ipsj_150318

発表用

Transcript

1. ฉ ຠᏴ H27
2. Background - ‘‘Food historians look at food as one of the most important elements of culture, reflecting the social and economic structure of society.’’ - Wikipedia - Food reflects culture, society etc. as well, - Food history reflects them more. § - Prediction of evolution of food preference is also interesting and challenging ! (future work)
3. Contribution to understand the evolution of people’s food preference or do the prediction distinguish classic(traditional) recipes from those are not proposed a specific novel ranking model based on the traditionality we defined need to method
4. Outline of method data source recipes clean ranking * clean - use of existing cooking ontology * ranking - proposing two sub ranking functions reflecting our assumptions about ingredients used in traditional (classic) recipes
5. Datasets & method recipes data source : cookpad.com data sets : ingredients lists of recipes (several kinds or food’s data downloaded, for each of them there are 5k up to 20k recipes)
6. Datasets & method
7. Datasets & method data clean
8. Datasets & method data clean text processing to delete symbols with no meaning making use of the ontology to transform and classify ingredients * * more explicit, we use the synonym dict of ontology to classify and specify our ingredients
9. Datasets & method ontology synonym the synonym dict was provided by [1]’s team and we 1. specify ingredients into their upper level synonyms 2. handle issue of many different words refer to same thing Nanba, H., Takezawa, T., Doi, Y., Sumiya, K. and Tsujita, M.: Construction of a cooking ontology from cooking recipes and patents, Proc. Ubicomp 2014, pp. 507–516 (2014). reference example: Ϗʔϑ ڇ೑ ڇ͢͡ ڇ΋΋೑ … -> … -> ೑
10. Ranking we need an original base for understanding how food recipe changes from the most traditional / typical ones and ranking can meet this need to find out that we want to distinguish traditional recipes from those are not, ranking the recipes just helps we rank the recipes based on recipe’s ingredients
11. Ranking * in some cases, ranking can also meet needs (custom search etc.) of some users by supplying the specific ranking results ex. young people would like to cook for the elders who prefer classic / traditional food mothers want to know the current or previous trend of cooking some kinds of food
12. Definition for ranking 2. top frequently used ingredients of on kind of food will be regarded as common ingredients 1. among one kind of food, rare ingredient refers to ingredient which were used by few recipes
13. Assumption for ranking our proposed method are based on several assumptions as below, 1. recipes contains (more) rare ingredients will be regarded as not traditional traditional traditional not rare ex. ೑ ͡ Ό ͕
14. Assumption for ranking our proposed method are based on several assumptions as below, 2. recipes contains less common ingredients will be regarded as not traditional (typical) traditional traditional not 10 common ingredient 9 ex. ೑ ͡ Ό ͕
15. Ranking Rare ingredient disuse ratio Rare ingredient disuse ratio function 1 function 1 n(a): total number of ingredients recipe(a) contains n(f): number of recipes belong to typical kind of food d(a)i: ith ingredient of recipe(a)’s frequency among all recipes of food(f) explain in detail with real example from ೑͡Ό͕ later ex. f = ೑͡Ό͕ ex. a = one recipe n(a) log 2 n( f ) d(a i ) i n(a) ∑
16. in left case: n(a): total number of ingredients recipe(a) contains SO, n(a) of recipe in the left equals to 14 Recipe a ai n(a) log 2 n( f ) d(a i ) i n(a) ∑ a1 a2 . . . a14 * a -> recipe a ai -> ith ingredient Rare ingredient disuse ratio Rare ingredient disuse ratio Ranking function 1 function 1 d(ai)
