Adaptive Autoencoders Exploiting Content Preference for Accurate Recommendation

Recommender systems have been extensively studied these days. We note that the contents (e.g., genres or tags of movies) of an item are an important factor to determine whether a user prefers the item. Therefore, if available, it would be promising to take advantage of this information to improve the quality of recommendation. Along this line, this paper presents a novel hybrid recommendation framework called Adaptive Autoencoders Exploiting Content Preference. We first propose to use a new notion of content preference, which implies a user's impressions over different contents. We also propose an adaptive learning algorithm for effective training of Autoencoders used in our framework. We verify the superiority of our framework by comparing it with several baselines using real-world data.