Learning pareto optimal solution of a multi-attribute bilateral negotiation using deep reinforcement

Abstract

Abstract This paper aims to design an intelligent buyer to learn how to decide in an incomplete information multi-attribute bilateral simultaneous negotiation. The buyer does not know the negotiation strategy of the seller and only have access to the historical data of the previous negotiations. Using the historical data and clustering method, the type of seller is identified online during the negotiation. Then, the deep reinforcement learning method is utilized to support the buyer to learn its optimal decision. In the complete information case, we prove that the negotiation admits a unique Nash bargaining solution with possibly asymmetric negotiation powers. In comprehensive simulation studies, the efficiency of the proposed learning agent is evaluated in different scenarios and we show that the learning negotiation with incomplete information is converged to a Pareto optimal solution. Then, using the concept of the Nash bargaining solution, the negotiation power of the buyer is assessed in negotiation.