CHATBOT RESEP MAKANAN KHAS BANDUNG BERBASIS TEXT DENGAN METODE NATURAL LANGUAGE PROCESSING

Abstract

Traditional food is processed food typical of the local area that is usually consumed by the surrounding community. These recipes can be searched from various media ranging from recipe books, the internet, and so on, because the sources are so many, making it difficult for users to determine the most appropriate recipe. Therefore there’s need an application or system that integrates these media and can assist in finding food recipes in order to save time in finding food recipe information. One solution is a chatbot, a chatbot is a system that imitates how humans communicate with a knowledge base stored in a database and is made to provide responses or answers that can be understood by humans. because the chatbot conveys information to be effective, easier, available at any time, can work 24 hours and provide information quickly but the problem that occurred in previous research was how the chatbot could choose the right response, response or answer from the response data set it has so that the answers provided match the user's questions. this research will build a chatbot that is able to understand questions about typical Bandung food recipes in detail, the system is made using preprocessing, mark removal, case folding and tokenizing, after that a training model is made from the chatbot response database using learning to get the intent of the question, then after getting the intent, then enter the manager dialog to match the intent with the intent in the chatbot response database so that the chatbot can issue a response according to the intent of the question, then select the response. The process of selecting responses from chatbots in this study uses a template base, testing in this study was carried out by answering according to the questions that had been previously scenariod as many as 40 questions, with the results obtained that 75% or 30 of the 40 questions were successfully answered, and 25% or 10 of the 40 questions answered were not in accordance with chatbots.