Dual gated attentive-autoencoder for content-aware Recommendation

Abstract In recent years, content-aware recommendation of personalized products using automatic encoders has become the mainstream technology in the field of personalized recommendation systems. But the technology still faces the following main problems: 1.The automatic encoder structure ignores the inherent duality of the model and separates the training encoder from the decoder.2.The automatic encoder structure makes use of implicit feedback data and the difficulty of combining heterogeneous data.3.The traditional content-aware recommendation based on the auto-encoder ignores the neighboring merchandise information of the merchandise. In order to solve these problems, this paper proposes a dual gated attentive-autoencoder based on the door attention mechanism. The model is compared to the current optimal outcome model on the three recommended system public datasets. The experimental results show that the recall rate and the normalized damage cumulative gain of the auto-encoder content perception recommendation based on the dual-gate attention mechanism are better than the current optimal result model.