Food recommendation systems play a crucial role in assisting discovery of new recipes for users based on preference, ingredients, and health. In this paper, we present a recipe recommendation algorithm that leverages embedding with clustering & similarity analysis for ingredient-based input. This approach has been used for several recommendation systems recently with results that compete with state-of-the-art recommendation systems. We will use multi-modal data to understand further similarities between recipes and ingredients and conduct exploratory analysis of both the data and the model with further preprocessing and fine-tuning. The performance of our model will be compared to state-of-the-art recommendation systems using precision, recall, F1 score, and NDCG in future steps. Further, we will cross-examine the generated ingredient recipe list with existing recipes both in the dataset and beyond. Through integration with smart devices, this project aims to assist users in making informed decisions about what to cook.

You can download Proposal PDF here

You can download the final research paper here