CADPP: An Effective Approach to Recommend Attentive and Diverse Long-tail Items

Abstract

As the long-tail items are widely seen in various recommendation related applications, e.g., E-commerce and music recommendation, the long-tail recommendation consequently becomes an important research issue attracting both academic and industrial attentions. Apparently, it is a very challenging practical issue and the corresponding key challenges to address this task is to find the long-tail items which best match users’ preferences but are sufficiently diverse to avoid recommending similar long-tail items. To address this issue, this paper proposes a novel long-tail item recommendation approach which is based on the multi-pointer co-attention mechanism and the determinant point process (abbreviated as CADPP). Specifically, we design the multi-pointer co-attention mechanism for extracting important feature embeddings to capture the common characteristics of multiple items clicked by the users. We also employ the determinant point process (DPP) to allow diverse long-tail items but are relevant to the target items. To evaluate the model performance, extensive experiments have been performed on two real-world datasets. The promising results have demonstrated that the proposed CADPP is superior to both baseline and the state-of-the-art approaches with respect to the widely adopted evaluation metrics.