AICF: Attentional Item-interaction Collaborative Filtering

Abstract

As of today, deep learning technology has been widely utilized in the field of recommendation algorithms. Many work utilizes the multilayer perceptrons to simulate inner product operations in matrix factorization (MF) models. Hence, those works inherit a weakness of MF, namely the ineffectiveness of identifying between a small group of closely related items. The association of items, such as high-order relations between items, endows the recommendation system with valuable information on the user’s final choices. In accordance with the above facts, this paper proposes an Attentional Item-Interaction Collaborative Filtering model (AICF). It can distinguish which items in the user’s historical interaction items play a more important role in the user’s choice of items. To examine the performance of AICF, we designed multiple experiments on two real world data sets.