Collaborative Filtering For Implicit Feedback

A common task of recommender systems is to improve customer experience through personalized recommendations based on prior implicit feedback. These systems passively track different sorts of user behavior in order to model user preferences. Collaborative filtering for implicit feedback mainly has following problems:1) data sparseness. The numbers of users and items in major e-commerce recommendation systems is very large. Even users that are very active result in consuming just a few of the total number of items available in a database and respectively, even very popular items result in having been consumed by only a few of the total number of users available in the database.2) no negative feedback. All of the negative examples and missing positive examples are mixed together and cannot be distinguished.3) the data can’t reflect user preferences. the unobserved user-item pairs can be interpreted in many different ways. For example, the reason why a user did not purchase a product could be that she was simply not aware of it. In order to solve the above problems, this paper presents collaborative filtering for implicit feedback based on factorization machine and one-class collaborative filtering takes two-step strategy.The main contributions are summarized as follows:(1) To overcome the sparsity problem of the data, we use a context-aware recommendation algorithm which exploiting the rich user information and item information.(2) Aiming at generating negative examples, we replace the user-oriented random sampling scheme with the user-oriented sampling based on matrix factorization.(3) To reflect the user preferences, we replace the uniform weighting scheme with the ranking weight function. The experimental results show that our approaches significantly outperform the baselines.