Dialog Systems

Open domain dialogues face the challenge of being repetitive and producing generic responses. In this paper, we demonstrate that we are able to interpret the discrete dialog attributes and composited attributes, and it helps to improve the model and results in various and interesting non-redundant responses. We propose to formulate the dialog attribute prediction and reinforcement learning (RL). Controlling existing RL approaches which formulate the token to a policy, to reduce the complexity of the policy to reduce the complexity of the problem.