Computational gastronomy: novel recipe generation with constraint optimization

Abstract

In response to the escalating demand for the generation of novel and diverse cooking recipes, this research in Natural Language Processing introduces a new tool—Ratatouille. The tool utilizes various Deep Learning models, including Long Short-Term Memory (LSTM) networks and the Generative Pre-trained Transformer-2 (GPT-2), to fulfill the need for creating authentic and inventive recipes based on user-specified ingredients. Trained on a substantial dataset of recipes, Ratatouille addresses the challenge of generating diverse culinary ideas. By incorporating models like character-level LSTM, word-level LSTM, and GPT-2, the tool provides users with a platform to explore and generate recipes. Evaluation using the BLEU score underscores the ongoing challenge of assessing the quality of generated recipes. Ratatouille serves as a user-friendly solution to meet the demand for diverse recipe generation, highlighting potential directions for future research in this dynamic field.