MediaEval 2014 Crowdsourcing Task: Crowdsorting multimedia comments
Martha Larson, Karthik Yadati, Mohammad Soleymani, Pavala Shakthi Nathan Chandrasekaran Ayyanathan · 2014
The MediaEval 2014 Crowdsorting task addresses the challenging of classifying timed-comments that users have associated with specific time-points in music tracks on a large social music sharing platform. Input to the classifier is a set of labels generated by human computation (i.e., input from the crowd). Alternately, the classifier can exploit labels generated by both human and automatic computation (i.e., text and/or audio signal analysis). The job of the classifier is to sort timed-comments that have been contributed by users listening to a particular song in order to isolate the comments that are the most useful (whereby 'useful' takes on different definitions dependent on the final application scenario in which the sorted comments are to be used).