Attention based collaborative filtering

Attention mechanism invention was an important milestone in the development of the Natural Language Processing domain. It found many applications in different fields, like churn prediction, computer vision, speech recognition, and so on. Many state-of-the-art models are based on attention mechanisms, especially in NLP. As this technique is very powerful, we decided to investigate its application in solving a collaborative filtering problem. In this paper, we propose a standard framework for developing a recommender system engine based on transformer architecture. We could not reproduce current state-of-the-art results on MovieLens datasets, but in our implementation attention based model achieves competitive scores on MovieLens 1M and MovieLens 10M datasets.