A study of Top-N recommendation on user behavior data

Traditional collaborative filtering algorithms are designed mostly for making rating predictions. Due to actual applications prefer directly item recommendation rather than the item rating and most of the existing systems do not have user explicit ratings on items, in this paper we focus on the implicit feedback on which filtering approach is constructed to provide users with Top-N recommendation. For the special nature of implicit feedback, a binary matrix is used to represent user-item information. On this basis, we analyze the effectiveness of the various existing typical binary similarity methods applied on the traditional collaborative filtering algorithm. Besides, as the current neighbor selection does not take into account the combined effects of neighbors, this paper adopts a new way that considers the neighbor similarity to avoid selecting too similar neighbors. The experiments conducte on the real dataset show that the traditional recommendation performance can be effectively improved by our methods.