Predicting the direction of the oil price trend using sentiment analysis

In this paper, we present a natural text processing model for predicting the price of exchange-traded products based on machine learning and general statistics. With the help of our model, we are forecasting the trend of one of the most important energy, the oil prices daily basis from tweets. The backbone of our model consists of transformer-based techniques in a recurrent neural network framework with corresponding hyperparameter optimization. The essence of our solution is to use the sentiment characteristics and vocabulary that can be extracted from the tweeter news. We have found that some of the news sources have better correlated to the oil price change which observation was used to refine the training corpus. Furthermore, we have applied noise filtering by removing the insignificant words from the textual information. In this way, we have generated a data source from which the sentiment values showed a high-precision correlation of 84.08% with the true direction of the oil price.