RSS item ranking based on implicit feedback

Really Simple Syndication(RSS) has been widely used in our daily lives, but RSS doesn´t always collect interesting articles, user has to sift through every subscription for articles they like. The ranking of unread RSS articles has the potential power to release user from this heavy burden. Although user preferences can be learned from explicit feedbacks such as rating or tagging, implicit feedback techniques which collect information by monitoring users´ natural behavior instead of asking them to explicitly give feedbacks become more attractive. There are several implicit feedbacks including scrolling and gazing, however, some of them may not be effective or require special equipments. In this article, we present a model-based approach for ranking unread RSS items and make recommendations via implicit behavior analysis. Since time spent on reading has been proved to be a useful measurement for user modelling, the approach aims at learning predictive models of user preferences from their implicit feedback of reading time only. User preference are expressed by latent semantic classes, and Bayesian inference with a set of heuristics is used to reveal the probability that an article may attract the user. Experimental evaluation shows substantial improvements over existing methods.