Hello, this is 
Figure 1. Various foods and IBM Watson
Can we discover new recipes through food? Can we innovate creative tastes by trying new recipes? Actually, as a result of contemplating such questions, IBM has launched Chef Watson. Chef Watson combines recipes by considering how to configure recipes according to the ingredients, how ingredients change, and how the foods will taste and match with other foods. To do this, it is necessary to make combinations based on Watson’s basic knowledge on food/ingredients and recipes created by people.
It can create recipes, etc. for vegetarians and people who are allergic to certain foods, and it can also create recipes that can reduce food waste. Through the development of such recipes, even the book called “Cognitive Cooking with Chef Watson: Recipes for Innovation from IBM & the Institute of Culinary Education” was published.
Figure 2. IBM Chef Watson’s main page for combining food
After picking and choosing ingredients, it recommends foods based on good combinations as shown below. When I picked strawberry, mango, etc., it first recommended strawberry pie.
Figure 3. After the ingredients have been picked, it provides a list of recommended foods
Should we try a little more difficult combination? Let’s take a look at the combination of strawberry, salmon, honey, and whisky. Aren’t you curious what kind of recipe it will create? Let’s look at the result.
Figure 4. Creation of a new recipe
Combination of strawberries and fish, what an original dish. Being the first of its kind of course, there is no photo of the food.(...) No matter what the food is, there must be cooking instructions to actually make the food, right?
So, Chef Watson presents the method.
Figure 5. Strawberry curry recipe (...)
Strawberry curry recipe, that’s original. Of course, I haven’t actually tried it, but I think it might be a pretty good combination. Sometimes people use apples in curry.
Emergence of such services reflects people’s interest in cooking while enabling them to be creative within the lines of providing taste and assistance to meet our demands.
Then, in order to provide such services, how can we reflect people’s opinions? There are many ways, but one of them is RDR (ripple down rules). In terms of applying various cases, it classifies which conditions meet or don’t meet the criteria. When a condition does not meet the criteria, a different label can be attached or a new label can be created.
Figure 6. Basic structure of RDR
When criteria 2 is met, it can go onto action 2 or criteria 3, but a sub-criteria can be added to criteria 2. Among the instance of criteria 2 being met, a new criteria 4 related to the subset is determined.
Figure 7. An example of RDR related to the correction of constraint conditions of meal planning
Example of RDR related to the correction of constraint conditions of meal planning For example, let’s assume the following process: 1) determine whether the user is a male, and if so, increase the calorie of the meal; 2) determine whether the user has underlying diseases, and if there is none, maintain the meal plan; 3) if the user has underlying diseases, first determine whether the user has diabetes and limit carbohydrates.
Also, rules related to pregnant women can be added. Here, we can think of two situations. First, this rule can be added after determining the diabetes condition, and the other is adding this rule as an exception handling in the part for determining underlying diseases. Each user or expert can add such rules anywhere. In this case, it is appropriate to make it into an exception handling. It is the process of “sub-classifying” the pregnancy as a detail added to the state of being normal. The advantage of exception handling is that additional details related to each situation can be assigned as a condition without breaking down the existing series of discriminant structures.
Also, the addition of such condition is ultimately related to adjusting the limiting conditions. That’s because the addition of rules and setting actions are the modification of an appropriate scope for calories or nutrients.
In summary, there are cases which a rule is satisfied, not satisfied (ELSE), or satisfied but an additional detailed classification is needed. Add to that, an additional (composed of criteria and action) rule can be created/modified/deleted when a criteria is not met or when a detailed classification is needed.
Accordingly, the following are the advantages of RDR in my opinion.
- The opinions of users and experts can be directly reflected in the decision structure.
- It is easy to describe the addition of a new criteria.
- General principle concerning meal planning can be extracted or organized in consideration of the type and characteristic of a rule.
Incidentally, the following are the disadvantages of RDR in my opinion.
- Because additional rules can be created continuously, the structure can become very complex and difficult to be organized.
- If a hierarchical structure is not thought of in advance, the vertical relationship between rules can get tangled and lead to inefficient decision and processing.
Ultimately, if rules are applied consecutively, a comprehensive rule can be formed as well.
< Combined rules >
Rule: 1) If the user feels hungry frequently, and 2) prefers to reduce fat consumption, Action: 1) Comprise foods that are easily digested and high in volume, and 2) substitute foods with high fat contents such as bacon and mayonnaise with salad or oriental dressing.
When each user’s recipes accumulate and information on rules reflected in such recipes accumulate, the set of rules that can occur in specific situation can be considered in order to create new recipes that reflect personal preferences and reflect the rules according to the situation.
In addition, because it is easy to add/delete/modify such rules when using RDR, it is possible to connect to the user’s external situations and conditions.
It is possible to accept various information of the user, process the information, and reflect it in the rules. We will delve further into this topic next time.
Figure reference (Reference, Copyright indication)
Fig 1, https://www.flickr.com/photos/ibm_es/24863070074 , Public domain
Fig 2, https://www.flickr.com/photos/ibm_research_zurich/18836492225/in/album-72157645397988715/ , CC-BY-ND 2.0
Fig 3, https://www.flickr.com/photos/ibm_research_zurich/18810247426/in/album-72157645397988715/ , CC-BY-ND 2.0
Fig 4, https://www.flickr.com/photos/ibm_research_zurich/18831404922/in/album-72157645397988715/, CC-BY-ND 2.0
Fig 5, https://www.flickr.com/photos/ibm_research_zurich/18831404652/in/album-72157645397988715/ , CC-BY-ND 2.0
Fig 6, Self-made
Fig 7, Self-made
Reference
[1] Khan AS, Hoffmann A. , Building a case-based diet recommendation system without a knowledge engineer, Artif Intell Med. 2003 Feb;27(2):155-79.
[2] Gaál B, Vassányi I, Kozmann G. , A novel artificial intelligence method for weekly dietary menu planning. , Methods Inf Med. 2005;44(5):655-64.
[3] Johan Aberg, Dealing with Malnutrition: A Meal Planning System for Elderly, AAAI Spring Symposium: Argumentation for Consumers of Healthcare, 2006
[4] Alessandro Mazzei, Luca Anselma, Franco De Michieli, Andrea Bolioli, Matteo Casu, Jelle Gerbrandy, Ivan Lunardi, Mobile Computing and Artificial Intelligence for Diet Management, ICIAP 2015 Workshops pp 342-349
[5] “왓슨과 함께 신메뉴 개발에 도전해 보세요!”, https://www-03.ibm.com/press/kr/ko/pressrelease/47402.wss
[6] Rodrigo Zenun Franco, Rosalind Fallaize, Julie A Lovegrove, and Faustina Hwang, Popular Nutrition-Related Mobile Apps: A Feature Assessment, JMIR Mhealth Uhealth. 2016 Jul-Sep; 4(3): e85.
[7] Gergely Kovásznai, Developing an expert system for diet recommendation, 2011 6th IEEE International Symposium on, Applied Computational Intelligence and Informatics (SACI)
