Detecting cyberbullying using latent semantic indexing

Abstract

Cyberbullying has proven consequential to youth Internet users and previous methods relied heavily on the use of man-ually developed dictionaries. This project describes prelimi-nary results for a system that uses Latent Semantic Indexing (LSI) for the detection of cyberbullying in a labeled collec-tion of posts from Formspring.me. After preprocessing to account for variations in spelling and use of emoticons, a search system was developed. Our system significantly out-performs the baseline with a very simple query and is not dependent on a dictionary of bullying terms.