Detox: NLP Based Classification And Euphemistic Text Substitution For Toxic Comments
Somil Jain, Garima Kaushik, Pulin Prabhu, Anand Godbole · 2021
The challenges of eliminating inadvertent hate online have grown more and more observable. With rapid increase in access to technology worldwide, the number of people present on the internet has increased exponentially. The rise in hateful and toxic content on the internet is many fold. Present solutions are largely based on censorship and removal of such content and can be the cause of mental disturbance to the concerned auditor. Biases of manual moderation can thwart the efforts taken to prevent the presence of hate speech on online platforms. However, posts that convey important meanings also fall prey to such systems, which should be avoided. Such systems fail to educate their authors what should have been done differently. Additionally, a system that combines detection and substitution algorithms is largely absent. A euphemistic substitution approach could prove to be more effective. In this project we have developed a classifier using Natural Language Processing and Machine Learning to detect toxic texts and provide euphemisms to erudite the user of words that can replace the toxicity present in the original texts. The classifier informs the online platform about the toxicity of the texts so they restrict such content to available on its platform. The euphemisms are aimed to make the user aware of the toxicity in the text and suggest replacements, which, if used, can make the text inoffensive. We aim to achieve self realisation on the user's part so we can target the issue at its source.