An Intelligent Chatbot for Negotiation Dialogues
Siqi Chen, Qisong Sun, Ran Su · 2022
In negotiation dialogue tasks, opponents may use various strategies. Such negotiations are challenging because our chatbot needs to be able to detect the behavior change and then to adapt its own policy accordingly. Moreover, instead of adopting stationary strategies, a more advanced opponent may demonstrate sophisticated behaviors by employing reasoning strategies to predict its opponent behavior. To address these challenges, this work proposes a novel chatbot for negotiation dialogues, which leverages the predictive power of Bayesian policy reuse (BPR) and the recursive reasoning ability of theory of mind (ToM). BPR can help the chatbot efficiently detect strategies of an opponent and select the best response policy, and ToM can efficiently identify whether an opponent is using stationary or higher-level reasoning strategies to improve policy selection. The performance of the proposed chatbot is evaluated on the CRAIGSLISTBARGAIN dataset against state-of-the-art baselines. The experimental results show that it outperforms existing agents on the task, and is also able to make efficient detection and optimal response against state-of-the-art baselines.