Desarrollo de un servicio web para la clasificación de noticias falsas, aplicando técnicas de aprendizaje automático

Abstract

The objective of this Curricular Integration Project was the development of a web service for the classification of false notifications in Spanish, integrating a machine learning model for the classification. We utilized the analytical and deductive methods in the development in order to determine the theoretical foundations, and we applied the SCRUM methodology for the web service, where we established the requirements, people and roles involved in the project. Seven technical stories and two user stories were defined, too and a sprint backlog was created to plan the period for the completion of the stories. On the other hand, we utilized the CRISP-DM methodology, covering tasks of data set collection, preprocessing, selection of the techniques to be implemented. Finally, we evaluated and selected the best performing technique for integration into the web service. Both the web service and the implementation of the model were carried out using the Python programming language, with the use of machine learning libraries. We based ourselves on the ISO/IEC 25010 standard to evaluate the performance efficiency of the web service, taking response time as a metric to evaluate the temporal behavior. The analysis determined that the web service has an average response time of 1.08 seconds. On the other hand, the F1 metric was utilized for the evaluation of the performance of the machine learning models, obtaining that the LSTM architecture has the best performance with a score of 0.746. Finally, we recommend that for the training of machine learning models, the size of the data set be taken into account, selecting data sets balanced in the number of existing data per class, so that the classification results will not be affected.