()

Size: px

Start display at page:

Download "()"

かずきにばし
5 years ago
Views:

1 NAIST-IS-MT

2 ()

3 (1) (2) (3) (1) (2) (3) (1) Web () 6, NAIST-IS-MT , i

4 (2) (3) ii

5 Implementation and Evaluation of A Recommendation Method for Dinner Recipes with Comprehensive Consideration of Multiple Factors Hiroaki Ihara Abstract In this thesis, we propose a recommender system for dinner recipes. The decision making of dinner recipes is troublesome because we need to consider (1) a lot of candidate recipes (2) by taking into account various factors, and (3) the decision occurs at every mealtime. Therefore, it would be useful if we could have a system that supports the decision making of recipes. Most of existing studies consider one or a few main recipe determinant factors, like nutrition or food preference, in their recipe recommendation systems. However, it is desirable to determine the best dinner recipe with comprehensive consideration of multiple factors such as nutritions, food preferences, foodstuff stocks, and so on. Although some existing studies take into account multiple factors in determining dinner recipes, they do not clarify how to obtain necessary information for identifying individual user s preference to each factor in candidate recipes. In this thesis, we propose a recommendation method for dinner recipes with comprehensive consideration of multiple factors and for collecting necessary information for the user s preference to each factor. To realize the proposed method, the Master s Thesis, Department of Information Science, Graduate School of Information Science, Nara Institute of Science and Technology, NAIST-IS-MT , March 13, iii

6 following three problems must be addressed: (1) how to calculate a quantitative score for each factor, (2) how to optimize the best weights among factors based on the user s sense of values or lifestyle, and (3) how to efficiently collect the necessary user information. To solve the problem (1), we suppose that the main factors considered when deciding dinner recipes consist of food preference, nutrition, calorie, ease of cooking, foodstuff stocks, and recent meal history (avoidance of continuous recommendation of the same dish). Then, we first devise a calculation method of a quantitative score for each of the six factors. The details to calculate the score of the factors are as follows: For food preference, the recipe with more of rare and frequently eating ingredients has higher score; For nutrition, the recipe with more necessary nutrition has higher score; For calorie, the recipe with calories closer to the ideal has higher score; For ease of cooking, the recipe which takes shorter time has higher score; For foodstuff stocks, the recipe which can be cooked with more of the foodstuff stocks has higher score; And for recent meal history, the recipe which has not been eaten for longer time has higher score. Also, to solve the problems (2) and (3), we devise a method for unifying six factors scores as the weighted sum. In the proposed method, we devise and adopt a method which optimizes the weighs for the six factors in short time from the user s subjective evaluation for multiple example recipes (e.g., 30 recipes). We implemented a prototype system according to the above-mentioned proposed method, and then we conducted an experiment where six subjects gave an evaluation point to each recommendation result. Through comparison of the proposed recommendation method with the recipe recommendation method considering only one factor and the random recommendation, we confirmed that the result of the proposed method is better than the other two methods where there are six subjects who evaluated that the proposed recommendation result is good, while there are only three subjects who evaluated the other methods recommendation is good. Keywords: iv

7 Cooking Recipe, Recommender System, Multiple Factors, Weighting, Personalized Service v

8 [] vi

9 vii

10 [] [] [] viii

11 UI ( ) ( 5 ) ix

12 ( 5 ) System1: System2: System3: System1: System2: System3: A B C x

13 17 D E F A B C D E F xi

14 1. Web [1] 69% (1) (2) (3) 3 [2] [3] TV [4] [5][6][7] 1

15 [8] 2 Web [1] [9] 85.4% A B () () 6 (1) (2) (3) (1) 2

16 (2) (3) 6 2 3

17 [5][6][10] [8][11] [12] [5][6][8][10][11][12] [7][13] [14] 6 [15] [16] 4

18 2.2 [17][18] [17] [19] DB [20][21] [20] 5

19 [21] 2.4 [22] [23][24] RFID [23] RFID RFID RFID [25][26] [25] [27] 1 6

20 [28] [28] [5] 7

21 Web [1] [9] [4] [3] [29] [2][30] TV Web [31] Slism [32] Web 8

22 (1) 3. (2)

