What do you want to want? Simulating User-Directed Preference Change in Recommender Systems
Keywords
Loading...
Authors
Issue Date
2022-06-25
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
In this thesis we reviewed the problem of recommender system-induced
preference change and presented user-directed preference change as a solution.
We argue that a reliance on behavior to infer user preferences can lead
to gaps and problems in the kind of preference change they tend to invoke in
users. User-directed preference change is a bottom-up approach which better
supports users to be in-the-loop of their own preference change. In order
to test the system-wide effects of user-directed preference change, we performed
a simulation study to model the effect of a novel “meta-preference”
mechanism by which users can provide corrective feedback to better align
recommended content with their preference change goals. Using the TRECS
simulation tool, we formalized user meta-preferences and present a
simple mechanism by which they were incorporated into the predicted user
preferences of simulated content-based recommender systems. In measuring
the degree and alignment of these changed preferences, we confirmed
that our mechanism was effective in causing user-directed preference change
within the simulation. We noted a trade-off with accuracy and found strong
homogenizing effects on recommendations and preferences throughout the
simulation. Though we believe that user-directed preference change is an
important consideration, we do not consider it to be a complete solution to
the preference change problem given its myopic focus on individual satisfaction.
Rather, this thesis explores one of many possible mechanisms which
are important in the development of technology which can be symbiotically
beneficial for individuals and society.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen