Memory Bank Augmented Long-tail Sequential Recommendation

Abstract

The goal of sequential recommendation is to predict the next item that a user would like to interact with, by capturing her dynamic historical behaviors. However, most existing sequential recommendation methods do not focus on solving the long-tail item recommendation problem that is caused by the imbalanced distribution of item data. To solve this problem, we propose a novel sequential recommendation framework, named MASR (ie Memory Bank Augmented Long-tail Sequential Recommendation). MASR is an "Open-book'' model that combines novel types of memory banks and a retriever-copy network to alleviate the long-tail problem. During inference, the designed retriever-copy network retrieves related sequences from the training samples and copies the useful information as a cue to improve the recommendation performance on tail items. Two designed memory banks provide reference samples to the retriever-copy network by memorizing the historical samples appearing in the training phase. Extensive experiments have been performed on five real-world datasets to demonstrate the effectiveness of the proposed MASR model. The experimental results indicate that MASR consistently outperforms baseline methods in terms of recommendation performance on tail items.