CoEA: A Cooperative–Competitive Evolutionary Algorithm for Bidirectional Recommendations

Abstract

In recent decades, recommender systems have been well studied and widely applied. However, most recommenders unilaterally optimize the results from the buying customers’ views without considering expectations of other participants, e.g., merchants. Unfortunately, the expectations of customers and merchants in recommendation are different or even conflicted. Especially for popular group-trading markets, customers and merchants are competing in trading, i.e., customers want to meet their preferences or obtain gains with personal favorite items, while merchants want to recommend wholesale items with setting group-trading terms or conditions. In addition, some practical constraints are not fully considered by prior systems. In this article, we propose a cooperative–competitive evolutionary algorithm (i.e., CoEA) for the bidirectional recommendations in group-trading markets. Specifically, we, respectively, formalize two subproblems with designed objectives for two-sided participants in markets, and integrate the cooperative–competitive optimizations into one framework. Second, CoEA designs a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">binary encoding matrix</i> for individual representation to integrate the two subproblems. Furthermore, by assembling game evolution process, CoEA designs cooperative–competitive evolution operators, i.e., the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">cooperative crossover</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">competitive mutation</i> , which guide the solutions to equilibrium by, respectively, bridging communication between two populations of subproblems and optimizing distinctive objective in each population. Finally, we construct two real applications involving bidirectional recommendations, i.e., the group buying and P2P lending, and conduct extensive experiments with the real-world datasets. By comparing CoEA with several representative recommendation algorithms and evolutionary algorithms, the experimental results clearly demonstrate the effectiveness of CoEA.