Análise de Desempenho de Modelos de Embeddings Multilínguas na Classificação de Notícias Falsas

Abstract

The fake news classification introduces some challenges for Natural Language Processing (NLP). Other non-textual representations, such as text embedding, can influence the performance of those classifications. The goal of this work is to evaluate these performances on two corpus databases in the context of fake news classification. Five different word embedding algorithms were applied, and subsequently, the data were classified by four classification algorithms. Our study demonstrates satisfactory performance for the representations, with the Multilingual-E5-large algorithm standing out with superior performance compared to the other tested models. This work may contribute to understanding the influence of multilingual word embedding algorithms on the performance of text classification algorithms. The experimental results can serve as a performance reference to guide the use of these models for text classification in similar studies.