Detection of Offensive Tweets: A Comparative Study


Niyati Aggrawal

With the growing popularity, Twitter has become a major platform for posting views via tweets. Tweets contain useful, relevant and offensive content as well. More than a decade of research has resulted in numerous techniques and models to detect offensive content. However, little is known about lexically offensive and contextual offensive content. In this research paper, lexical offensive contents have been identified using two techniques- Rule-Based Naive Bayes (RNB) and a collaborative model of LDA with Naïve Bayes (LDANB). LDANB provides better results as compared to RNB for lexical offensive tweet detection. Further, contextually offensive contents are detected using newly devised Adjective Based approach. Contextual offensive content results prove to be better with Adjective based approach than Cosine similarity based results. To validate results of applied offensive tweet detection techniques three performance metricsprecision, Accuracy and recall are used.


Share this article

Get the App