DEEP GRAPH EMBEDDINGS IN RECOMMENDER SYSTEMS: A SURVEY

Abstract

In recent decades, the increasing quantity of products and services offered the possibility of collecting significant amounts of data, which require new techniques to sort it. Rather than manually filter these large quantities of information, it provides changes over time. Recommender systems help suggest articles to read on smartphones, posts to watch on Facebook, books to buy on Amazon. Their goal is to personalize data to increase the use of a service or to enable more sales. Their influence is not just a technical and commercial necessity. They have also become a part of the evolution of the human mass and his\her ideas. Because a book or a newspaper is not just a commercial object, many current recommendation techniques are challenged by information overload, which poses many issues like high cost, slow processing of data, and low time complexity. For this reason, many researchers in this field use graph embeddings algorithms in the recommendation area, as the last few years have also seen the success of these algorithms, especially the ones based on deep learning. Current recommender systems based on these algorithms have shown that they can obtain exciting results and improve the quality of recommendations offered to users. In this survey, we present an overview of recommender systems and graph embeddings based on deep learning. Then we provide a literature review of recent recommendations works based on deep graph embeddings to make a pragmatic analysis and showings common limitations.