Dialogue System Based On Mixed Mode Of End-to-end And Case Inference

Dialogue system is a human-computer interaction system which communicates through natural language with people.It is positioned as a gateway to various services in the future.Dialogue system has a wide range of applications in the areas of online customer service,entertainment,education,and personal assistants.It has great research significance and application value.The traditional dialogue system is generally designed based on the retrieval method,there are problems such as difficult to maintain the knowledge base,poor scalability,etc.To solve this problem,this paper designs the dialogue system based on the combination of the deep learning generation method and the case-based reasoning method,which effectively reduces the cost of maintaining the knowledge base and the rules,and at the same time improves the extensibility of the dialogue system.First of all,this paper uses the sequence-to-sequence method to build a dialogue model,and then uses reinforcement learning to further optimize the sequence-to-sequence model.Finally,the generated dialogue system is combined with the CBR inference engine,which is applied to the intelligent robot fruit sorting system.The main work includes the following sections:(1)The influence of model structure and model solving algorithm on the Seq2 Seq model generation dialogue is studied.By selecting the optimal network structure,parameter search strategy,parameter initialization,introducing attention mechanism,optimizing the objective function and other means,the accuracy of model generation dialogue is improved.(2)Reinforcement learning is used to improve the problem of short conversations and meaningless responses in the Seq2 Seq model.And a new reward functions are designed based on the actual needs of intelligent robots.Using the Chinese subtitle corpora training model,the effect of the model was improved by word segmentation,and a triple-based assessment method was used to evaluate the model.(3)Combining the CBR inference engine with the Seq2 Seq model,a dialogue system based on deep learning was designed.The dialogue system was used in the intelligent robot fruit sorting system.Finally,the dialogue system was tested from the three aspects of task completion rate,response fluency,and user satisfaction,which verified the reliability and intelligence of the dialogue system.