Prevendo a flutuação de preço do Ether através da análise de sentimento do twitter

Abstract

Cryptocurrencies, an innovative technology that has gained a lot of traction and visibility, are still surrounded by mistrust, but as the social and economic impact of cryptocurrencies continues to grow rapidly, the number of articles on the web, television reports, and comments on social networks are also growing. With the popularization and growth of the social network Twitter, the opinions and sentiments expressed in the form of of tweets represent a large amount of data that can be studied. In this paper, we analyze tweets with the aim of establishing whether their sentiment is a useful tool for predicting price fluctuations in a cryptocurrency called ETH. The dataset consists of 34,558 tweets collected using the TWINT library, which consists of tweets containing the "ETH"sign. The collected tweets are converted into scores that represent positivity/negativity of the tweet, this is done by an analysis tool called VADER. The scores of the tweets are then summed up to represent a collective sentiment per hour, acting as a predictor variable for the price of cryptocurrency price. We use in our experiments supervised learning techniques training a Random Forest Classifier and for comparison purposes, we also trained a model using Multiple Linear Regression.