Exploring Deep Neural Networks and Decision Tree for Spanish Text Classification

Abstract

Nowadays, huge amounts of information are available on social networks, blogs, websites, and digital libraries. Most of this information is in unstructured text format, so text mining approaches have become increasingly studied to process all this data. Text classification aims to automatically classify documents into predetermined categories, applying machine learning (ML) algorithms. In this paper, we collected a dataset set related to reviews of a food store in Peru and compared different vectorization models, such as Term Frequency Inverse Document Frequency (TF-IDF), Bag of Words (BoW), and classification algorithms, such as traditional ML classifiers SVM, Decision Tree, MLP, KNN, Naive Bayes and a recent approach "deep jointly informed neural networks"(DJINN) that initialize deep feedforward neural networks based on decision trees. The results show DJINN gets a F1-score higher than traditional ML, being a promising technique for text classification.