Towards platform profit-aware fairness in personalized recommendation

The remarkable progress of machine learning has had a significant impact on decision-making, thus fairness is an important topic. Existing fair recommendation methods generally design a constraint framework between consumer satisfaction or item exposure in the optimization model. However, these methods only focus on the fairness of consumers and items, but fail to ensure the fairness of platforms. Ignoring platform fairness may lead to an imbalance between consumer satisfaction and platform profit. In this paper, we propose a novel recommendation method based on fairness called CPG-FairRec, which consists of three modules: Data division module, Global fairness-aware module and Local fairness-aware module. Data division module divides the consumers and items into two groups based on spendings and prices, respectively. Global fairness-aware module learns the difference_score of the item group and consumer group based on the consumer’s historical behavior, and continuously balance the satisfaction of consumer group and the exposure of item group. Local fairness-aware module aims to optimize the exposure of individual item through the greedy algorithm based on the item price to achieve higher platform profit. We conduct experiments on two real-world datasets and results demonstrate the superiority of the CPG-FairRec in recommendation quality and profitability.