Abstract:
Jakarta as a city with a vision to advance its city and make its people happy is always creating innovations in achieving its goals. One important indicator in achieving ...Show MoreMetadata
Abstract:
Jakarta as a city with a vision to advance its city and make its people happy is always creating innovations in achieving its goals. One important indicator in achieving this goal is the provision of interactive services to the community. Jakarta Kini (JAKI) is a product of the City of Jakarta that provides various services that can be accessed by the community. One of the JAKI features intended to expand the affordability of use is a chatbot. Chatbot provides an alternative form of interaction that is more flexible by natural human conversation. However, before this research, Jakarta Smart City (JSC) as the developer of JAKI did not have its chatbot model. Therefore, an intelligent chatbot model is needed that can understand various requests or questions from the public based on the Indonesian language with flexibility. This study aims to build an Indonesian-language chatbot in the public service domain in the JAKI application. Experiments will include the use of various embeddings and classifiers in the intent classification pipeline model. Embeddings tried to include Word2Vec, GloVe, FastText, and BERT. The classifiers used were the built-in Rasa framework DIET Classifier and SVM as the mathematical model. The best model is obtained by a pipeline model that uses a DIET classifier with IndoBERT contextual embedding witan h F1-score performance of 0.93.
Published in: 2023 IEEE International Smart Cities Conference (ISC2)
Date of Conference: 24-27 September 2023
Date Added to IEEE Xplore: 31 October 2023
ISBN Information: