Deteksi Intensi Chatbot Berbahasa Indonesia dengan Menggunakan Metode Capsule Network

Abstract

Intent detection is a process of classifying customer intention from given sentence or chatting. One of the uses of intent detection is in a chatbot. With intent detection, the chatbot can detect the customer intent. However, currently the use of intent detection has not been implemented by most companies in Indonesia. A good intent detection method for chatbots is one that is able to classify user intentions accurately and quickly. This study aims to perform intent detection of messages from the Indonesian language chatbot dataset obtained from customer conversations of PT. Kazee using the Capsule Network(CapsNet) method. With this research, it is hoped that the chatbot of PT. Kazee can respond to customers more appropriately. On this study we conducted experiments and analyze the use of Capsule Network (CapsNet) method in detecting the intent of PT. Kazee customers conversation. The dataset of the experiments contains questions about PT. Kazee. There are two types of datasets—a dataset with six intentions and a dataset with 18 intentions. We compare the result of this experiment with the BERT intent detection model we used previously. The experiment show that the execution time of CapsNet method is faster than that of BERT. However, BERT is still superior to the CapsNet for the ability to respond appropriately. The CapsNet method can be considered for use on chatbots that are more concerned with execution speed.