Klasifikasi Intent pada Chatbot Pembelajaran Tenun Ulos dengan Menggunakan RASA Framework dan BERT Language Model

Abstract

Ulos is an important cultural heritage of North Sumatra that should be preserved. This can be accomplished by making knowledge about Ulos easily accessible to users. An approach that can be utilized in this case is the use of chatbots. There are several terms in the culture of Ulos that have similar meanings, but they have different terminologies, which often causes confusion for those who wish to learn more about Ulos. In order to address this issue, the chatbot must be able to identify user intents effectively. This study employs a BERT pre-trained model for intent classification of the Ulos information chatbot that was developed in RASA Framework. As a result of this research, the chatbot was tested with RASA Test and end to end testing. a chatbot is able to predict and answer user questions correctly with an accuracy of 96%. The chatbot was also evaluated for its ability to predict the user's intent through RASA’s test_stories test and answer it with an accuracy of 79.34%. For the end to end testing result, the chatbot’s accuracy reach 81,5% for identify user’s intent and reach 80.9% F1 score.