A Similarity-based Chinese Restaurant Process for Social Event Detection.

In this paper, we present our approach for the Social Event Detection task of Medieval 2013 [2]. The goal of the task was to group similar multimedia items into event clusters, based on their metadata (e.g. title, description, tags). Since the number of the event clusters in the test set was not known in advance, we formulated a non-parametric algorithm which resembles the Dirchlet process clustering. More specically, we developed a similarity-based version of the Chinese Restaurant Process (CRP) which exploits the similarities among the media items. Our approach achieved a F1 score of 0.2364.