News Classification on Twitter Using Naive Bayes and Hypernym-Hyponym Based Feature Expansion

The large number of posted news in form of tweets resulted in scattered tweets and appearing on the Twitter homepage very diverse and not classified by categories such as health, sports, technology, economics, tourism and so on. The absence of categorization causes the user difficulty to read or retrieve information related to certain desired categories. Solution that can be done is by the method of text classification, which in the process of classification is able to classify automatically against some categories on unstructured text with natural language. In this research, classification process will be conducted using Naive Bayes method with additional feature expansion to add term in initial document. The addition of term aims to optimize the classification process because the tweet is a short text that can lead to ambiguity of classification class. The additions made are hyponym and hypernym from original documents extracted from WordNet. Accuracy calculation method used is k-fold that aims to avoid over fitting on the document. The accuracy obtained was 72% for the classification without feature expansion, 65.75% for hyponym and hypernym addition, 66.3% for hyponym addition, and 67.5% for hypernym addition. It can be concluded that the addition of queries made less effective to improve the accuracy of the classification process.