Combining the best of two worlds, the “sorted
Transcription
Combining the best of two worlds, the “sorted
Combining the best of two worlds, the “sorted napping” Marine Cadoret, Sébastien Lê, Jérôme Pagès Applied mathematics department Agrocampus Ouest, France SPISE 2009 Ho Chi Minh 1City 7 August 2009 Products • 8 smoothies • 4 flavors: – Strawberry-RaspberryBlueberry (SRB) – Pineapple-Banana-Coco (PBC) – Strawberry-Banana (SB) – Mango-Passion fruits (MP) • 2 types of brands: – Distributor brand name: Casino, Carrefour – Great brand name: Immedia, Innocent 2 Holistic approaches • • • • • Categorization, napping Global approaches Require no special competences Can be used by experts or consumers Each judge uses his own criteria 3 Categorization • The judges are asked to group the products according to their resemblances Banana Sweet Strawberry Bitter Thickly Simplicity 4 Napping • The judges are asked to provide a tablecloth with the products Banana Banana Sweet Thickly Strawberry Strawberry More nuanced 5 Sorted napping • The judges are asked to provide a tablecloth with the products that have been possibly gathered Banana Sweet Thickly Strawberry Bitter 6 Method 7 The dataset: Napping X Y 5 24 Y 31 29 21 24 13 31 20 10 57 20 29 6 5 13 X 38 20 50 33 8 The dataset: Categorization C 1 2 1 2 3 2 2 3 9 The dataset: Categorization Banana Sweet Thickly C thickly strawberry-bitter thickly strawberry-bitter banana-sweet Strawberry Bitter strawberry-bitter strawberry-bitter banana-sweet10 The dataset: Sorted napping Y C 5 24 1 29 21 2 13 31 1 20 10 2 57 20 3 29 6 2 38 20 2 50 33 3 11 X The dataset: Sorted napping X Banana Thickly Strawberry Y C 5 24 thickly 29 21 strawberry 13 31 thickly 20 10 strawberry 57 20 banana 29 6 strawberry 38 20 strawberry 50 33 banana 12 The dataset • For each judge, we have: – 2 columns for the coordinates of the products on the tablecloth – 1 column for the partition X Y C X Y C X Y C 5 24 1 12 14 1 45 23 3 29 21 2 2 33 1 12 17 1 13 31 1 23 7 1 37 12 3 20 10 2 45 20 3 23 34 1 57 20 3 34 33 3 18 5 1 29 6 2 55 15 2 58 25 3 38 20 2 23 28 2 6 18 2 50 33 3 5 7 2 43 20 2 13 The dataset Judge 1 Judge j Judge J X1 Y1 C1 Xj Yj Cj XJ YJ CJ 1 i Xj (i)Yj (i)Cj (i) I 14 Nested partitions Judge 1 X1 Y1 C1 Judge j Xj Yj Judge J Cj XJ YJ CJ Let’s run Hierarchical Multiple Factor Analysis (HMFA) 15 HMFA • Takes into account the hierarchy • Balances: – The role of each judge – The role of each appoach • Classical outputs: representation of the products, etc • Specific ouputs: representation of each node, partial representation for each node 16 Results 17 6 Representation of the smoothies Carrefour_SB 2 0 Casino_SRB -2 Immedia_MP Carrefour_MP Immedia_SRB Innocent_PBC Casino_PBC -4 Dim 2 (19.4 %) 4 Innocent_SB -6 -4 -2 0 2 Dim 1 (33.17 %) 4 6 18 Representation of the words Dim 2 (19.4 %) 0 2 4 6 indeterminate_taste doughy-a_little_tasteless-slightly_sweet thick_in_the_mouth-slightly_sweet bitter unknown_taste-heavy tastelessly-neutral undefinable strong_taste_of_strawberry_raspberry strawberry-thick strawberry-acid more_soft-less_sweet mixture too_much_bitter strawberry-raspberry Carrefour_SB very_creamy-light_fruity_taste marked_taste Innocent_SB tasteless not_sweet-liquid-tasteless more_or_less_tasteless liquid-fruit_juice-unknown_taste bitterness soft_fruits-soft-sweet -4 -2 pineapple banana-strawberry-thick strawberry banana-spicy-citrus fruit Casino_SRB peach-apricot soft-tasteless-disagreeablet pulp-fruity acid soft sweet banana-coconut-soft-sweet sweet orange Carrefour_MP soft_fruits exotic exotic pulp-sweet taste_of_fruits-cold Immedia_MP soft-smooth_in_the_mouth-not_soft_fruits orange_juice strawberry-thick-sweet Immedia_SRB acid-sweet-citrus_fruit Innocent_PBCbanana-sweet-doughy