Bias disparity in recommendation systems (Master thesis)

Τσίντζου, Βιργινία

Recommender systems have been applied successfully in a number of different domains, such as, entertainment, commerce, and employment. Their success lies in their ability to exploit the collective behavior of users in order to deliver highly targeted, personalized recommendations. Given that recommenders learn from user preferences, they incorporate different biases that users exhibit in the input data. More importantly, there are cases where recommenders may amplify such biases, leading to the phenomenon of bias disparity. Amplifying bias for different groups of users can lead to isolating sensitive groups or indirect discrimination. The goal of this thesis is to study bias disparity in recommender systems. To this end, we define metrics for bias and bias disparity for recommendation systems. Then, we consider variants of the K-Nearest Neighbors recommendation algorithms, and we perform a systematic analysis of their behavior using synthetic data. The goal is to understand the conditions under which those algorithms exhibit bias disparity, and the long-term effect of recommendations on data bias. We observe that even moderate amount of bias, and small biased groups can lead to significant bias amplification. Using the Movielens dataset, we also present cases of real data where bias is observed and confirm bias disparity on recommendations. To address the problem of bias disparity, two algorithms that post-process recommendations are considered. The algorithms re-rank the results of any recommendation algorithm in order to produce new sets of recommendations where bias disparity is eliminated. Each bias correcting algorithm aims at providing useful recommendations by targeting the utility of the user group or the least satisfied user in the group. We conclude that correcting bias in recommendations slows down the polarization of users in the long-term.
Institution and School/Department of submitter: Πανεπιστήμιο Ιωαννίνων. Πολυτεχνική Σχολή. Τμήμα Μηχανικών Ηλεκτρονικών Υπολογιστών και Πληροφορικής
Subject classification: Recommendation systems
Keywords: Recommendation systems
Appears in Collections:Διατριβές Μεταπτυχιακής Έρευνας (Masters)

Files in This Item:
File Description SizeFormat 
Μ.Ε. ΤΣΙΝΤΖΟΥ ΒΙΡΓΙΝΙΑ 2018.pdf415.28 kBAdobe PDFView/Open

 Please use this identifier to cite or link to this item:
  This item is a favorite for 0 people.

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.