Fair Recommendations for Online Barter Exchange Networks

Of late online social networks have become popular, with interest spanning various aspects including search, analysis/mining, and their potential use for item barter exchange markets. The idea is that users can leverage their social network for exchanging items they possess with other users. The problem of generating recommendations for item exchanges between users, consisting of synchronous exchange cycles has been investigated[2]. In this paper, we identify the shortcomings of the above exchange model and propose an asynchronous model that makes use of credit points. Rather than insist on exchanging items synchronously, we award points to users whenever they give items to other users, which can be redeemed later. Points and their redemption raise an issue of fairness which intuitively means users who contribute more should have a greater priority over others for receiving items they wish for. We focus on fairness maximization and prove that it is NPhard and cannot be approximated within any factor in polynomial time unless P=NP. We then develop efficient heuristic algorithms, and experimentally demonstrate their effectiveness and scalability on both synthetic data and a real dataset from readitswapit.co.uk.