Deepfair: Deep learning for improving fairness in recommender systems

Abstract

The lack of bias management in Recommender Systems leads to minority groups receiving unfair recommendations. Moreover, the trade-off between equity and precision makes it difficult to obtain recommendations that meet both criteria. Here we propose a Deep Learning based Collaborative Filtering algorithm that provides recommendations with an optimum balance between fairness and accuracy. Furthermore, in the recommendation stage, this balance does not require an initial knowledge of the users’ demographic information. The proposed architecture incorporates four abstraction levels: raw ratings and demographic information, minority indexes, accurate predictions, and fair recommendations. Last two levels use the classical Probabilistic Matrix Factorization (PMF) model to obtain users and items hidden factors, and a Multi-Layer Network (MLN) to combine those factors with a ‘fairness’ (ß) parameter. Several experiments have been conducted using two types of minority sets: gender and age. Experimental results show that it is possible to make fair recommendations without losing a significant proportion of accuracy.