Detección del engaño en notas de opinión a través de técnicas tradicionales de clasificación automática de textos

This work studies the scope of traditional techniques used in automatic text classification (v. gr. bag of words) for the deceptive detection. Commonly, traditional techniques work well in thematic classification. However, it is desired to know the performance of these techniques in an intuitively non-thematic task. The collection used is a set of English notes of hotel reviews, including truthful and deceptive notes. Experiments were performed using bag of words with binary weighing schemes, tf and tf-idf and training a probabilistic classifier. The results show that deception can be detected with the traditional approach. A first analysis of these results identifies those elements on which discrimination fell.