CookRecipe

: towards a versatile and fully-fledged recipe analysis and learning system

Abstract

The recent explosive proliferation of interesting and useful data over the Web including various cooking recipes, while providing people with practically useful information, brings out a few challenging issues on how to manage and best utilize such non-conventional data effectively. To respond the challenges, this thesis presents a brand-new recipe analysis and learning system, the CookRecipe, which is developed to manage recipes in a novel way radically different from existing online recipe systems. More specifically, a RecipeCrawler is established to automatically collect a large number of recipes from various online resources such as websites, blogs and forums. A recipe model is devised to well represent recipe data, which not only covers the static recipe attributes but also elucidates the dynamic cooking behaviors. The cooking procedure of each recipe is represented and visualized as a graph, thereby providing a detailed and formalized description of how a dish is cooked. With such a recipe model and a set of corresponding properties proposed, each raw recipe crawled from the Web can be logically determined whether it contains errors or not. Such a mechanism can be further extended to logically identify the correctness for huge amounts of raw data on the Web that share similar dynamic behaviors with recipe data such as instructional data and workflow data. Several knowledge level facilities such as rules and constraints, discovery of cooking techniques and user profiling are also presented to accommodate the tasks of analyzing and comparing recipes, based on which users are able to learn basic cooking techniques as well as specific techniques involved in a particular recipe. Such knowledge facilities further lay down a solid foundation for solving some practical problems such as personalized recipe recommendation and recipe adaptation at the application level. The usefulness and elegancy of CookRecipe are demonstrated through a prototype system called RecipeView, in which a novel recipe retrieval method is developed for users to query similar recipes by example. To increase the query efficiency on user modified graphs, a domain-specific graph segmentation method is proposed for cooking graph indexing in accordance with a filtering algorithm focusing on the distinguishable structures and the shared percentage of graphs. The experiment results show that the filtering algorithm significantly reduces the graphs to be accessed and exhibits a good combination with the cooking graph mining system. Furthermore, CookRecipe prototype is also shown to be portable to mobile devices, which offers PDA users a series of basic services such as browsing, editing recipes and keyword-based recipe search.

Date of Award	15 Jul 2008
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Qing LI (Supervisor)