LAUSR

Sequential recommendation tasks predict items to be interacted at the next moment according to users’ historical behavior sequences. A large number of studies have shown that accuracy is not the only evaluation metric in the sequential recommendation tasks. Diversity can measure homogeneity of items to capture the changes of users’ interests in the sequences. Integrating multiple interests of users has become the focus of current research. The existing multi-interest sequential recommendation methods adopt the method of self-attention, but it is based on the self-attention of transformer, which lacks the consideration of the correlation between different samples. Therefore, we propose an Enhanced Attention (EA) framework, which is based on two linear layers and two norm layers. Compared with self-attention, it not only reduces the high computational complexity, but also obtains the correlation between different samples. Multi-head mechanism is also applied to EA framework. We conduct experiments for the sequential recommendation on three real-world datasets, Amazon, Taobao and MovieLens-1M. The experimental results show that the EA framework is significantly improved compared with the current state-of-the-art models.