INGCF: An Improved Recommendation Algorithm Based on NGCF

Strengthening the representation and learning of user vector and item vector is the key of recommendation system. Neural Graph Collaborative Filtering (NGCF) has the problem of insufficient feature extraction of user vector and item vector. In addition, it uses the linear combination of the final embedding vectors of user and item to calculate the inner product, which is difficult to accurately obtain the user-item prediction score. In this article, we propose a graph neural network collaborative filtering algorithm based on NGCF, named INGCF. This new algorithm uses a 4-layer IndRNN layer and a feature extraction layer constructed by a self-attention mechanism to enhance the feature extraction capabilities of NGCF. And INGCF optimizes the user-item prediction score method to capture the complex structure of user-item interaction data to improve the accuracy of score calculations. We compare several indicators of INGCF, NGCF, PinSage and GCN by using same data sets and perform ablation experiments. The experimental results show that the INGCF algorithm has a better recommendation effect.