A Human-computer Negotiation System Based On Big Data

With the rise of the artificial intelligence boom and the popularisation of the Internet,ecommerce has ushered in a new opportunity for development.At present,artificial intelligence can be applied to all aspects of e-commerce,such as intelligent customer service in the field of e-commerce.However,current intelligent systems of customer service only can simply respond to customers according to their input keywords,and they can not conduct the automated negotiation.On the other hand,the current e-commerce practice generates a lot of data every day,which is very valuable in e-commerce.For example,we can use big data to recommend commodities to customers.Besides,big data can also improve the negotiation between sellers and buyers and help them make more or better deals.Nevertheless,little work utilises big data in human-computer negotiation.To this end,in this thesis,we develop a human-computer negotiation based on big data.The system can autonomously negotiate with customers in natural language.Specifically,first,we propose a method for the negotiation system to understand and extract the intent and information in the natural language input of human customers.Then we propose a set of methods for analysing customers' big data,including historical orders analysis,overall consumption analysis,customer distribution analysis and browsing history analysis.Moreover,we design a price negotiation decision algorithm based on big data analysis.Finally,we propose a reply generation method,which generates a natural language reply to a customer according to the output of the price negotiation decision algorithm.We do a lot of experiments to evaluate our system's quality and usability.Specifically,we compare outcomes of the customer intent recognition module and the reply generation module evaluated by various classifiers to assess our system's quality criteria.Then,we conduct 100 simulated human-computer negotiations with big data and 100 simulated human-computer negotiations without big data to analyse the usability of our system.Our system's usability is measured by the success rate,the yield rates of different sellers and customers,and the overall tendency of the system.In conclusion,our evaluation experiments show that our system is effective in human-computer negotiation.