Ontology based Approach for Reducing Toxicity and Maintaining Similarity

Many online social media platforms have particular community guidelines for comment sections. The platforms that maintain commentary sections in various posts, videos, and blogs need to adhere to these guidelines. These comment sections may have specific comments that fail to satisfy the rules and regulations to maintain societal norms of communication. These comments are classified as toxic comments. Google’s Perspective API defines toxic comments as comments that are rude, offensive, and likely to make someone leave the conversation. In this paper, we propose a Detoxification module for toxic comments - in- put to which will be a toxic comment. We will get a sentence detoxified, having a similar meaning to the original sentence as output. This module consists of an Ontology that we have built for toxic words. As per our knowledge, this Ontology is the first ontology built for toxic words. This Ontology consists of toxic words and their antonyms and synonyms in in- creasing order of their toxicity levels. We can traverse this Ontology and find the best-suited word with less toxicity and similar meaning. We are experimenting with the Toxic Comment Classification Challenge dataset from the Kaggle competition which consists of 1,59,571 comments out of which we filtered 15,294 toxic comments. Our approach is relatively straightforward and simple - and is effective in reducing toxicity in online comments.