In our work we report our models, their architectures and their respective performances in the Big MM Challenge MeToo. We showed how the ensemble model, which was a collection of 3 different deep learning models, and used two different sets of Embedding vocabularies performed better than the individual models. In the future, we would try to give the problem a more defined approach, given the bias in positive and negative classes for the sentiment classes, we can try to pre-process the dataset in a better way so as to accommodate all training examples, while making sure that the outputs from negative classes do not dominate over the positive outputs from the respective classes. Some other areas which could be improved include, data scraping from the attached urls to the tweets, which can be helpful to find the relevance and judge the authenticity and perform better language understanding.