A phrase-based model to discover hidden factors and hidden topics in recommender systems
Liping Guan, Md. Hijbul Alam, Woo-Jong Ryu, SangKeun Lee · 2016
The majority of existing recommender systems focus on modeling the ratings; however, these systems ignore a large number of reviews. Existing rating based recommender systems are hard to discover the hidden dimensions in human feedback that can identify user preferences. In this study, we combine collaborative filtering with latent review topics to generate a new model called phraseHFT. We apply reviews to the phrase-level document, and use the phrase-based topic model to discover the review topics that are embedded in the review text. The interpretable topics which are learned and presented as phrases can help us understand characteristics of users or items. The conducted experiment shows that our approach outperforms the state-of-the-art techniques in perplexity and topic visualization due to the strong topic learning functionality.