The Food Oracle: A Knowledge-Rich and Dynamic Recipe System



The Food Oracle: A Knowledge-Rich and Dynamic Recipe Systems

Content areas: AI and creativity, computer-aided education, art and music,

Abstract

Artificial Iintelliegence successes have often been in solving specific problems such as configureingconfiguring computer systems or playing chess. In the search for general uses of reasoning and learning systems we took cooking as a window into culture. The foodFood oracleOracle is a set of sytemstools we are creating that weave learning and reasoning about cooking and food together into a resource to help people explore everyday activitys from the huge corpuses of food and cookingcreative and intuitive cooking. Using the partially structured knowledge informal internet informationmined from the Internet, we have constructed a number of prototype explorations for various kinds of thinking about food.Today Artificial Intelligence is being applied to a myriad of very specific domains like aircraft maintenance, but we often lose sight of opportunities to assist in and enrich everyday tasks. Food jumps out as an under-explored, impactful, yet still plausible domain of investigation, given its daily relevance to our lives, and the recent accessibility of large-scale computational resources for food on the World Wide Web. In the Food Oracle project, Wwe weave together disparate knowledge mined from the Web and a culinaryn expert’s database into a single rich resource. The system reasons r intelligently about recipe-sensitive ingredient substitutionseasons over a intgrediant substitution list, about thean essesnce of recipee “essences” of recipes, system, and a reasoner abouand tover a database of food and culture. that we . This rich resource, along with tools for smart recipe substitutions, and “gisting” the essence of a dish, is being integrated to create adaptive and dynamic recipes that can also deliver just-in-time educational information to the user. Our goal is to create a smart recipe system that will foster users to think creatively and critically about food as they explore adaptive and dynamically generated recipes laden with a wealth of practical and historical information about food.to create semantic substitution through automated means for large corpoi of knowledge In the mean teim we are shapingshape the Food Oracle into an education an educational al and playful tool that towill empower novice cooks to think creatively and flexibly about food.

1 Introduction[1]

Food is a representation of a culutureculture, of its customs, of the roles of its people, and of the efforts we takemake for each other. Cooking represents a vast interwoven knowledge of nutrition, ingredients, processes and technology. It is also It is more than a worthy representation of human knowledge, and of course a In today’s fast-paced, commerce driven world, the art of cooking often seems to be lost in the shuffle. Noas no one can deny the beauty of a home-cooked meal. , yetYet sadly, fewer and fewer people have the confidence to cook. Those who do cook often follow recipes to letter, and prefer recipe books that “dumb down” the ingredients and procedures.

While cultural pessimists see an inevitable decline, we see an opportunity for AI technology to help people forage for food and cook from recipes in a more flexible and creative way. Such a technology can educate and empower people, giving them the confidence to be creative in the kitchen again. Through learning and teaching about food, we expect to learn about knowledge as well.Such a technology can educate and empower people, giving them the confidence to be creative in the kitchen again.

So how do we begin to embark upon such an ambitious venture? There is already a vast resource of knowledge about food, ingredients, and recipes available on the World Wide Web. Case and point: A Google search for “fettucine alfredo” returns over 8,230 000 web pages of, inter alia, recipes, menus, and historical accounts; the Searchable Online Archive of Recipes (SOAR) database [SOAR Website] boasts over 70,000 machine-readable recipes; Lori Alden’s Cook’s Thesaurus [Cook’s Thesaurus Website] is a machine-readable encyclopedia of many hundreds of ingredients, with pictures, taxonomic and functional information, and ideas for substitutions.

However, these informational gems are scattered across the Web, are wholly unconnected to each other, and therefore, they do not compel people to use the Web in everyday food foraging, and cooking. The import of Artificial Intelligence to this venture is to gather diverse sources of information over the Web, “make sense” of the knowledge by performing semantic analysis and data correlation across different sources, synthesize knowledge into a coherent and compelling story about food, and present this information to the user in a useful and context-sensitive way. The Food Oracle project at the MIT Media Laboratory Counter Intelligence Group aims to do precisely this.

While work on the Food Oracle is ongoing, the ensuing sections of this paper represents a prospectus on a set of diverse resources we have built to be a “society” of counterbalances to each other in reasoning about recipes.small society of informational resources and tools tha t we have already built to help people “make sense” about food. In the remainder of the paper, we discuss relevant work and future directions for this project.

2 Cultural Mining of the Web

In order to weave together Food knowledge is aa compelling and rich tapestry of culture, processes, technology, and ingredients. We follow the multiple intelligences approach to attempt to make a computer see food in this richness. food knowledge,W we first mined the Web and our own proprietary database about culinary history.

Barbara Wheaton, who is the honorary curator of the Culinary History collection at the Harvard-Radcliffe Schlesinger Library has been compiling a computer database of the historical uses of food, which for some 40 years. Our goal is to be able to answer food queries with the facility of Barbara. Starting from her database, we were available to mine knowledge about ingredients and their occurrences in historical recipes through Western Culinary History.

We augment her personal view withbuilt several semantic agents to crawl and scrape mine several freely available recipe databases off of the Web. From each plaintext recipe, we created a parsed representation with the following components:

1. Title and taxonomic information about the geographic, dietary, and ethnic context of the recipe.

2. List of ingredients, with parsed measurements, and explicit substitutions.

3. List of procedural steps, indexed by cooking procedure keywords (e.g. “poach”) and the ingredients involved in that procedure (e.g. “1 cup sliced pears”)

Similarly for ingredients, we compiled together a parsed knowledge base of ingredients, their possible substitutions sorted by purpose (e.g. color, texture, spice, etc.), their place in the overall functional and structural taxonomy of ingredients, and some educational, cultural, and historical context of that ingredient.

