A comparison of text weighting schemes on sentiment analysis of government policies: a case study of replacement of national examinations

The National Examination (UN) is a system of evaluation of education standards for elementary and secondary schools conducted nationally and is also used to equalize the quality between education levels. The national examination aims to determine graduation, national education quality mapping, and also for selection to higher education levels. Over the years the UN has become a benchmark for the standardization of education in Indonesia, meaning that the UN is very much needed to find out the size of the quality of student education and the quality of teaching of a school. The government’s policy regarding the plan to remove the UN system has received public attention. The removal of the UN is planned to be replaced with a competency assessment and character survey. In order to know the public’s sentiments regarding this policy, research needs to be done, one of which is to analyze public sentiment through social media Twitter. In text mining tasks such as text classification and sentiment analysis, careful selection of a term weighting scheme (TWS) can have a significant impact on effectiveness. We tested the effectiveness of six classification algorithms by varying the TWS in the dataset obtained from Twitter. The experimental results showed that overall TF-IGM outperformed TF-IDF on four classification algorithms. Finally, the sentiment analysis of the discourse on the removal of the UN is expected to provide a general picture to the government regarding public opinion from the point of view of data coming from social media.