Using NLP Methods for Cryptocurrencies Price Prediction

The paper considers the possibility of using Natural Language Processing methods as applied to tweets to search for interrelations and the possibility of predicting the price of cryptocurrencies. In view of the novelty of the cryptocurrency market, the issue of using tools from the classical financial market, especially the relatively new ones, is particularly acute. The work consists of three chapters covering the tasks: the study of existing predictive cryptocurrency indicators, as well as the use of sentiment analysis for predictions; overview of NLP methods that can be used; constructing a structure for processing tweets and searching for interrelations on historical data. In the course of the work, NLP methods were used to build the pipeline, as well as BERT neural network. Taken together, it was possible to get a feature from tweets to predict the price. On cryptocurrencies with a small capitalization (TRX, BNB), the result is better than in currencies with a large (BTC, XRP). The logical continuation of the work is to create a model for predicting prices based on the feature obtained in this work, as well as others.