Leveraging attribute latent features for addressing new item cold-start issue

A recommender system employs an information filtering technology aiming to recommend items that are likely to be of interest to users, based on user behavior, rating feedback of items, or item characteristics. The cold-start problem occurs when the recommender system has difficulties in drawing any recommendations due to lack of information. In this paper, we focus on the item cold-start problem. When new items are added to the catalogue, they can easily be overlooked because no feedback information is accessible for recommending them to users, impeding the promotion of new products online. We propose a hybrid recommender system, called ALFNCF (Attribute Latent Features with Neural Collaborative Filtering), which combines the advantages of collaborative filtering, content-based filtering, and neural network technologies, to address the item cold-start issue. Like collaborative filtering, ALFNCF considers the interaction between users with similar behavior and items with similar feedback, while it also takes into account the influence of user's and item's attributes on individual behaviors. As in content-based filtering, attribute information is used in ALFNCF to establish links between new and old items. ALFNCF not only can retain the linear interactions between users and items, but also can learn the non-linear interactions by neural network units. Through the training on users' past rating feedback information for old items, ALFNCF can predict the user's ratings for new items. Experimental results show that our proposed method is effective and superior to other methods in the promotion of new items.

Automated summary

This paper discusses the cold-start problem in recommender systems and proposes a hybrid recommender system called ALFNCF to address this issue. ALFNCF combines collaborative filtering, content-based filtering, and neural network technologies, and predicts user ratings for new items based on training on past rating feedback information.