What is it about?
When hit by the sudden outbreak of COVID-19, a large number of users are likely to panic and imitate others buying a lot of goods that they do not need. Users' unconscious mimicry often does not reflect their real preferences. This large-scale and sudden synergic shift in user behavior will heavily impair recommender systems' performance. This paper evaluates how collaborative filtering recommenders behave when facing such changes and also identifies two factors of herd behavior.
Featured Image
Photo by Heather Morse on Unsplash
Why is it important?
We introduce a new concept drift in user behavior due to the pandemic-like event. This helps recommender systems identify herd effects among users and generate better recommendations accordingly. Our evaluation shows: a) an extreme popularity bias in collaborative filtering models in the context of population-scale concept drift. Given that, the accuracy optimizing strategy often fails to generate novel and diverse recommendations. b) The growth of self-experience is an important factor to alleviate the herd effects.
Perspectives
COVID-19 has brought a profound impact on our life. I hope readers can make wise decisions when facing such extreme outlier events. Also, I hope this paper broadens researchers' thoughts on user modeling, user preference dynamics, and adaptive recommendations. User behavior may not always be rational and reflect their real preference. This is a key but often overlooked hypothesis.
Chenglong Ma
RMIT University
Read the Original
This page is a summary of: Evaluation of Herd Behavior Caused by Population-scale Concept Drift in Collaborative Filtering, July 2022, ACM (Association for Computing Machinery),
DOI: 10.1145/3477495.3531792.
You can read the full text:
Contributors
The following have contributed to this page
