Sentiment Analysis of BPJS Kesehatan’s Services Based on Affective Models

BPJS Kesehatan is a corporation in Indonesia which aim organizing health insurance program. By increasing the number of BPJS Kesehatan’s members every year, BPJS Kesehatan should be able to do all its services well so that its members can get their rights. BPJS Kesehatan performance can be assessed from the public response, one of the social media used by public to share their responses to BPJS Kesehtaan’s service is Twitter. BPJS Kesehatan can use these responses to find out people's opinions on their services. Therefore, this study proposes a new approach to analyzing public opinion using the field of scientific computational linguistics. Specifically by making a computing system with features, 1) Sentiment analysis using the effective models method which sees a different degree for each adjective in the commentary. Affective model is a new approach in Indonesian Language that evaluates each adjective has a different level of pleasure and arousal. This method collects adjectives in Indonesian into a context and assigns different values to each adjective. This value is obtained from the adjective mapping results from Russel's Circumplex model of affect, we also sees words that have affect polarity in a sentence and words that affect the degree of affection in a sentence. 2) Categorization, this feature is to categorize comments into types of BPJS Kesehatan’s services. There are 10 service categories, each of categories has keywords. System identified the keywords in each comment and calculated similarity with existing categories. Total data that has been obtained is SS3S2 tweets. For each data obtained sentiment value will be calculated and categorized, this system will show which service category has positive or negative sentiment. The test method uses data that has been labeled manually before and then is tested using a program. From 211 tweets that have been labeled manually, the sentiment analysis program has succeeded in achieving an accuracy of 83.4% and the categorization program produces an accuracy of 81.05%.