Diversified Recommendation Generation Using Graph Convolution Neural Network

Abstract

Many methods have been proposed for recommendation generation using graph neural networks (GNNs). The advantage of using GNN in a recommendation system is learning the structural information of the user and item and their interaction more efficiently than traditional learning techniques. Most of the proposed models for recommendation generation are concerned only about their accuracy enhancement. Besides accuracy, novelty, diversity, and serendipity in the recommendation are often desirable for a better user experience in a real-world application. Earlier diversity in the recommendation system is achieved using the re-ranking algorithms. These approaches often compromise with accuracy to include diversity in the recommendation. Here, we proposed a methodology for diversity inclusion in the recommendation system using the GNN. We proposed a method based on Cluster-GCN for diversification of the recommendation. In our proposed method, we cluster users’ nodes based on their dissimilarity, and further, their subgraph is used for their neighborhood-based representation learning using graph convolution neural network (GCN). The novelty of the work is the clustering for the user’s pre-trained diversity enhancement in the recommendation generation. The proposed diversified cluster graph convolution neural network (Div-ClusGCN) model is trained for diversified recommendation generation. We achieved around 7% more diverse recommendations from the other state-of-the-art models.