Computer Cooking Contest September the 1 st 2008 ---The Taaable - - PowerPoint PPT Presentation

2

Computer Cooking Contest September the 1st 2008

• --The Taaable Team---

3

4

5

6

7

Roadmap Roadmap

Interface Interface Knowledge Knowledge acquisition acquisition CBR CBR engine engine Future Future work work Facts Facts

8

Facts Facts

9

The Team

People trainees, PhD students, professors, ...

3 labs

10

A CBR Web Application

Accessible Easy to use Web oriented Case-Based Reasoning Information retrieval Ontology engineering Knowledge representation Knowledge extraction Interface ergonomics Knowledge acquisition CBR Based Knowledge intensive

11

Architecture

12

Facts Facts CBR CBR engine engine

13

Knowledge representation

Banana Butterfinger Cake

Recipe Rex Index Idx(Rex) Language L: polynomial fragment of propositional logic Ontology O in L: set of axioms a ⇒ b banana ⇒ fruit ∈ O Generalization cost associated to each axiom in O a1∧a2∧ ... : closed world assumption on indexes

Case base : a set of recipes

banana ⇒ food ∈ O

Idx(Rex) = banana ∧ ... ∧ chocolate_chips ∧a1∧a2∧

14

Knowledge representation

If (banana ⇒ fruit, chocolate_chips ⇒ chocolate) ∈ O then Idx(Rex) ⇒ Qex Query

I want a cake with chocolate and fruits

Query Qex

Qex = chocolate ∧ fruit ∧ cake

Idx and Q are expressed in L A recipe R answers the query Q if, given the ontology O Idx(R) ⇒ Q

Banana Butterfinger Cake

Recipe Rex

Idx(Rex) = banana ∧ ... ∧ chocolate_chips ∧a1∧a2∧

Index Idx(Rex)

O

15

Retrieval principles

cost(lemon ⇒ citrus) 

K(#RecipesWith(citrus) – #RecipesWith(lemon))

Strong classification of Q

Using hierarchy search techniques, Find Idx(R) such that Idx(R) Q ⇒

Smooth classification of Q

While strong classification fails SP = next similarity path SP = (a1 b1) o ... o (aq  bq) (ai bi): substitution by generalisation, (ai ⇒ bi) ∈ O

In Taaable SP generated by increasing cost cost(SP)= ∑ cost(ai⇒bi)

i = 1..q

Results: SP (if any) and Idx(R) such that Idx(R) Q ⇒

16

Example of retrieval result

butter ⇒ butter_or_margarine margarine ⇒ butter_or_margarine chocolate_chips ⇒ chocolate almond ⇒ nut pecan ⇒ nut almond chocolate butter

Q = ∧ ∧

margarine pecan chocolate_chips cream_cheese egg sugar

a1 a2

...

Idx(R) = ∧ ∧ ∧ ∧ ∧ ∧

vanilla

∧ ∧ ∧ SP =

butter  butter_or_margarine almond  nut

• SP(Q) =

nut chocolate butter_or_margarine

∧ ∧

Elements of ontology

17

Adaptation principles

• 1. Substitution by generalizations from Idx(R) to SP(Q)

margarine pecan chocolate_chips

AP1 = ∧

• 

nut chocolate butter_or_margarine

 

butter_or_margarine butter



nuts



almond nuts nut

AP =

• 3. Composition of AP1 and AP2 and simplification of the result
• 2. AP2 = SP-1

chocolate_chips



chocolate butter_or_margarine



pecan nut



AP2 =

margarine

Following the adaptation path: left to users

18

CBR CBR engine engine Facts Facts Knowledge Knowledge acquisition acquisition

19

Towards a cooking ontology

Cooking Ingredient Ustensil Action DishType DishOrigin DishMoment ...

Knowledge modelling

20

Towards a cooking ontology

Cooking Ingredient Ustensil Action DishType DishOrigin DishMoment ...