17. in left case: so n(f) of ೑͡Ό͕ equals to 6520 n(f): total number of recipes belong to one typical kind of food Here, food f refers to ೑͡Ό͕, n(a) log 2 n( f ) d(a i ) i n(a) ∑ Rare ingredient disuse ratio Rare ingredient disuse ratio Ranking function 1 function 1 Recipe a d(ai)
18. in left case: ຯḩ was used in 302 recipes among all the recipes of ೑͡Ό ͕ɼthus d(a1) here equals to 302 d(ai): frequency of ith ingredient used among all recipes of food(f) n(a) log 2 n( f ) d(a i ) i n(a) ∑ Recipe a d(ai) d(a1) d(a2) . . . d(a14) * a -> recipe a ai -> ith ingredient Rare ingredient disuse ratio Rare ingredient disuse ratio Ranking function 1 function 1
19. Rare ingredient disuse ratio Rare ingredient disuse ratio Ranking function 1 function 1 in left case: 14 302 6520 4678 6520 3396 6520 + + + … Recipe a n(a) log 2 n( f ) d(a i ) i n(a) ∑ more rare ingredients much rarer they are + function 1 function 1 d(ai)
20. Common ingredient usage ratio Common ingredient usage ratio |Cf|: number of common ingredients of food(f) n(f): number of recipes belong to typical kind of food d(a)i: ith ingredient of recipe(a)’s frequency among all recipes of food(f) ex. f = ೑͡Ό͕ ex. a = one recipe d(a i ) n( f ) cf ∑ | C f | Ranking function 2 function 2
21. amount of ingredients of the recipe amount of the recipe Most of the recipes contain 10 ingredients, So we take top 10 ingredients as the so-called common ingredients Ranking function 2 function 2 Common ingredient usage ratio Common ingredient usage ratio ex. ೑͡Ό͕
22. For {೑͡Ό͕ nikujaga}, its common ingredients are : δϟΨΠϞ 5246 ೑ 5197 ۄೢ 4867 ে༉ 4678 ࠭౶ 4122 χϯδϯ 3978 ञ 3396 ΈΓΜ 3221 ༉ 2781 ҿྉਫ 2624 ©
23. Ranking function 2 function 2 Common ingredient usage ratio Common ingredient usage ratio 10 4678 6520 3396 6520 2624 6520 + + + … Recipe a d(a i ) n( f ) cf ∑ | C f | more common ingredients the recipe contains function 2 function 2 one common ingredient Cf
24. Datasets & method Ranking value of function 2 value of function 1 + by
25. Case Study NO. 1 / 6520
26. Case Study NO. 1500 / 6520 
    ΊΜͭΏ
    ʢPOF
    LJOE
    PG
    4PVQ
    GPS
    DPPLJOHʣ
27. Case Study NO. 3000 / 6520 
    ͍Μ͛Μ
    ʢ,JEOFZ
    CFBOTʣ
28. Case Study NO. 4500 / 6520 
    ΕΜ͜Μ
    ʢ-PUVT
    SPPUʣ 
    ͝·
    ʢ4FTBNFʣ
29. Case Study NO. 6000 / 6520
30. Case Study recipe containing only or more common ingredients will rank higher recipe containing more rare ingredients will rank lower rare ingredient plays more significant role than common ones ©
31. misc / trial ೑͡Ό͕ ೑͡Ό͕ಲ೑ ೑͡Ό͕ڇ೑ 2006 17 38 2007 188 196 2008 394 304 2009 616 778 2010 828 924 2011 1212 997 2012 2886 2209 2013 4538 2951 2014 4295 2574 0 1250 2500 3750 5000 2006 2007 2008 2009 2010 2011 2012 2013 2014 ೑͡Ό͕ಲ೑ ೑͡Ό͕ڇ೑ (without normalization)
32. misc / trial ࠭౶ in? ೑͡Ό͕ಲ೑ ೑͡Ό͕ಲ೑ ʴ࠭౶ ೑͡Ό͕ಲ೑ percentage 2000 0 0 - 2001 0 3 0 2002 0 0 - 2003 3 3 1 2004 1 1 1 2005 253 254 99.9 2006 508 538 96.2 2007 5564 5659 98.3 2008 1147 1469 78.1 2009 252 365 69.0 2010 528 729 72.4 2011 2580 2762 93.4 2012 1789 2053 87.1 2013 1248 2069 60.0 2014 378 576 65.6 0 25 50 75 100 2000 2002 2004 2006 2008 2010 2012 2014 0 0 0 1 1 99.9 96.298.3 78.1 69 72.4 93.4 87.1 60 65.6 percentage
33. Conclusion of current stage’s research 0. method with only ingredient as factor works not bad 1. recipes were ranked and classified in a relatively coarse granularity which is not enough 2. it was just the beginning
34. Fin
35. Q & A
36. Q & A Why was not ideal? score amount of ingredients of each recipe
37. Q & A score amount of ingredients of each recipe
38. Q & A classical vs traditional vs typical - Yes, now our ranking seems just to be like finding typical recipes of specific kind of food. - But Typical recipes among a long time period may be more likely to be traditional. - While classical ones means old and a bit like traditional, but may not be popular now. - Our target now is to find recipes which are traditional.