This year’s workshop focuses on log-based methods, for application in search systems and recommender systems, that tailor results for the current user. The logs can be used to uncover characteristics, patterns and preferences in users and cohorts of users. Logs can also be used to train the personalized search/recommender system. The goal is to improve results by showing different results to different users.
In one application, a search system might detect a user’s current interest in rock music, so reorder the results for the query ‘chicago’ to promote the band above the city. In another application, a news recommender system might use a contextual bandit approach to learn from click logs and personalize news article recommendation.
The workshop is a forum for discussion of current work and problems in the area. For this reason we solicit position papers:
Specific models for log-based personalization
Data mining methods for identifying and characterizing interests of users in logged data
Insights into personalization evidence and features derived from logs. For example via demographics, topical interests or memorizing specific past actions
Specific application(s) of log-based personalization. How useful is it? What works? What doesn't?
Does log-based personalization narrow options into a filter bubble?
Or, does log-based personalization broaden people's options like on Amazon
What machine learning problems arise in training log-based personalized models
What IR evaluation problems arise in testing log-based models
What datasets are available, or should be available, for studying log-based personalization
Submissions: | December 23, 2013 |
Notification: | January 10, 2013 |
Event: | February 28, 2014-February 28, 2014 |