We design a bot trading technique based on machine learning on twitter sentiment analysis. We compare sentiment models like Vader, Naive Bayesian, and BERT to see which performs best on tweet sentiment analysis. We then use these tweets, their sentiment, and popularity of the user to assign a modified sentiment score. This modified score is one of several input features for several models that aim to calculate the best action of buy or selling a stock to maximize profit. In the end, we gain 5% advantage compared to base line model using an LSTM model.