Identificação de escritor por transformada SIFT e SVM-Linear na língua portuguesa

Abstract

This work investigates the feasibility of the SIFT Transform (Scale-Invariant Feature Transform) and the SVM-Linear algorithm (Support Vector Machine-Linear) for identifying authors in texts in portuguese. The research contributes to the area of forensic authorship, information security and style analysis, seeking to provide a robust and efficient method for identifying authors in different contexts. The proposed methodology uses the SIFT Transform together with the Bag-of-Words concept to extract relevant characteristics from texts and SVM-Linear as a classifier to identify authors. The research was carried out on a BFL (Brazilian Forensic Letter Database) dataset, composed of forensic documents in Portuguese, to evaluate the effectiveness of the methodology. The proposed methodology obtained promising results, with an average accuracy rate of up to 98.73% in identifying authors.