The Research Of Emotional Dialogue Model Based On Recurrent Neural Networks And Reinforcement Learning

The neural dialogue generation is very popular recently.The Seq2 Seq model based on LSTM as one of the typical models,although succeeds in the dialogue generation,there are also highly general response problems,and fall into the problem of endless loops easily and the uncontrollable emotional tendency of dialogue generation.The research topic of this article will research the topic of uncontrollable emotional tendencies in the dialogue generation.Different from the current research,this paper will takes a different approach and uses Generative Adversarial Networks as the theoretical to build an emotional dialogue generation model based on recurrent neural networks and reinforcement learning.The main contributions of this article are as follows:(1)Summarizing the research progress in the generation of dialogue models in recent years,and summarizing their advantages and disadvantages,pointing out the significance of emotional factors in dialogue systems.(2)In the research of emotion control of the generative dialogue model,the emotional tendency of the generated model is guided through the GANs and the reinforcement learning.Compared with other research methods in this direction,this method proposes a new idea and is effective and convenient.(3)This article has designed a brand-new discriminant model.A feature weight model based on the combination of TextRank and Word2 Vec is proposed.Through this model,the feature weights of short texts can be calculated.In addition,the text feature weights are spliced and combined as a vector and a vector obtained from a hierarchical classification model to obtain one.Synthetic vectors of text semantics and text features,thereby improving the classification accuracy of the classification model.The experimental data used in this article was obtained from open source data OpenSubtitles.We designed and trained two generative dialogue models with positive and negative affective based the dataset and the model.These two models are Emotional control under setting has achieved very good results.