Automatic Multi-label Classification for GDP Economic-phenomenon News

摘要

GDP is a measure of a country's economy. One of the variables used in the process of compiling GDP is the news analysis of economic phenomena. Briefly, the GDP figures are getting better if they are in line with the economic phenomena that occur. Analysis of this economic phenomenon can also be a supporting component of GDP publicity. This study aims to investigate the best classification model for economic phenomenon news to speed up the analysis process. We use a multi-label classification method because each news item has one or more categories. The label in this study corresponds to the GDP by Industry which refers to the Indonesian Standard Industrial Classification (KBLI). We use the Problem Transformation approach in combination with several single label classification algorithms. Label Power-set method combined with Linear SVC showed a better result among others, it reaches 75% for F-Measure and 0.021 for Hamming Loss in 5-fold cross-validation. However, the model performance increase to 84% in testing mode.