Automatic Categorization of Tagalog Documents Using Support Vector Machines

31st Pacific Asia Conference on Language, Information and Computation (PACLIC 31), pages 346–353 Cebu City, Philippines, November 16-18, 2017 Copyright c⃝2017 April Dae Bation, Aileen Joan Vicente and Erlyn Manguilimotan Automatic Categorization of Tagalog Documents Using Support Vector Machines April Dae C. Bation Erlyn Q. Manguilimotan Aileen Joan O. Vicente Department of Computer Science University of the Philippines Cebu Cebu City, Philippines 6000 acbation@up.edu.ph AI Innovation Center Weather News, Inc. Chiba, Japan 261-0023 erlynqm@gmail.com Department of Computer Science University of the Philippines Cebu Cebu City, Philippines 6000 aovicente@up.edu.ph Abstract Automatic document classification is now a growing research topic in Natural Language Processing. Several techniques were incorporated to build a classifier that can categorize documents written in specific languages into their designated categories. This study builds an automatic document classifier using machine learning which is suited for Tagalog documents. The documents used were news articles scraped from Tagalog news portals. These documents were manually annotated into different categories and later on, underwent preprocessing techniques such as stemming and removal of stopwords. Different document representations were also used to explore which representation performed best with the classifiers. The SVM classifier using the stemmed dataset which was represented using TF-IDF values yielded an F-score of 91.99% and an overall accuracy of 92%. It outperformed all other combinations of document representations and classifiers. 1 Introduction Due to the explosive growth of documents in digital form, automatic text categorization has become an important area of research. It is the task of assigning documents, based solely on its contents, to predefined classes or categories. Through time, approaches to this field of study evolved from knowledge engineering to machine learning. In the machine learning approach, the defining characteristics of each document are learned by the model from a set of annotated documents used as “training” data. Such includes Naïve Bayes and Support Vector Machine classifiers. Different standard machine learning techniques treat text categorization as a standard classification problem, and thereby reducing the learning process into two steps — feature selection and classification learning over the feature space (Peng et. al., 2003). Of these two steps, feature selection is more critical since identifying the right features will guarantee any reasonable machine learning technique or classifier to perform well (Scott & Matwin, 1999). However, feature selection is language-dependent. Several preprocessing methods such as stopword removal, lemmatization and root-word extraction require domain knowledge of the language used (Peng et. al., 2003). Methodologies used in researches concerning automatic document categorization are unique from language to language, depending on the structure and morphological rules of the specific language. Although automatic text categorization is becoming a great area of research in most languages aside from English such as Chinese and Arabic, researchers have paid little to no attention 346 in categorizing Tagalog documents. Tagalog exhibits morphological phenomena that makes it a little different than the English language. Thus, this study aims to investigate the factors and explore on different methods that will affect the process of building a Tagalog document classifier. Specifically, this …

Content truncated to 500 words