Unintended biases in machine learning (ML) models are among the major concerns that must be addressed to maintain public trust in ML. In this paper, we address process fairness of ML models that consists in reducing the dependence of models on sensitive features, without compromising their performance. We revisit the framework FixOut that is inspired in the approach “fairness through unawareness” to build fairer models. We introduce several improvements such as automating the choice of FixOut's parameters. Also, FixOut was originally proposed to improve fairness of ML models on tabular data. We also demonstrate the feasibility of FixOut's workflow for models on textual data. We present several experimental results that illustrate the fact that FixOut improves process fairness on different classification settings.

A requirement for a great user experience is to meet the exact needs for the usage of a recommender system. Such systems need user's historical preferences to reasonably perform, which might not be the case for a cold-start user. This paper presents VP-Rec, a hybrid image recommender system that addresses the new user cold-start problem. VP-Rec combines user visual perception and pairwise preferences as source of information to perform recommendations. First, we infer pairwise preferences from users ratings. Next, we build visual perception networks linking users according to their visual attention similarities. From these two inferred structures, we build consensual prediction models, so that when a new user enters the system, we capture his visual attention and choose the best model that fits him. The system has been tested on two image datasets, getting important improvements in terms of ranking quality (nDCG) when applied to new user cold-start scenario against state-of-art recommender systems.

In this article we propose CPrefMiner, a mining technique for learning a Bayesian Preference Network (BPN) from a given sample of user choices. In our approach, user preferences are not static and may vary according to a multitude of user contexts. So, we name them Contextual Preferences. Contextual Preferences can be naturally expressed by a BPN. The method has been evaluated in a series of experiments executed on synthetic and real-world datasets and proved to be efficient to discover user contextual preferences.