Sentiment analysis based direction prediction in bitcoin using deep learning algorithms and word embedding models

Abstract

Sentiment analysis is a considerable research field to investigate enormous quantity of knowledge and specify user opinions on many subjects and is resumed as the extraction of ideas from the textual data. Like sentiment analysis, Bitcoin which is a digital cryptocurrency also attracts the researchers considerably in the domain of cryptography, computer science, and economics. The objective of this work is to forecast the direction of Bitcoin price by analyzing user opinions in social media such as Twitter. To our knowledge, this is the very first attempt which estimates the direction of Bitcoin price fluctuations by using word embedding models in addition to deep learning techniques in the state-of-the-art studies. For the purpose of estimating the direction of Bitcoin, recurrent neural networks (RNNs), long-short term memory networks (LSTMs), and convolutional neural networks (CNNs) are used as deep learning architectures and Word2Vec, GloVe, and FastText are employed as word embedding models in the experiments. In order to demonstrate the contribution of our work, experiments are carried out on English Twitter dataset. Experiment results show that the usage of FastText model as a word embedding model outperforms other models with 89.13% accuracy value to estimate the direction of Bitcoin price.