Overview of the International Sexual Predator Identification Competition at PAN-2012.

This contribution presents the evaluation methodology for the identi- fication of potential "sexual predators" in online conversations as part of PAN 2012, and provides details of the realized collection and analyse the submissions of the participants.

Abstract

Abstract This contribution presents the evaluation methodology for the identi-fication of potential “sexual predators ” in online conversations as part of PAN 2012. We provide details of the realized collection and analyse the submissions of the participants, who had to solve two problems: identify the predators among all the users in the different conversations and identify the part (the lines) of the predator conversations which are the most distinctive of the predator bad be-haviour. The methods proposed by the 16 teams participating in the contest made possible the recognition of common pattern for predator identification (e.g. no preprocessing of the conversations, lexical and behavioral analysis, blacklisting of predator terms) as well as possible extension to existing systems (e.g. victim-predator distinction, pre-filtering of not relevant conversations). 1