CFN-AI: Um classificador de Notícias falsas baseado em Inteligência Artificial

Abstract

The democratic and free profile turned the Internet into a common place for propagation information, making this a commonly used medium for sharing of false news, the so-called fake news. These news have a writing method differentiated from journalistic content focused only on the fact itself. Thus, this research seeks to affirm whether it is possible to minimize the impact of false news through a textual analysis, using Artificial Intelligence (AI), analyzing already implemented AI algorithms for ranking fake news. Search also develop a model that classifies news as true or false with usage of Artificial Intelligence, comparing the developed model with other works correlated. The research for the composition of the related works took place through a systematic review of the literature. The model was developed in Python, on Google Collaboration After importing the dataset, the pre-processing of the data was performed with tokenization using tensorflow datasets, model building using the Keras library, analysis and definition of the hyperparameters, model training, evaluation using cross-validation and experimental ranking tests with news not in the dataset used for training. The model’s accuracy reached a percentage of 95.76%, but when submitted to other news outside the database used for training, the value was 80%. In the results, it is possible to verify that the model proposed through Networks Convolutional Neurals showed good preliminary results when compared to other searches for news classification. But there are some to be observed nonconformities when algorithms are subjected to real situations after training.