Using graph embeddings to improve the performance of embedding based recommender Systems

This paper discusses the use of node2vec graph embeddings to improve the performance of wide and deep recommenders by substituting the embedding layer with graph embeddings to leverage its representational power without major investments in migrating from current recommender systems. First, recommender systems importance in modern day platforms is discussed. Then, the magnitude of investment needed to deploy a recommender system into a production environment is discussed and finally, wide and deep recommenders and their importance in industry is outlined along with our experiments and results in improving its performance using graph embeddings,