Predicting Stock Prices Using Neural Models Based on Financial Textual Information

Pablo Alvarez, Andrés Morales, Rodrigo Seguel, John Atkinson

Producción científica: Contribución a una revistaArtículorevisión exhaustiva


The investment market has evolved with the integration of information technologies and the creation of new assets and strategies. Machine learning techniques have been used in finance to develop efficient applications. Furthermore, social media platforms have a significant impact on the market, with studies showing a strong correlation between stock prices and opinions on social media. Hence, machine learning techniques have been used to predict stock behavior based on the implicit sentiment in social media opinions. Accordingly, this article proposes a combined model for stock price prediction that uses natural language processing techniques and neural time-series prediction models based on long short-term memory. The approach combines stock price data from a week with related social media sentiments and uses natural language processing techniques based on pretrained transformers for the financial sector and predictors based on long short-term memory networks. Additionally, a convolutional neural network–based classifier is designed to analyze sentiments and determine a bullish or bearish factor based on polarity. The experiments with real stocks from the NYSE demonstrate that polarity prediction can improve stock price prediction by 26% compared to traditional predictive models.

Idioma originalInglés
Páginas (desde-hasta)74-87
Número de páginas14
PublicaciónJournal of Financial Data Science
EstadoPublicada - 1 mar. 2024
Publicado de forma externa


Profundice en los temas de investigación de 'Predicting Stock Prices Using Neural Models Based on Financial Textual Information'. En conjunto forman una huella única.

Citar esto