Key points are not available for this paper at this time.
Trust-based recommendation systems study how people form opinions via trusted friends, so as to predict unknown ratings based on the ratings expressed by trusted friends. Most of the existing work only considers the ratings at the current time slot. In real life, a user's opinion evolves over time, since he receives the influence of different opinions sequentially. In addition, existing work usually targets a single user at a time; there is a need to predict multiple ratings for multiple connected users. To reach these ends, we propose a novel multiple-rating prediction scheme, FluidRating, which uses fluid dynamics theory to reveal the time-evolving formulation process of human opinions. In this scheme, each user corresponds to a container, and several containers are connected through single directional pipes, corresponding to influence relations. We identify three features of human personality in the opinion formulation and propagation process: “persistency” represents how much one insists on his opinion, “persuasiveness” represents the ability to impact others, and “forgetting” reflects the common truth that people have limited memory. The recommendation (or influence) is modeled as fluid with two dimensions: its temperature is taken as the “opinion/rating,” and its height is deemed as the persistency. When new opinions emerge, each person refines his opinion through a round of fluid exchange with neighbors. Opinions of multiple rounds are aggregated to gain a final prediction. Experimental evaluation in a real data set validates the feasibility and the effectiveness of the proposed model.
Jiang et al. (Fri,) studied this question.