21

Four hierarchies

DishMoment MainDish Snack Dessert DishType BakedGood Cake Muffin Sweet Pizza DishOrigin Europe Italia Mediterranean IngredientType Oil Tiramisu Asia PancakeMix BakingMix Pancake FatAndOil Fruit OliveOil Size: 10 Size: 60 Size: 4000 Size: 68

22

Four hierarchies - linked

DishMoment MainDish Snack Dessert DishType BakedGood Cake Muffin Sweet Pizza DishOrigin Europe Italia Mediterranean Tiramisu Asia Pancake Size: 10 Size: 60 Size: 68 IngredientType Oil PancakeMix BakingMix FatAndOil Fruit OliveOil Size: 4000

23

Four hierarchies - linked

DishMoment MainDish Snack Dessert DishType BakedGood Cake Muffin Sweet Pizza DishOrigin Europe Italia Mediterranean Tiramisu Asia Pancake Size: 10 Size: 60 Size: 68 IngredientType Oil PancakeMix BakingMix FatAndOil Fruit OliveOil Size: 4000

24

Four hierarchies - linked

DishMoment MainDish Snack Dessert DishType BakedGood Cake Muffin Sweet Pizza DishOrigin Europe Italia Mediterranean IngredientType Tiramisu Asia Pancake Size: 10 Size: 60 Size: 68 IngredientType Oil PancakeMix BakingMix FatAndOil Fruit OliveOil Size: 4000

25

Recipe annotation

Ingredients DishType DishOrigin DishMoment flour egg banana ... dessert cake baked-good sweet ? Banana Butterfinger Cake

Semi-automatic annotation External ressources

Recipesource: http://www.recipesource.com/

26

Ingredient annotation

Quantity Unit Modifier IngredientType 1 1/2 granulated sugar c <IN>1 ½ c sugar</IN>

Annotation by Controlled-indexing 8000 lexical variants Cooking Thesaurus

Cook's Thesaurus: http://www.foodsubs.com

27

Ingredient annotation

Quantity Unit Modifier IngredientType 2 banana c Not recognized: “about 5”

Cook's Thesaurus: http://www.foodsubs.com

mashed, ripe <IN>2 c mashed ripe bananas (about 5) </IN>

28

Type annotation

Goal: at least one dish moment and one dish type Methodology: controlled-indexing and manual validation Source: recipesource

Recipesource: http://www.recipesource.com/

29

Type annotation

• 1. From the

folder hierarchie of recipesource

• 2. Categories

available in recipesource

• 3. Terms

recognized in the title

• 4. Association

rules

• b. Manual

validation

• 5. Additionnal

annotation

• a. Help!

30

Knowledge Knowledge acquisition acquisition CBR CBR engine engine Facts Facts Interface Interface

31

Easy to use

32

Consistency validation

33

Advanced features

34

Result display features

35

Adapted recipe

36

Interface Interface Knowledge Knowledge acquisition acquisition CBR CBR engine engine Facts Facts Future Future work work

37

Test and evaluation

Knowledge acquisition process % of ingredient identification % of correctly typed recipes with each method

38

Test and evaluation

Knowledge acquisition process % of ingredient identification % of correctly typed recipes with each method CBR engine Response time (not relevant) ??

39

Test and evaluation

Knowledge acquisition process % of ingredient identification % of correctly typed recipes with each method CBR engine Response time (not relevant) ?? How to evaluate Taaable? Computer Cooking Contest

40

Taaable, a CCC participant

=

Knowledge base of the cooking domain

+

Ad-hoc CBR engine Web Interface

Compulsory task

+

Negation challenge Menu challenge (quick and dirty) Best costume challenge

41

Ongoing work

Time

Improve existing knowledge base Improve the knowledge representation features Code optimization Interactive knowledge acquisition tool Debug Interactive annotation interface Recipe sharing tool

Knowledge base CBR engine Web interface

Integrate an existing inference engine?

We leave that for the CCC 2 (or 3) !

Increase customization parameters

42