,and studies dialogue generation algorithms combined with dialogue context and knowledge graph in the open field,in the end,the evaluation of the model in this thesis on both Chinese and English datasets is outstanding,outperforming the state-of-the-art method CCM,and through case analysis,the dialogue generation model based on background knowledge perception constructed in this thesis adds knowledge information to dialogue generation,fundamentally solves the single problem of dialogue generation and response.In general,the research on dialogue context and background knowledge perception in this thesis promotes the humanized development of dialogue generation systems,and also provides research value for subsequent research on artificial intelligence products.