Research on GRU and CNN Combined Recommendation Model with Self-Attention Mechanism

Abstract

In recent years, recommendation models have been widely used in various fields, but the existing recommendation models mostly focus on user historical behavior learning or the analysis of the internal relationship of items, and it is difficult to take into account the effective extraction of deep-level user features and item features. In view of this, this paper studies and proposes a GRU and CNN combined recommendation model with Self-Attention mechanism. The model uses GRU and Self-Attention mechanisms to extract user features, combines with CNN to capture the local relevant features of the item. Then through the full connection calculation of each feature, the prediction rating is obtained and the recommendation is generated. This paper uses MovieLens 1M data set and Amazon Digital Music data set for experiments. The results show that compared with other deep learning-based recommendation models and traditional recommendation models, the model proposed in this paper has achieved better results in both MSE and MAE indicators.