Classificador de fake news utilizando um modelo de aprendizado de máquina com técnicas de processamento de linguagem natural

Abstract

Fake news has an important social impact currently. Detect fake news has been an objective of several researches and interest in tools to categorize it has grown. Machine learning methods havs been gaining emphasis. Therefore, this study aimed to implemente the Passive Agressive Classifer classification technique together with natural language processing to classify news as false or true. For this purpose, four datasets were used, containing mandatorily title, text and label. The training was divided into 70% training and 30% test, then they were treated with TfidfVectorizer and submitted to the PassiveAgressiveClassifier to obtain the results. The metrics used were accuracy, precision and recall. All databases obtained rates above 78%, being considered excellent. Among the bases chosen, the FK3 dataset had the best performance, presenting a rate of 99%. The results obtained in this work show the potential of the classifier and the natural language processing as an alternative to the detection of fake news.