Prompt-Based Monte-Carlo Tree Search for Goal-Oriented Dialogue Policy Planning
This addresses the problem of data scarcity and noisy annotations in goal-oriented dialogue planning for AI systems, offering a training-free approach that is incremental over existing MCTS methods.
The paper tackles goal-oriented dialogue policy planning by introducing GDP-Zero, which uses Open-Loop MCTS with a large language model as a policy prior, value function, user simulator, and system model, eliminating the need for model training. Results show that GDP-Zero's responses are preferred over ChatGPT up to 59.32% of the time and rated more persuasive in interactive evaluations on the PersuasionForGood task.
Planning for goal-oriented dialogue often requires simulating future dialogue interactions and estimating task progress. Many approaches thus consider training neural networks to perform look-ahead search algorithms such as A* search and Monte Carlo Tree Search (MCTS). However, this training often requires abundant annotated data, which creates challenges when faced with noisy annotations or low-resource settings. We introduce GDP-Zero, an approach using Open-Loop MCTS to perform goal-oriented dialogue policy planning without any model training. GDP-Zero prompts a large language model to act as a policy prior, value function, user simulator, and system model during the tree search. We evaluate GDP-Zero on the goal-oriented task PersuasionForGood, and find that its responses are preferred over ChatGPT up to 59.32% of the time, and are rated more persuasive than ChatGPT during interactive evaluations.