Research Of Human-Machine Emotional Conversation Based On Hybrid Neural Network

With the development of AI research,most works enhance human-machine interaction by humanizing machine.Humanizing machine means that machine can communicate with user at the human level,can understand human emotion and feedback to use at the emotion level.There are many human-machine conversation techniques,and the most widely used techniques are sequence to sequence model based on deep neural network.But these techniques are realized by single neural network,and don't consider emotion factor.Because of these problems,we propose a study on human-machine conversation based on hybrid neural network.The main works are as follows:We construct a human-machine conversation model based on the hybrid neural network.In order to solve seq2seq problem better and achieve better human-machine conversation,we analyze advantages and disadvantages of three kinds of neural network and make best use of their advantages.First of all,according to specialties of each neural network,we divide the original corpus.Then on different data set,models which are made up of a variety of neural networks are trained tested.Finally,analyzing results and evaluating performance of each model.Study the importance of emotion factor in human-machine emotional conversation.In order to make conversation more humanized,the influence of emotion factor is considered,and concrete manifestation of emotion factor is emotion type of conversation text.On the basis of human-machine conversation based on hybrid neural network,adding utterance and response's emotional type,then observing the effect of the introduction of emotion type on emotional transition of conversation.In order to explore the effect of emotion classification accuracy on conversational emotion level,we propose a text emotion classification model based on extended multi-modality feature and deep belief network.We reclassify emotional conversation corpus by using this model,then train human-machine emotional conversation model again.Finally,we explain the influence of emotional classification accuracy on emotional conversation.