Emotional conversation generation with heterogeneous graph neural network

The successful emotional conversation system depends on sufficient perception and appropriate expression of emotions. In a real-life conversation, humans firstly instinctively perceive emotions from multi-source information, including the emotion flow hidden in dialogue history, facial expressions, audio, personalities speakers. Then, they convey suitable according to their personalities, but these multiple types information are insufficiently exploited fields. To address this issue, paper, we propose heterogeneous graph-based model for generation. Firstly, design Heterogeneous Graph-Based Encoder represent content (i.e., its flow, speakers' personalities) with graph neural network, then predict feedback. Secondly, employ an Emotion-Personality-Aware Decoder generate response relevant context as well emotions, through taking encoded representations, predicted by encoder personality current speaker inputs. Experiments both automatic human evaluation show that our method can effectively knowledge satisfactory response. Furthermore, based up-to-date text generator BART, still achieve consistent improvement, which significantly outperforms some existing state-of-the-art models.