FICOBU: Filipino WordNet construction using decision tree and language modeling

Abstract

The paper discusses the approach in creating a Filipino WordNet. A semi-supervised learning approach using Decision Tree and Language Modeling. This will take advantage on the information found on the web. It will help future NLP researchers in Filipino language. The approach uses words from a dictionary as preliminary data and as seed for the search engine to start crawling the WWW. To decide if the word is part of Filipino language, the word will first undergo in Code-Switching Points Module (CSPD). CSPD scores the word by using the frequency counts of word bigrams and unigrams from language models which were trained from an existing and available corpus. After scoring, Filipino Stemmer will get the stem of the word and examine if the stem word is part of the said language. Once the words were scored and stemmed, the archive will evaluate if the word is Filipino. To test the accuracy of the system, we collected different articles around the web and then grouped it into two groups - Plain Filipino and Bilingual. The result shows the F-measure for Plain Filipino Category range between 65.65% - 96.85% with an average of 85.64% while for Bilingual range between 60% - 100% with an average of 88.17%.