DynFluid: Predicting Time-Evolving Rating in Recommendation Systems via Fluid Dynamics

In trust-based recommendation systems, if a user is predicted to have a high rating of a product, then this product is recommended to that user for shopping potential. Therefore, rating predictions are critical for qualified recommendations. In this paper, based on the fluid dynamics theory, we propose a novel rating prediction scheme called DynFluid. The key observation is that the rating of a user depends on his/her user experience, as well as the ratings of other users. For example, users may refer to friends' ratings upon rating a product, themselves. DynFluid analogizes the rating reference among the users to the fluid flow among containers: each user is represented by a container, the rating of a user is mapped to be the fluid temperature in the corresponding container. Two user characteristics, persistency and persuasiveness, are also incorporated into DynFluid. Finally, real data-driven experiments in Epinions and Ciao validate the efficiency and effectiveness of the proposed DynFluid.