Research On Bilateral Interdependent-Issues Negotiation Model Based On Offline Learning Mechnism

In the field of E-Commerce, agent automated negotiation theories and techniquesbecome an important mean for participants to achieve their respective interests and toresolve the dispute that may exist in the service side and client side. Bilateralmulti-issue negotiation is an important research topic in this field. Most of negotiationmodels that based on the agenda, similarity, case and other traditional models did notfully consider the interdependence between the values of issues, which willundoubtedly increase the complexity of the multi-issue negotiation problem.Therefore, constructing efficient bilateral multi-issue interdependent negotiationmodel has become a hot topic in the negotiation research field in recent years.In recent years, there are many achievements of multi-issue interdependentnegotiation model appeared. For example, transforming the multi-issue negotiationproblem into the multi-objective optimization problem, which use space searching to get asolution.But this method did not describe the interdependences and its computationalcomplexity was high. In the utility graphs-based negotiation model, issues are usuallybinary value (0or1). If each issue has many values, this model will be limited in thenegotiation process. The GAI network can represent the interdependence betweeneach subset of issues, but if issues’ value spaces are continuous or large, theGAI-based model will be limited to some restrictions.In this thesis, the proposed bilateral multi-issue negotiation model builds on thebasis of offline learning mechanism and covers some important points in the processof multi-issue negotiation in case of interdependence between the values of issues.Firstly, we propose the internal framework of bilateral negotiation model andformalize all parts of the model. Secondly, we propose a new offline learningmechanism, which mainly consists of two components: the construction of the libraryassociation rules and C-GNN algorithm. For the construction of the libraryassociation rules, we use the association rule generation mechanism and theassociation rules optimization mechanism to achieve and use the association rules thatdescribe the interdependence between the values of issues. At the same time, theC-GNN algorithm is used to obtain the corresponding predictable neural networkstructure through the study of historical cases as a first step, and then use thecorresponding predictable neural network to predict the value effectively, which is an acceptable result of the association issues as a second step. Thirdly, we propose twoextended time-dependent negotiation strategies: time-dependent strategy based onassociation rules and segmentation time-dependent strategy based on the predictivevalue. These two extended strategies built on the basis of time-dependent negotiationstrategy combined with association rules between issues and values, which aregenerated by the corresponding association predictable neural network. Finally,comparing with the traditional agenda-based model, the experimental results showthat the proposed bilateral interdependent-issues negotiation model improves theoverall utility and efficiency. Thus, we further verify the feasibility and effectivenessof the model.