Perceiving the World: Question-guided Reinforcement Learning for Text-based Games
This addresses challenges in applying deep reinforcement learning to real-world text-based games, though it is incremental as it builds on existing methods.
The paper tackles low sample efficiency and large action space in text-based games by introducing world-perceiving modules that decompose tasks and prune actions, resulting in significant performance and sample efficiency improvements with robustness against errors and limited data.
Text-based games provide an interactive way to study natural language processing. While deep reinforcement learning has shown effectiveness in developing the game playing agent, the low sample efficiency and the large action space remain to be the two major challenges that hinder the DRL from being applied in the real world. In this paper, we address the challenges by introducing world-perceiving modules, which automatically decompose tasks and prune actions by answering questions about the environment. We then propose a two-phase training framework to decouple language learning from reinforcement learning, which further improves the sample efficiency. The experimental results show that the proposed method significantly improves the performance and sample efficiency. Besides, it shows robustness against compound error and limited pre-training data.