Improving LSTM price prediction of Bitcoin with sentiment analysis of Twitter post

Abstract

The Covid-19 pandemic has seen a significant increase in retail investors across all age groups. Out of all the asset classes, cryptocurrencies like Bitcoin gained a lot of attention and surged by 300% in 2020 due to speculation in the financial market. Unlike traditional asset classes that offer various channels for newcomers to learn (books, news, courses etc.), crypto investors are highly dependent on social media for information and knowledge. These social media include YouTube, Twitter and Reddit, with some communities using Facebook and Discord groups to interact and exchange information. This information provides the basis for sentiment analysis to predict the prices of Bitcoin. This paper aims to make use of sentiment analysis via the SenticNet APIs and investigate if adding sentiment scores as a feature will improve the accuracy of LSTM price prediction models for Bitcoins.