Improving Trust in Recommender Systems through Context Clues

The bachelor thesis was written by Tobias Sippl.

Abstract

Providing serendipitous recommendations in a recommender system can lead to various problems. Not only is it hard to find an item which is both novel and relevant to the user, but the user should also engage with the item instead of discarding it as irrelevant. Serendipitous items tend to appear as unusual or unfitting at first, but turn out to be a pleasent surprise once investigated. So it is not only necessary to find a serendipitous item, but also to present it in a way which increases the likelihood that a user engages with it. An experiment was conducted in which participants interacted with a novel book recommender system and were assigned to either an experimental or a base group. The experimental group interacted with a visually enhanced version which was designed to trigger various heuristics aiming to increase user engagement and trust, while the base group used a comparatively bland interface. Trust and serendipity were then measured for each item and the recommender system itself. It was found that while it did not increase trust in the recommender system, serendipity was rated higher for books the participants did not know.

Supervision

Supervisor: Irina Nalis-Neuner, Co-Supervisor: Julia Neidhardt

Thesis

Bachelor_Thesis.pdf