Sparse Feature Factorization for Recommender Systems with Knowledge Graphs

RecSys 2021 Sparse Feature Factorization for Recommender Systems with Knowledge Graphs

Authors: Vito Walter Anelli, Polytechnic University of Bari | Tommaso Di Noia, Polytechnic University of Bari | Eugenio Di Sciascio, Politecnico di Bari | Antonio Ferrara, Politecnico di Bari | Alberto Carlo Maria Mancino, Politecnico di Bari

Abstract: Deep Learning and factorization-based collaborative filtering recommendation models have undoubtedly dominated the scene of recommender systems in recent years. However, although their outstanding performance, these methods require a computational effort proportional to the size of the embeddings.
This paper addresses this problem by presenting a sparse factorization approach, KGFlex, that grants an even greater degree of expressiveness while reducing the computational cost.
To achieve this result KGFlex analyzes the historical data to understand the dimensions the user decisions depend on (e.g., movie direction, musical genre, nationality of book writer).
KGFlex represents each item feature as an embedding and models user-item interactions as a factorized entropy-driven combination of the item features relevant to the user. Only these features are involved in the training, thus reducing the computational effort.
Feature embeddings are updated only by knowledgeable users that base their decisions on them.
In other words, the user-item prediction is mediated by the user's personal view that considers only relevant features.
An extensive experimental evaluation shows the approach's effectiveness, considering the recommendation results' accuracy, diversity, and induced bias.

DOI: https://doi.org/10.1145/3460231.3474243