In this contemporary world, people use social media tremendously to communicate and have fun with their friends and family. Over the past few years, the ethical use of social media has been exploited and people have started to post comments or tweets expressing their hatred towards other communities. While some people claim them to be sarcastic comments, others feel they are offensive and depict hate speech. This has become a major issue as these are dependent on an individual’s perspective. We have constructed a dataset that contains sentences from one of the four classes-” Hate Speech”, “Offensive”, “Sarcasm”, “Normal”. The dataset has been evaluated using various state-of-the-art models like Bag of Words, TF-IDF, BERT and ERNIE. The best accuracy was achieved by BERT accounting to around 80%.