Abstract:
In this project, we ventured into computational gastronomy to craft AI-generated recipes that embody diverse regional cooking styles. Our methodology initially centered on a novel tokenization approach, using the GPT-2-based AI model Ratatouille and integrating cuisine-specific tokens within the comprehensive RecipeDB. This strategy yielded decent results in reflecting regional culinary styles. However, recognizing the potential for enhancement, we shifted our focus towards advanced recipe encoding methods. Our challenge was to capture the unique culinary essence of various regions, a task with no straightforward solution. Utilizing the comprehensive RecipeDB and the AI model Ratatouille, based on GPT-2, we experimented with different recipe encoding methods. These included both semantic and contextual models to generate ’embeddings’ – numerical representations of recipes that computers can analyze. The effectiveness of these embeddings was vital in assessing mirroring distinct regional flavors, drawing from language processing research to represent diverse cooking styles.