These threeethree resources are iWe were also fortunate enough to work with a proprietary database of the historical uses of food, compiled by Barbara Wheaton, who is the honorary curator of the Culinary History collection at the Harvard-Radcliffe Schlesinger Library. From this database, we were available to mine out a wealth of knowledge about ingredients and their occurrences in historical recipes through Western Culinary History.

Integrated with ing these three resources, we created a user interface tool called HyperRecipes – information-laden recipes which allow the user to click on particular ingredients within any given recipe to learn more about ingredients, recipes or, or to click on historical occurrences of a recipe, ingredient, or procedure in earlier timesuses of foods.

3 Smart Recipe Substitutions

We implementedWe are developing heuristics and an algorithmapproaches for evaluating the appropriateness of certain ingredient substitutions in the context of a particular recipe. The larger goal is to enable the flexible adaptation of recipes to ingredients available in the user’s kitchen.

One tTThe traditional methods for deciding on appropriate substitutions is to consult a chart of substitutions. However, we argue that it actually takes a great deal of common sense about cooking to pick the appropriate substitution for each dish and situation. Our system’s goal is to understand the common sense of a good cook and teach it to novices.While experienced cooks might possess some of this knowledge, novice cooks generally do not.

In addition toThe system consultsing the usual substitution charts, and we tests each out-of-context substitutions for plausibility. It uses In effect, we asktThe Wweb becomesas a common sense oracle that lookslooks to see the Web, if has anyone ever cooked this same (or similar)food recipea particular dish using the a particularcandidate substitution.? In addition to unary ingredient substitutions, our algorithm also considers the case where there is a compositional substitution, e.g. Italian seasoning substitutes for oregano, plus basil, plus thyme, etc. Using this algorithm, we are able to suggest much more reliable substitutions, sensitive to the context of each recipe. Currently weWe are working on measurement equivalencies of substitutes, automaticallyed rewriting the cooking procedures, to take into account the substitution, and integrating the smart substitution technology into a recipe browser that adapts ingredients to those in the user’s kitchen.

4 Gisting the Essence of Food

There What is common among theare hundreds of recipes for guacamole on the Web?.. However, when novice cooks encounter any single recipe, they are often unable to distinguish between what is essential to the guacamole, and what is extra. To help educate and impart intuition to novice cooks, we built a mechanism to identify the essence of food, that is to say, the ingredients which seem to be at the “core” of guacamole, and the functional roles that each ingredient plays (e.g. the avocado is for a pasty texture, and the chopped onions and garlic is for spicy taste and chunky texture).

“Gisting” the essence of food is done by semantic correlation of ingredients common across variants of a recipe to determine the saliency of ingredients.. It is also desirable to abstract away from correlating just ingredients alone, and acknowledge that sometimes, it is not an ingredient in the core that constitutes the essence of a dish per se, but rather, a functional role such as a taste or texture (e.g. we can use anything spicy in guacamole, and not just paprika). Using the functional and structural taxonomies we mined from the Web, the Gister generates functional hypotheses about the role of each ingredient in a dish, and compares variants of recipes for the dish to assess an ingredient’s likely functional role. Salient Iingredients and functional roles common toacross a recipe’s all variants are placed atform the essence of that e dish.

We are currently working on improving and integrating the identified food essences into a recipe such that vital ingredients and are syntactically highlighted to facilitate the user in visualizing what’s important about a recipe, and vital functional roles are used to inform the smart substitution mechanism.

5 Related Work

A preponderance of commercial software exists for viewing and searching recipes using standard HTML forms. A representative program is AccuChef [AccuChef Website], which, in addition to allowing recipes to be searched by keywords, performs some US-to-metric conversions and scales recipes by quantity. By comparison, with the Food Oracle, we are interested in creating smart recipes which can adapt to a user’s available ingredients, indicate the importance and functional role of ingredients, and hyperlink to a wealth of knowledge and history of ingredients and dishes.

Another related work is The Interactive Chef [Chen et al.], a voice-interfaced task-sensitive cooking assistant which steps through the procedures in a recipe, providing useful information and videos along the way. Our work focuses on a the complementary issue of recipe understanding and integrating many diverse sources of knowledge for education and creative exploration.

6 Conclusion

We have built several recipe-understanding tools to: synthesize together disparate knowledge about recipes, ingredients, and food history knowledge; generate smart, context-savvy substitutions; and gist the essential ingredients and functional roles of a dish. In Oour ongoing work on the Food Oracle , our ultimate goalstrives for is a completelay flexible recipe system that can simplify itself, add cultural influence, adapt to available ingredients, and provide in-context knowledge about food. We want to empower novice cooks with intuition and the confidence to think creatively and playfully about the art ofabout cooking.

Most importantly, we see food knowledge as vast and as differentiated as knowledge itself. We are building a “society” of reasoning and learning systems that understand and teach users about food types, recipes and ways of making food. We expect From building these prototypical tools, we expect to glean methodological insights to help us understand to help us understand large corpus- based reasoning inabout other domains with common senses.

References

[AccuChef Website] AccuChef Website. .

[Chen et al.] Leonard Chen, Sandra Cheng, Larry Birnbaum, and Kristian J. Hammond. The Interactive Chef: A Task-Sensitive Assistant. In Proceedings of IUI 2002.

[Cook’s Thesaurus Website] Cook’s Thesaurus Website.

[SOAR Website] SOAR Website.

-----------------------

Copyright © 2003, American Association for Artificial Intelligence

(). All rights reserved.

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download