B1709
Title: Continuous-time recommender system with evolutionary temporal feature process embedding
Authors: Xiwei Tang - University of Virginia (United States) [presenting]
Abstract: Large volumes of temporal event data are drawing increasing attention in a wide variety of applications, such as in analyzing social media data, healthcare records, online consumption, and product recommendation. For the recommender system, traditional models based on static latent features or discretized time epochs usually fail to capture the important temporal dynamics in user-item interactions. We propose a novel evolutionary recommender system by leveraging the temporal mechanism on the continuous-time user-item interactive events. The proposed approach can effectively capture the long- and short-term preferences from the sequential historical data with informative dynamic feature embeddings. We develop an efficient algorithm for learning the model parameters with outstanding scalability and computational effectiveness. Using both synthetic and real-world datasets, we show the outperformance of the proposed model in learning sequential user behaviors and achieving better predictive power in the recommendation.
