Supervised-Topic-Model-Based Hybrid Filtering for Recommender Systems
Mimu Kawai, Takayuki Shiohama, Hiroyuki Sato · 2017
Recommender systems are beneficial to both service providers and users, as they offer item recommendations to individual users based on their preferences. In this paper, we propose a hybrid recommender system based on a widely used topic modeling method: supervised latent Dirichlet allocation (sLDA). This supervised topic model provides a simple clustering method for analyzing large volumes of unlabeled data from potential items. We use matrix factorization based on sLDA for topic finding in item-feature spaces, enabling hybrid recommendations using topic distributions and a user rating matrix. The proposed model can learn latent factors for predicting unknown ratings and recommending new items, thus addressing the cold-start problem, as any user can be recommended items that have not yet been provided with ratings. Our results show that the proposed method improves upon several baselines for the Jester dataset.