User Profile Minimization Investigating the Potential of Data Augmentation on Recommender System Performance
Keywords
Loading...
Authors
Issue Date
2022-06-30
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Recommender systems have quickly become a part of daily life, whether we
notice their presence or not. These systems continue to collect an increasing
amount of user data. This thesis argues that, not only does this practice
clash with the data minimization principle of the European General Data
Protection Regulation (GDPR), but also that there exists a way to achieve
the desired recommendation performance other than the endless collection
of more user data.
This work focuses on the potential of augmented data in recommender
systems by testing its effects on the recommendation performance of a stateof-
the-art rating prediction algorithm. We start by reproducing previous
work on the susceptibility of this algorithm to data scarcity, then improve
its current data minimization strategy, and finally test how augmented data
affects the recommendation performance. Our results indicate that both the
choice of data minimization strategy as well as the usage of augmented data
in the training set can have a significantly positive effect on recommendation
performance of the algorithm.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen