IDENTIFY AND CLASSIFY TOXIC ONLINE COMMENTS

Abstract

The main purpose ofthe project is to build an analysis module that can analyse toxic sentiments then classify it into different categories. Toxic classifier automatically categorizes comments or tweets. This model can categorize toxicity into 6 categories that are Toxic, Obscene, Insult, Identity hate, Threat, Severe Toxic. A large number of tweets in form of super vised dataset is used for training and testing of classifiers, the model is capable of doing multi label classification. The model is working in 4 stages, in first stage pre-processing ofdata is being done by cleaning unnecessary words, links, emoticons, punctuation and tagging of parts of speech (POS) after that words is being lemmatized. In second stage making of feature vector is being done from which important words and features are extracted from data set, in third stage classifier is being trained which will later then classify the tweets in to toxic category, in last stage evaluation of classifier is being done which will show the accuracy of classifier that how well classifier performed. Recommendations for development in future and conclusion is included in the report