LinRec: Linear Attention Mechanism for Long-term Sequential Recommender Systems

What is it about?

The sequential recommendation is to recommend user items based on their historical interactions (clicks, views, paid, etc.). This is a work about how to use a novel linear attention mechanism in sequential recommendation. Literally, 'linear' means we reduce the computational complexity from square to linear, which may save a lot of GPU memory and time cost without losing the recommendation accuracy.

Featured Image

Why is it important?

The square complexity is the bottleneck of applying the transformer, while the transformer is the most popular sequential recommender systems (SRSs) technique. So solving the computational complexity issue is beneficial and necessary for the implementation of SRSs in industries.