Reinforcement Learning for Dialogue Generation: A Systematic Literature Review

Abstract

With the rapid advancement in personal assistants such as Siri, Alexa, and Cortana among many others, dialogue systems have become a very prominent issue in recent years. Open-ended dialogue systems function best in languages with a large volume of text and speech data.. Conversational agents in constrained resource languages are still not quite as developed due to the difficulty of data sparsity, which is one of several challenges to establishing robust and effective dialogue systems. To get a great result, traditional RL techniques take thousands of dialogues. To carry out this SLR, 28 research papers were carefully chosen based on predefined selection criteria. In this study, we presented a taxonomy of reinforcement learning considering its components, goals, and basis and discuss the reinforcement learning framework which may provide researchers with understandings of the dialogue generating aspect. The purpose of this study is to identify and discuss current research on dialogue generation, as well as to choose valuable methodologies and requirements for future study. The findings as well as the research questions are addressed. Finally, challenges and issues have been given to give research with possible future paths in the domain of dialogue generation.