A Hybrid Aspect Based Latent Factor Model for Recommendation

Abstract

Recommender system has been recognized as a superior way for solving personal information overload problem. More and more aspect-based models are leveraging user ratings and extracting information from review texts to support recommendation. Aspect-based latent factor model predicts user ratings relying on latent aspect inferred from user reviews. It usually constructs only a single global model for all users, which may be not sufficient to capture the diversity of users' preferences and leave some items or users be badly modeled. We propose a Hybrid aspect-based latent factor model (HALFM), which jointly optimizes the Global aspect-based latent factor model (GALFM) and the Local Aspect-based Latent Factor Models (LALFM), their user-specific combination, and the assignment of users to the LALFMs. HALFM makes prediction by combining user-specific of GALFM and many LALFMs. Experimental results demonstrate that the proposed HALFM outperforms most of aspectbased recommendation techniques in rating prediction.