Transferência de conhecimento para detecção automática de fake news com aprendizagem profunda

Abstract

The use of social networks for communication and non-measured exchange of information results in the spread of fake news. Because it has the capacity to influence and manipulate the course of reality, the sharing of fake news needs to be restrained. Considering the massive volume of information circulating on social networks and its speed of propagation, the review by trained professionals, to verify the occurrence of unreliable content, is not viable. In this sense, computational tools capable of automatically detecting the occurrence of fake news are necessary. In this work, we deal with the problem of classifying information in text format as false or true, considering only the input. So, as an approach, using Machine Learning classification algorithms. In particular, we focus on approaches based on Deep Learning, for the ability to abstract complex concepts and recognize semantic patterns. In this type of approach, it relies on labeled data sets, with sufficient volume to allow the recognition of patterns related to the problem. As there are no large data sets available in the Portuguese language. We use Knowledge Transfer techniques to improve the performance of the algorithms. As for the methodology, it is an exploratory and quantitative research, in which we investigate the performance, through computational experiments, of state-of-the-art methods to solve the problem. Regarding the data, the base Fake.br will be used.