Analysis of social networks publications for stock market movement prediction: methodology and case study in a Spanish-speaking country

Predicting the stock market has been a problem that has caught the attention of the scientific community since any new information is quickly incorporated into the share price. Then, it has been sought to find new sources of information that can be useful for machine learning models and allow to predict with greater precision the movement of the action. One of these novel data sources has been the estimation of the mood of the population, approximating it by means of polarity analysis on Twitter. The scientific community has focused on the study of this relationship in the stock markets of the United States and China, but no work has been reported in Spanish-speaking markets. This paper presents a methodology for capturing, cleaning, and creating indexes derived from the polarity in Twitter messages, adapted to the Spanish language. For the calculation of indicators, more than 8 million tweets published in Colombia were analyzed during the period 8-2020 to 8-2021, their polarity was calculated using the set of words from the AFINN lexicon translated into Spanish. There were calculated 5 social indicators derived from the Twitter message's polarity. The Logistic Regression, Support Vector Machines, and Artificial Neural Networks models were trained, where the latter had the best performance, with an accuracy of 58%. Then this article opens the discussion of the applicability of this type of technique in the Spanish-speaking markets.