an online evaluation of explicit feedback mechanisms for recommender systems
DESCRIPTION
Poster about an online feedback experiment as presented during the WEBIST 2011 conference in Noordwijkerhout (The Netherlands), May 7, 2011 by Simon Dooms.TRANSCRIPT
An Online Evaluation of Explicit Feedback Mechanisms for Recommender Systems
Simon Dooms [email protected]
Toon De Pessemier [email protected]
WiCa, Wireless & Cable, www.wica.intec.ugent.beGaston Crommenlaan 8 box 201, 9050 Ghent, BelgiumGhent University, Department of Information Technology
Introduction1
Recommender algorithms need feedback data
Feedback systems collect this input from the user
Can be explicit (ratings) or implicit (logs in the background)
Ratings can be dynamic (JavaScript) or static (HTML forms)
The Experiment2
On the pages of a popular Belgian cultural website we
randomly put one of the four most used rating mechanisms
(dynamic thumbs, dynamic stars, static thumbs and
static stars) and logged the interaction of the users. Next to
this explicit feedback, we tracked also implicit feedback in
the form of click and browse behavior.
The Results3
6 months of data
8 100 explicit ratings
200 000 implicit ratings
800 000 pageviews
5-Star
(dynamic)
Thumbs
(static)
Thumbs
(dynamic)
5-Star
(static)
1330 1694 2101 2976
16 % 21 % 26 % 37%
Conclusion4
Users don’t prefer dynamic over static systems
Most (explicit) ratings are provided anonymously
The 5-stars systems are used as thumbs systems(The extreme star values i.e. 1 and 5 are most commonly used)
Explicit feedback is hard to get, implicit feedback is easy
The static 5-stars rating system collected the most feedback
Future Work5
Optimize feedback collectionHow can users be encouraged to rate?How to combine explicit and implicit feedback?
What feedback system collects most valuable feedback?Quantify the relevancy of explicit versus implicit feedback
Extract events dataset for future research
Rating Values Distribution for Thumbs and Stars
Explicit Ratings per day
Implicit Ratings per day
Cumulative number of Explicit RatingsExplicit Feedback Anonymous Vs User