Hybrid Semantics-Aware Recommendations Exploiting Knowledge Graph Embeddings.

Graph-based recommendation methods represent an established research line in the area of recommender systems. Basically, these approaches provide users with personalized suggestions by modeling a bipartite graph that connects the users to the items they like and exploit such connections to identify items that are interesting for the target user. In this work we propose a hybrid semantics-aware recommendation method that aims to improve classical graph-based approaches in a twofold way: (i) we extend and enhance the representation by modeling a tripartite graph, that also includes descriptive properties of the items in the form of DBPEDIA entities. (ii) we run graph embedding techniques over the resulting graph, in order to obtain a vector-space representation of the items to be recommended. Given such a representation, we use the resulting embeddings to cast the recommendation problem to a classification one. In particular, we learn a classification model by exploiting positive and negative embeddings (the items the user liked and those she did not like, respectively), and we use such a model to classify new items as interesting or not interesting for the target user. In the experimental evaluation we evaluated the effectiveness of our method on varying of different graph embedding techniques and on several topologies of the graph. Results show that the embeddings learnt by combining collaborative data points with the information gathered from DBPEDIA led to the best results and also beat several state-of-the-art techniques.