Sentiment Identification on Tweets to Forecast Cryptocurrency’s Volatility

Abstract

Cryptocurrencies have had a huge presence on social media since their creation. In current days, the constant increase of the mass of data produced by this environment has attracted several researchers to try to identify patterns with the potential to allow identification of the volatility in the crypto market before it happens. This approach involves the concept of the wisdom of the crowds, a popular theory in the economy field that in the current days may have the perfect tools to prove itself true. This scenario creates an opportunity to unite two new technologies, social media, and cryptocurrencies to the newest Natural Language Processing (NLP) tools, and produces a study in a rich and unexplored field. Executing a detailed sentiment analysis, this study intents to analyze the forecast of the volatility of cryptocurrencies through the detection and evaluation of several categories of sentiments on messages on twitter when it is associated with a specific cryptocurrency. To achieve this, an NLP model was trained with the GoEmotions dataset to identify and categorize emotions, and results were used to calculate the forecast of the cryptocurrency. Index terms cryptocurrency, social media, Natural Language Processing (NLP), GoEmotions.