Intent Detection Dan Slot Filling Secara Joint Learning Pada Chatbot Rekomendasi Bahan Aktif Skincare Menggunakan Indobert-Crf

Abstract

Skincare has become an essential aspect, particularly among women. The growing market demand and interest have led skincare manufacturers to launch various products with different functions and purposes. The main component that determines the effectiveness of skincare products is the suitability of their ingredients with the user’s skin condition. This research aims to implement a Natural Language Processing (NLP) task—intent detection and slot filling—using a IndoBERT-CRF model in a chatbot designed to provide ingredient recommendations related to skincare. The dataset consists of 2,802 user requests expressing curiosity about skincare ingredients, such as suitable ingredient recommendations, ingredients to avoid, skincare steps, ingredient information, and skincare routines. The data were preprocessed and used for model training. The results show that the chatbot utilizing intent detection and slot filling with the IndoBERT-CRF model can provide accurate responses, achieving an average respon time 570,27 ms, accuracy 93%, recall 97.2%, precission 95.8%, and F1 Score 96.49%. These results indicate that the chatbot can accurately respond to user queries.