Technical report on Automatic Identification of Paedophile

Accurate and up-to-date knowledge of keywords entered by users who search or provide paedophile content is a key resource for filtering purposes and for monitoring by law enforcement institutions. However, such keywords are often hidden and may change frequently, and our current knowledge about them relies on manual inspection and field expertise. We explore here the possibility to help in improving this situation by applying various keyword analysis methods. Using a large-scale real-world collection of paedophile and non-paedophile file names, we construct lists of keywords suspected to be used as paedophile keywords. We evaluate the relevance and interest of these lists by submitting them to experts, thus showing that automatic approaches are indeed of great interest for this task.