Analysis of Economic Indicators Through News and Twitter Using Text Mining, Machine Learning and Multiagent Systems

Abstract

This research proposes a new analysis approach for economic phenomena, including data from news and social networks as external information to predict commodity values (LBMA GOLD and Brent oil) and the USD/COP currency, clas-sify the sources of information and model multi-agent systems. Information was collected from 166 news sources through RSS and Twitter for 8 months (from October 2020 to May 2021). Linear regressions and assembly machine learning techniques such as XGBoost and Random Forest are used to predict daily changes. The analysis is complemented with the construction of a socio-inspired multi-agent system that evolves using external information, at the end presents patterns typical of complex systems.