An Emotion-Aware Human-Computer Negotiation Model Powered by Pretrained Language Model

The growth of e-commerce has led to increasing online transactions, inevitably requiring online negotiations. However, manual negotiation cannot meet the growing need. As a result, automated negotiation attracts lots of researchers. However, most work of this kind is on computer-computer negotiation and a little on human-computer one. Moreover, even the research on human-computer negotiation tends to ignore the emotional factors of human negotiators. So, it cannot deal with human emotions during a negotiation, which may significantly influence its outcome. To this end, this paper proposes a novel human-computer negotiation model. First, we fine-tune the ERNIE pretrained language model on a dataset we create. Then, the negotiating agent uses it to understand the intents and emotions of human dialogue in negotiation. Finally, the agent responds to humans according to the sentiment-related negotiation strategy we designed in this paper. Our extensive experiments show the effectiveness of our model.