Data Driven Approach for Optimizing the Quality in Gastronomic Recipe Recommendation

Abstract

Food is playing an ultimate role in the human life not only by being a biological need, but by being a defining factor of the human civilization. Preparation of the food makes it more digestible and healthy as well as it has led to have unique clusters of food preparation patterns around the world. The choice of the food is mainly depended on its flavour and nutrient but the biasness towards the flavour factor when choosing food has lead human to effect badly on their healthier lifestyle. Several health promotion works have been done but proven unsuccessful due to inadequate acceptance in real world situations. The major reason for this rejection is sacrificing flavour factor over the nutrient enhancement food. Therefore, healthier foods tend to be unpalatable which make them unaccepted by the majority in the society. The personalized nature of food choices have also affected this problem. Due to the unprecedented growth of internet, food recipe sharing is also grown which has enabled the access to a large amount of worldwide recipe data to explore. We are utilizing a large food recipe dataset and analyse the patterns of food combinations which are popular among several cuisines. Food pairing hypothesis which has analysed the cuisine specificity of a range of raw food items has laid the foundation to our study. Food pairing patterns are then interpreted using several indicators provided by the Food Pairing Hypothesis and use to assess the novel ingredient combinations. We are defining flavour and nutrient as the quality variables of the food recipes and suggesting a framework that optimizes the quality of the food recipe when altering a single ingredient. Altering a single ingredient is a common phenomenon occurred in kitchens every day and recipe recommendation using the proposed quality variables has not been tried before. Our data driven framework consists of classification model which have achieved 79.38% prediction accuracy when detecting the cuisine. Framework then optimizes the changing ingredient according to the original recipe’s classified cuisine. The framework encompasses of flavour optimization phase and nutrient optimization phase which utilizes a generalized flavour recommendation and a personalized nutrient recommendation approach. Finally, the evaluation of the generated candidate ingredients for the altering ingredient based on data driven testing and by expert opinion provided approximate acceptance ratings