Predictive Analysis on the Price of Dogecoin Using Tweet Sentiments and Volume of Tweets with LSTM

Cryptocurrencies are a type of digital currency that have been gathering attention as they are becoming more accepted as an investment medium and societies have become more receptive to cashless payments around the world. Previous studies have predominantly concentrated on the mainstream cryptocurrencies like Bitcoin and Ethereum. In contrast, this research focused on the correlation between meme coins, specifically Dogecoin, and tweets related to Dogecoin. The researchers have used a tweet dataset related to Dogecoin from Kaggle. VADER was used to get the sentiment scoring of the tweets. The input features used were tweet sentiments, tweet volume, and lag values at 1,3,8, and 12 hours. A total of 12 LSTM models using different combinations of the input features were developed utilizing the Bayesian Optimization algorithm. Half of the models were optimized for R-squared, and the other half were optimized for mean squared error (MSE). The models were evaluated using root mean squared error, mean directional accuracy, mean absolute error, R-squared, and percentage of duplicate values. The input features of the best-performing models had a combination of tweet sentiments, tweet volume, and all 4 lag values. The one optimized for R-squared is slightly better than the one optimized for MSE. However, the best-performing models still fell short since their R-squared was a negative value at −6.60364 and −6.21998 for those optimized for MSE and R-squared, respectively. This outcome may be attributed to the lack of available data since the tweet dataset used spans only approximately 4 months.