INCORPORATING INDOBERT INTO DIET CLASSIFIER TO ENHANCE INTENT AND ENTITY RECOGNITION IN CULINARY CHATBOT

Dining out, working from cafes, and socializing in restaurants become the modern lifestyles, creating a growing need for accurate tools like culinary chatbots. This paper introduces Restbot, a restaurant-related chatbot. Unlike traditional heuristic or machine learning methods, which often struggle with dynamic data or complex queries, Restbot leverages a customized Dual Intent Entity Transformer (DIET) model integrated with IndoBERT. It improves intent detection and entity recognition, particularly for nonstandard and slang Indonesian language inputs common in the culinary domain. Using a dataset of 4,011 sentences across 17 intents and 1,079 menus from 203 restaurants, the DIET+IndoBERT model achieves significant performance compared with DIET: 96% F1-score for entity recognition (4% higher) and 92% F1-score for intent detection (3% improvement). This research contributes a novel approach to enhancing chatbot accuracy in specialized domains, paving the way for more intuitive and user-friendly culinary assistants. ICIC International