5-Ions: An Automatic Method for Predicting Entity Performance

• Vanderson Sampaio UFSC https://orcid.org/0000-0002-0509-4638
• Renato Fileto UFSC
• Douglas Macedo UFSC

DOI: https://doi.org/10.5753/sbsi.2020.13759

Resumo

Textual documents on the Web (e.g., news, comments) may impact public opinion towards certain entities (e.g., currencies, people, institutions, products). Several indicators can be extracted from these documents, correlated to real measures of entity performance (e.g., currency quotation, company stock valuation) and used to explain or estimate real performance of some entities. This paper proposes an automatic method, named 5-Ions, which employes state-of-the-art off-the-shelf natural language processing tools to find relevant entity mentions along with their associated sentiment in text, and use them to calculate consolidated performance indicators for entities that are related in a semantic network (e.g. entities related to a particular currency). The results of the experiments with text documents about the Brazilian economy show that these consolidated metrics are better correlated with the real quotation of the Brazilian currency and allow better predictions of currency fluctuations than measures calculated for single entities.

Palavras-chave: Entity performance correlation, entity performance prediction, semantic relatedness