Text Generation for Low-Calorie Food Recipes Using IndoBART

Awareness of maintaining a healthy lifestyle is increasing in society. Often, maintaining a healthy life is difficult in maintaining food calories due to a lack of knowledge or boring low-calorie foods. Therefore, through Natural Language Processing technology, this study aims to develop an automatic system to generate variations of low-calorie food recipes in the Indonesian language. This study uses the Indonesian Benchmarking Algorithms for Data Repairing and Translation (IndoBART) and Rule-Based algorithms that generate recipes and estimate total calories based on the food ingredients used. IndoBART is used to create recipe texts, while the rule-based method is used to estimate total calories. The experimental results demonstrate that the IndoBART model is effective in generating recipe texts based on the training data, resulting in a significant decrease in both training loss and validation loss, with an accuracy of between 88% and 92%. The total calorie estimates produced by the rule-based method are also consistent compared to manual calculations. The results of this study contribute to the development of low-calorie food recipes using transformer technology, which demonstrates good performance.