Health-aware Food Planner: A Personalized Recipe Generation Approach Based on GPT-2

What to eat today? With the flourish of Internet, more and more people nowadays are inclined to find an answer to this most problematic question online. The recent explosion of food networks; however, produces large volumes of recipes, making it even harder to make an informed decision. This yields the need for advanced decision-making algorithms and efficient recommendation systems. Conventional recommender systems are not feasible anymore as food is a complicated feature that presents unique challenges and is less studied. For example, it can be one of the main reasons for obesity and many other chronic diseases. Food recommender system has the potential to urge users to change their eating behaviors by adding a healthiness component as another factor in the recommendation procedure. Text generation, a hot area in machine learning, can be used as a part of a food recommender system to explore new recipes. However, existing works do not include the factors of users’ preferences, nutritional needs, and knowledge of the ingredients. In this work, we tackle this issue by proposing a new task of healthy and personalized recipe generation given only a few ingredients. We also suggest personalizing the ingredient list by integrating the user profile extracted from the previous history. Specifically, our model consists of three main components: 1) completing the given initial ingredient list by predicting the most relevant, healthy, and personalized ingredients, 2) fine-tuning GPT-2 model to generate a new recipe given the ingredients, 3) finding and recommending the top similar recipes to the generated one. In contrast to other recipe generation models, we expand the final output to be the generated recipes in addition to the top-k similar recipes from the dataset. All the proposed solutions in this work have been evaluated separately to compare their evaluation against their related works using suitable metrics. In addition to that, we did further analysis to study the hyperparameters and design options. By doing so, we intend to show our model’s ability to recommend new yet logical recipes that balance the preferences with the healthiness