Rational Emotive Human-Machine Negotiation Based On Pre-trained Language Models

The growth of online transactions in e-commerce inevitably necessitates online negotiations.However,manual negotiation cannot meet the growing demand.As a result,automated negotiation has attracted numerous researchers.Nevertheless,most of the work of this kind focuses on computer-to-computer negotiation,with little attention given to human-computer interactions.Furthermore,even research on human-computer negotiation tends to overlook the emotional factors of the human negotiator.Consequently,it fails to address human emotions during negotiations,which can significantly influence the outcome.To address the issue,this paper proposes a novel human-computer negotiation model.Firstly,a human-computer negotiation system must understand what human opponents express in a natural language during the negotiation process.Therefore,in this thesis,we survey Pretrained Language Models(PLMs)and the success of large-scale PLM,such as BERT and GPT,which are essential techniques for solving problems in natural language process.Specifically,we provide an overview of the main machine-learning methods employed by PLM.Furthermore,we explore early PLMs,discuss the current state-of-the-art PLMs,review several Chinese PLMs,and compare their performance.Additionally,we outline the applications of PLMs and offer insights into their future development.Secondly,since we need to detect and respond to human emotions during negotiation,we survey the applications of PLMs in sentiment analysis,which detects,analyses,and extracts the polarity of sentiment expressed in both mono-lingual and cross-lingual texts.Specifically,we outline these methods,compare their performance,and identify challenges for future research.Lastly,we propose a novel human-computer negotiation model.To be precise,we fine-tune PLM ERNIE 3.0 on a dataset we create,enabling the negotiating agent to understand the humans’ intents and analyse their sentiments during negotiation.Moreover,we design fuzzy negotiation strategies to respond to human opponents’ emotions detected during the negotiation process.We conduct numerous experiments to demonstrate the effectiveness of our model.