ABSTRACT

Twitter has more than 313 million users, is considered a subjective social network of content generation on the Internet, not only used to disclose personal information, but shares opinions and information about events and events in general, being a great source of research for the discovery of knowledge among relevant data. In this context, the present work presents a study on the implementation (application) of the Recurrent Neural Network LSTM for the analysis and classification of tweets that are related to crime and not crime. The set of data that is the object of the survey comes from an account of a newspaper in the city of Belém, containing 20000 records in a period of two months. The results obtained with the use of RNR LSTM proved to be quite satisfactory, reaching accuracy in some cases of 90% correctness. Based on the results of this research, we observed the effectiveness of this method in the analysis of feelings in comparison with other algorithms such as Nayve Bayes, Convolutional Neural Networks used in the literature. Because it is a network that uses a Long-Term Memory technique, it can adapt to large time intervals, such as Twitter texts.