24 2 UI

25 ID ( 4 ) 0,1,2 12

26 ID kcal ID 2 2 ID ID ID ID 13

27 ID ID ID 3 3 ID ID ID ID 4 ID ID ID 1.0 or or ID 14

28 4 ID ID ID 1 test L 2 test L 3 hiroaki D ID ID ID ID ID ID 1 test test test

29 3.5 R S R S R F S R N S R C S R E S R S S R L (1) C i (i = 1, 2,..., 6) S R (1) S R = C 1 S F R + C 2 S N R + C 3 S C R +C 4 S E R + C 5 S S R + C 6 S L R (1) C i 3.6 C i

30 4. () [5]

31 4.1.2 [5] i p i u i R j I j S I (R j ) (2) i I S I (R j ) = j u i p i α (2) i I j p i S I (R j ) 0 1 i I j u i p i (2) α (= u i ) (3) α = 7 c=1 c 1 c (3) i p i M i M i (4) u i t ( t = 7 ) c r c (5) p i = log 10 ( M i M ) (4) u i = t c 1 is included(i, r c ) c (5) c=1 is included(i, r) (6) 1 (i r) is included(i, r) = 0 (otherwise) (6) 18

32 [14] [] [14],, (,, ) (,,, ) (,, ) (,, ), 80kcal=1, 25 (2000kcal), 20 (1600kcal).,, 3,

33 1 3 4 i P Nut i ( Nut i (u) < 3.0 i ) X X=0,1,2,3 R S Nut (R) (7) 0.0 (X = 0) S Nut (R) = 3 i=1 P Nut i (otherwise) X i Nut i R Nut R i i P i Nut (8) 0.0 (Nut i = 0) P i Nut = Nut R i Nut i (Nut R i Nut i ) (8) 1.0 (Nut R i > Nut i ) u i Nut i (u) Nut i (9) (9) kcal 3.0 Nut i (u) (Nut i (u) 3.0) Nut i = 0.0 (Nut i (u) > 3.0) (7) (9)

34 4.3.2 C R C R R S C (R) (10) 1.0 C C R (C C R < 2C) S C (R) = (10) 0.0 (otherwise) u C u C (11) 2000 C u () 1600 C C = u () (11) (C u 400 )

35 R CT (R) R S T (R) (12) CT (R) 1.0 (CT (R) 120) 120 S T (R) = 0.0 (CT (R) > 120) (12) I R R S S (R) (13) S S (R) = k I R W (k) I R (13) I stk W (k) (14) 1 (k I stk ) W (k) = 0 (otherwise) (14) 22

36 ( ) X Y R D R S S (R) D (15) 0 (D < X) S S (R) = D X (X D Y ) (15) Y X 1 (D > Y or ) 23

39 6 / DB / 64(= 2 6 [2: / 2,6:]) DB ( )

40 S R = C 1 S F R + C 2 S N R + C 3 S C R +C 4 S E R + C 5 S S R + C 6 S L R (16) S R S R F, S R N, S R C, S R E, S R S, S R L C i (i = 1, 2,..., 6) 6 C i (i = 1, 2,..., 6) S R S R F, S R N, S R C, S R E, S R S, S R L S R, S R F, S R N, S R C, S R E, S R S, S R L S R S R F, S R N, S R C, S R E, S R S, S R L C i (i = 1, 2,..., 6) R

41 S R = ( 1.02) S F R S N R S C R S E R + ( 0.52) S S R + ( 0.39) S L R (17) 28

42 () 5 DB 29

43 ( or or ) DB

44 7 DB DB DB

45 8 32

46 9 33

48 11 ( ) 35

49 12 1 DB 4 4 system1 4 system1 system2 system3 1 system4 DB

51 :

54 ??

