An empirical analysis of how users with different genders are not equally affected by the recommendations

2022 International Conference on Engineering Technologies (ICENTE22), Konya, Türkiye, 17 - 20 Kasım 2022, ss.1-5

Yayın Türü: Bildiri / Tam Metin Bildiri
Basıldığı Şehir: Konya
Basıldığı Ülke: Türkiye
Sayfa Sayıları: ss.1-5
Sivas Cumhuriyet Üniversitesi Adresli: Evet

Özet

One of the main concerns related to personalized recommendations in recent years is how fair the provided referrals are for individuals differentiating in terms of particular features. In this study, we consider a critical protected attribute related to individuals, i.e., gender, and aim to analyze how the most prominent recommenders show varying performance for users of different genders. The experimental studies conducted on a real-world benchmark dataset and selecting eight state-of-the-art algorithms from different families demonstrate that males considerably receive more accurate recommendations than females. However, the obtained results also show that the recommendations produced for women are usually more qualified in diversity and novelty aspects when compared to those generated for men. These findings denote the unfair nature of the recommenders across users of different genders, and such disparities can occur differently for a specific aspect of the recommendations.