E-commerce Product Recommendation by Personalized Promotion and Total Surplus Maximization

Existing recommendation algorithms treat recommendation problem as rating prediction and the recommendation quality is measured by RMSE or other similar metrics. However, we argued that when it comes to E-commerce product recommendation, recommendation is more than rating prediction by realizing the fact price plays a critical role in recommendation result. In this work, we propose to build E-commerce product recommender systems based on fundamental economic notions. We first proposed an incentive compatible method that can effectively elicit consumer's willingness-to-pay in a typical E-commerce setting and in a further step, we formalize the recommendation problem as maximizing total surplus. We validated the proposed WTP elicitation algorithm through crowd sourcing and the results demonstrated that the proposed approach can achieve higher seller profit by personalizing promotion. We also proposed a total surplus maximization (TSM) based recommendation framework. We specified TSM by three of the most representative settings - e-commerce where the product quantity can be viewed as infinity, P2P lending where the resource is bounded and freelancer marketing where the resource (job) can be assigned to one freelancer. The experimental results of the corresponding datasets shows that TSM exceeds existing approach in terms of total surplus.