57 ( 5 ) 44

58 B A C E F D B F B B B

59 8 ( 5 ) 1 A B C D E F , >1.0 8 B C D

60 A B C D E F B C D A E F B C D A E F 47

61 1 2 48

63 2 50

64 [1] vol.4. php?r=site/page&view=henshubu.question_vol_4. [2]. jp/homecook. [3]. [4]. [5],,,,. Ff-irf ().. MVE,, Vol. 107, No. 454, pp , jan [6] Mayumi Ueda, Mari Takahata, and Shinsuke Nakajima. User s food preference extraction for personalized cooking recipe recommendation, [7],,,,,,.., Vol. 8, No. 4, pp. 1 6, [8],. ( )., Vol. 36, No. 8, pp , [9] Panasonic. http: //panasonic.co.jp/corp/news/official.data/data.dir/2013/04/ jn /jn html. [10],.,

65 [11],,.. 3 (DEIM), [12] Ippei Tomoda, Fumitaro Goto, Hiroshi Masui, Fumito Masui, Masakiyo Suzuki, and Yasunari Maeda. A note on supporting method for making recipe using local foods with constraints. IJCSNS, Vol. 11, No. 11, p. 49, [13],,.. SVBL, Vol. 8, pp , mar [14] gun-TOP.html. [15]. info02d.html. [16],.. Journal of Japan Society for Fuzzy Theory and Intelligent Informatics, Vol. 24, No. 1, pp , [17] Jill Freyne and Shlomo Berkovsky. Intelligent food planning: personalized recipe recommendation. In Proceedings of the 15th international conference on Intelligent user interfaces, pp ACM, [18] Gediminas Adomavicius and YoungOk Kwon. Improving aggregate recommendation diversity using ranking-based techniques. Knowledge and Data Engineering, IEEE Transactions on, Vol. 24, No. 5, pp , [19],,,.. 52

66 [20],. Graphicalrecipes:., Vol. 7, No. 2, pp , Web (DBWeb2007) [21].. 14, 2008, pp , [22],,. - -., Vol. 63, No. 3, pp , [23]. Rfid [24],,. ( ). = IEICE technical report:, Vol. 113, No. 168, pp , [25],,,,,,. ( (1),,,,, ).. MVE,, Vol. 111, No. 479, pp , [26],,,,,,. (, < > ).. A,, Vol. 94, No. 7, pp , [27],,. ( ).. AI,, Vol. 110, No. 301, pp ,

67 [28] Nguyen Tuan Ngoc,. 5s-1 ( (2),,, 50 )., Vol. 72, No. 1, [29]. [30] &. recipe/. [31]! and. eiyoukeisan.com/calorie/gramphoto/index_gram.html. [32] slism. 54

68 8. [] (or ) 3 55

69 (or ) (or )

70 57

71 9. [] System1 2 System5 6 58

72 11 System1: System2: System3: System1 system5 System A B C E F 6 59

73 12 System1: System2: System3: System1 System5 System6 A ( 5 ) B ( 5 ) C ( 5 ) E ( 5 ) F ( 5 ) G ( 5 ) H ( 5 ) I ( 5 ) ?? 12 F A 6 60

74 System5 System6 F System5 61

75 10. [] B 2013/12/ A 2013/12/ C 2013/12/ E 2013/12/ F 2013/12/ D 2013/12/ B 2013/12/ F 2013/12/ B 2013/12/ B 2013/12/ B 2013/12/

76 14 A /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ B /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/

77 16 C /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ D /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/

78 18 E /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ F /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/ /12/

79 20 A (2013/12/17) 66

80 21 B 1 22 C 1 67

81 23 D 1 24 E 1 68

82 25 F 1 69

DEIM Forum 2009 E

DEIM Forum 2009 E5-3 464-8601 1 606-8501 464 8601 1 E-mail: lifushi@arch.itc.nagoya-u.ac.jp, mayumi@mm.media.kyoto-u.ac.jp, {hirano,kajita,mase}@itc.nagoya-u.ac.jp Abstract Study on a Recipe Recommendation