Deep Learning Generative Indonesian Response Model Chatbot for JKN-KIS

Jaminan Kesehatan Nasional- Kartu Indonesia Sehat (JKN-KIS) is a government program for the Ministry of Health Indonesia. Information of all the participant is the key to the sustainability of the JKN-KIS program. Lack of socialization and reading guidebooks desire are the main reasons why JKN-KIS services do not run properly and has many problem in the community. A question answering system (chatbot) has many benefits because it can automatically responds to questions and help the government with public communication. Unfortunately, most chatbot services today do not offer a completely automated solution. There are lots of chatbots out still static, so the resulting response must comply with certain rules. That chatbot is not scalable and less effective when applied to large JKN-KIS data. However, this research develops a generative model for an automatic question answering system (chatbot) in the Indonesian language with domain Indonesia JKN-KIS. The generative chatbot model is built with the Sequence to Sequence (Seq2Seq) model, encoder and decoder, with the LSTM Multiplicative Attention method. The dataset used is question and answer pairs from JKN-KIS guidebook. The results show that the chatbot able to solved the problem with a large and varied Indonesia dataset without certain rules. With a 20 thousand question and answer conversation pairs dataset, this research get the best chatbot results with the parameter at 6000 iterations for 15 batch sizes and 1000 hidden size architectures. The results show that the resulting loss value is 0.23 with a BLEU Score of unigram 0.86 and a BLEU Score of bigrams 0.85.