very_sweet banana without_pulp-exotic I_hate creamy-soft acid-thick Casino_PBC more_or_less_sweet banana_aroma-heavy sweet-fluid coconut strawberry-good-thick banana soft_fruits-liquid-acid -6 -4 -2 0 2 4 6 banana-very_sweet-thick-soft 19 spicy thick-pure-banana-vanilla Dim 1 (33.17 %) tasteless Representation of the words Dim 2 (19.4 %) 0 2 4 6 indeterminate_taste doughy-a_little_tasteless-slightly_sweet thick_in_the_mouth-slightly_sweet bitter unknown_taste-heavy tastelessly-neutral undefinable strong_taste_of_strawberry_raspberry strawberry-thick strawberry-acid more_soft-less_sweet mixture too_much_bitter strawberry-raspberry Carrefour_SB very_creamy-light_fruity_taste marked_taste Innocent_SB tasteless not_sweet-liquid-tasteless more_or_less_tasteless liquid-fruit_juice-unknown_taste bitterness soft_fruits-soft-sweet -4 -2 pineapple banana-strawberry-thick strawberry banana-spicy-citrus fruit Casino_SRB peach-apricot soft-tasteless-disagreeablet pulp-fruity acid soft sweet banana-coconut-soft-sweet sweet orange Carrefour_MP soft_fruits exotic exotic pulp-sweet taste_of_fruits-cold Immedia_MP soft-smooth_in_the_mouth-not_soft_fruits orange_juice strawberry-thick-sweet Immedia_SRB acid-sweet-citrus_fruit Innocent_PBCbanana-sweet-doughy very_sweet banana without_pulp-exotic I_hate creamy-soft acid-thick Casino_PBC more_or_less_sweet banana_aroma-heavy sweet-fluid coconut strawberry-good-thick banana soft_fruits-liquid-acid -6 -4 -2 0 2 4 6 banana-very_sweet-thick-soft 20 spicy thick-pure-banana-vanilla Dim 1 (33.17 %) tasteless 6 Representation of the smoothies and their confidence ellipses 2 0 Casino_SRB -2 Immedia_MP Carrefour_MP Immedia_SRB Innocent_PBC Casino_PBC -4 Dim 2 (19.4 %) 4 Carrefour_SB Innocent_SB -6 -4 -2 0 2 Dim 1 (33.17 %) 4 6 21 1.0 Representation of the judges 5 4 0.6 14 0.4 12 2 21 9 8 0.2 7 6 16 22 23 20 13 3 18 24 11 1 15 19 10 0.0 Dim 2 (19.4 %) 0.8 17 0.0 0.2 0.4 0.6 Dim 1 (33.17 %) 0.8 1.0 22 Innocent_SB 10 Carrefour_SB Immedia_MP Casino_SRB Innocent_PBC Carrefour_SB Innocent_SB Casino_SRB Immedia_MP Carrefour_MP Immedia_SRB -4 0 Casino_PBC Dim 2 (19.4 %) -2 0 6 2 4 Immedia_SRB Carrefour_MP 20 30 40 Categorized tablecloth of judge 5 0 10 20 30 40 50 60 -6 -4 -2 Innocent_PBC Casino_PBC 0 2 4 Dim 1 (33.17 %) 6 23 1.0 Representation of the judges 5 4 0.6 14 0.4 12 2 21 9 8 0.2 7 6 16 22 23 20 13 3 18 24 11 1 15 19 10 0.0 Dim 2 (19.4 %) 0.8 17 0.0 0.2 0.4 0.6 Dim 1 (33.17 %) 0.8 1.0 24 Partial representations of Carrefour_SB 5 12 8 14 4 17 23 21 6 16 20 7 2 Carrefour_SB Innocent_SB 18 11 9 223 24 19 13 0 Casino_SRB Carrefour_MP Immedia_MP Immedia_SRB 10 15 1 Innocent_PBC Casino_PBC -5 Dim 2 (19.4 %) 10 5 -5 0 5 Dim 1 (33.17 %) 10 25 Sorted tablecloths of judges 8 & 22 Judge 22 40 40 Judge 8 Carrefour_SB Carrefour_MP Immedia_MP Carrefour_SB Immedia_SRB Innocent_PBC Carrefour_MP 10 Immedia_SRB Casino_SRB 10 20 Casino_PBC 30 Innocent_SB Innocent_SB 20 30 Casino_SRB Immedia_MP Innocent_PBC 0 0 Casino_PBC 0 10 20 30 40 50 60 0 10 20 30 40 50 60 26 Partial representations of Carrefour_SB 5 12 8 14 4 17 23 21 6 16 20 7 2 Carrefour_SB Innocent_SB 18 11 9 223 24 19 13 0 Casino_SRB Carrefour_MP Immedia_MP Immedia_SRB 10 15 1 Innocent_PBC Casino_PBC -5 Dim 2 (19.4 %) 10 5 -5 0 5 Dim 1 (33.17 %) 10 27 Conclusion 28 Conclusion • Sorted napping: combines the simplicity of categorization and the nuance of napping • Can be used by experts or consumers • Realizable in practice • Data analyzed by HMFA • HMFA provides useful and easily interpretable results 29 SensoMineR a package for sensory data analysis Journal of sensory studies (2008) FactoMineR: an R package for multivariate analysis Journal of statistical software (2008) http://www.agrocampus-rennes.fr/math/ 30