Author(s): Miftahul Islam and Muhammad Mahfuz Hasan
Abstract: Scrambled words that are formed due to the disorganization of one or more internal letters of a word have been analyzed by psycholinguists to be discernible by most of those who know of the language it is used in. These scrambled words can be generated by unmoderated users for spreading hate speech, curse words and expletives with the specific intent of bypassing existing censor filters. This paper presents a proposal to bolster detection of curse words and hate speech from internally scrambled words along with additional user preferred censor words. Our model has been developed to make general users safer from the atrocities of manipulated hate speech and curse word usage. Further improvements have been made to the detection of user manipulated offensive expletives through the use of permutations and their proper censorship and also through the application of different filters in various services used both in online and offline text-based media.
Keywords: Natural Language, Online Interaction, Swearing, Cursing, Social Media, Scrambled Words