Recipe recommendation using ingredient networks
Chun-Yuen Teng
Yu-Ru Lin
Lada A. Adamic
School of Information
University of Michigan
Ann Arbor, MI, USA
IQSS, Harvard University
CCS, Northeastern University
Boston, MA
School of Information
University of Michigan
Ann Arbor, MI, USA
arXiv:1111.3919v1 [cs.SI] 16 Nov 2011
[email protected]
[email protected]
[email protected]
ABSTRACT
1.
The recording and sharing of cooking recipes, a human activity dating back thousands of years, naturally became an
early and prominent social use of the web. The resulting
online recipe collections are repositories of ingredient combinations and cooking methods whose large-scale and variety yield interesting insights about both the fundamentals of
cooking and user preferences. These insights include preferences for cooking methods depending on the nutritional
value extracted from food, and the geographic region from
which the recipe originates. At the level of an individual ingredient we measure whether it tends to be essential or can
be dropped or added, and whether its quantity can be modified. We also construct two types of networks to capture the
relationships between ingredients. The complement network
captures which ingredients tend to co-occur frequently, and
is composed of two large communities: one savory, the other
sweet. The substitute network, derived from user generated
suggestions for modifications, can be decomposed into many
communities of functionally equivalent ingredients, and captures users’ preference for healthier variants a recipe. Our
experiments reveal that recipe ratings can be well predicted
with features derived from combinations of ingredient networks and nutrition information.
The web enables individuals to collaboratively share knowledge and recipe websites are one of the earliest examples of
collaborative knowledge sharing on the web. Allrecipes.com,
the subject of our present study, was founded in 1997, years
ahead of other collaborative websites such as the Wikipedia.
Recipe sites thrive because individuals are eager to share
their recipes, from family recipes that had been passed down
for generations, to new concoctions that they created that
afternoon, having been motivated in part by the ability to
share the result online. Once shared, the recipes are implemented and evaluated by other users, who supply ratings
and comments.
The desire to look up recipes online may at first appear
odd given that tombs of printed recipes can be found in almost every kitchen. The Joy of Cooking [12] alone contains
4,500 recipes spread over 1,000 pages. There is, however,
substantial additional value in online recipes, beyond being able to look them up. While the Joy of Cooking contains a single recipe for Swedish meatballs, allrecipes.com
has “Swedish Meatballs I”, “II”, and “III”, submitted by different users, along with 4 other variants, including “The
Amazing Swedish Meatball”. Each variant has been reviewed, from 329 reviews for “Swedish Meatballs I” to 5
reviews for “Swedish Meatballs III”. The reviews not only
provide a crowd-sourced ranking of the different recipes, but
also many suggestions on how to modify them, e.g. using
ground turkey instead of beef, skipping the “cream of wheat”
because it is rarely on hand, etc.
The wealth of information captured by online collaborative recipe sharing sites is revealing not only of the fundamentals of cooking, but also user preferences. The cooccurrence of ingredients in tens of thousands of recipes provides information about which ingredients go well together,
and when a pairing is unusual. Users’ reviews provide clues
as to the flexibility of a recipe, and the ingredients within
it. Can the amount of cinnamon be doubled? Can the nutmeg be omitted? If one is lacking a certain ingredient, can a
substitute be found among supplies at hand without a trip
to the grocery store? Unlike cookbooks, which will contain
vetted but perhaps not the best variants for some individuals’ tastes, ratings assigned to user-submitted recipes allow
for the evaluation of what works and what does not.
In this paper, we seek to distill the collective knowlege and
preference about cooking through mining a popular recipesharing website. To extract such information, we first parse
the unstructured texts from the recipes and users’ reviews.
We construct two types of networks that reflect different re-
Categories and Subject Descriptors
H.2.8 [Database Management]: Database applications—
Data mining
General Terms
Measurement; Experimentation
Keywords
ingredient networks, recipe recommendation
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. To copy otherwise, to
republish, to post on servers or to redistribute to lists, requires prior specific
permission and/or a fee.
Copyright 20XX ACM X-XXXXX-XX-X/XX/XX ...$10.00.
INTRODUCTION
lationships between ingredients, in order to capture users’
knowledge about how to combine ingredients. The complement network captures which ingredients tend to co-occur
frequently, and is composed of two large communities: one
savory, the other sweet. The substitute network, derived
from user generated suggestions for modifications, can be
decomposed into many communities of functionally equivalent ingredients, and captures users’ preference for healthier
variants a recipe. Our experiments reveal that recipe ratings
can be well predicted by features derived from combinations
of ingredient networks and nutrition information (with accuracy .792), while most of the prediction power comes from
the ingredient networks (84%).
The rest of the paper is organized as follows. Section 2 reviews the related work. Section 3 describes the dataset and
section 4 provides descriptive analysis of the data. Section 5
discusses the extraction of ingredient complement network
and the characteristics of this network. Section 6 presents
the extraction of recipe modification information, as well as
the construction and characteristics of the ingredient substitute network. Section 7 presents our experiments on recipe
recommendation and section 8 concludes.
2.
RELATED WORK
Recipe recommendation has been the subject of much
prior work. Typically the goal has been to suggest recipes to
users based on their past recipe ratings [5][15][3] or browsing/cooking history [16]. The algorithms then find similar recipes based on overlapping ingredients, either treating
each ingredient equally [4] or according to the focus of the
recipe [20]. For example, for “grilled chicken with basil
dressing”, chicken is assigned a higher weight than basil. Instead of modeling recipes using ingredients, Wang et al. [17]
represent the recipes as graphs which are built on ingredients and cooking directions, and they demonstrate that
graph representations can be used to easily aggregate Chinese dishes by the flow of cooking steps and the sequence of
added ingredients. However, their approach only models the
occurrence of ingredients or cooking methods, and doesn’t
take into account the relationships between ingredients. In
contrast, in this paper we incorporate the likelihood of ingredients to co-occur, as well as the potential of one ingredient
to act as a substitute for another.
Another branch of research has focused on recommending recipes based on desired nutritional intake or promoting
healthy food choices. Geleijnse et al. [8] designed a prototype of a personalized recipe advice system, which suggests
recipes to users based on their past food selections and nutrition intake. In addition to nutrition information, Kamieth et
al. [10] built a personalized recipe recommendation system
based on availability of ingredients and personal nutritional
needs. Shidochi et al. [14] proposed an algorithm to extract
replaceable ingredients from recipes in order to satisfy users’
various demands, such as calorie constraints and food availability. Their method identifies substitutable ingredients by
matching the cooking actions that correspond to ingredient
names. However, their assumption that substitutable ingredients are subject to the same processing methods is less direct and specific than extracting substitutions directly from
user-contributed suggestions. In the present paper, we use
a data-driven approach to construct a substitute ingredient network, and derive clusters of substitutable ingredients
from this network.
3.
DATASET
Allrecipes is one of the most popular recipe-sharing websites, where novice and expert cooks alike can upload and
rate cooking recipes. Since its launch in 1997, it has received more than 535 milliion annual visits from over 5 million members. It hosts 16 customized international sites
for users to share their recipes in their native languages.
Recipes uploaded to the site contain specific instructions
on how to prepare a dish: the list of ingredients, preparation steps, preparation and cook time, the number of servings produced, nutrition information, serving directions, and
photos of the prepared dish. The uploaded recipes are enriched with user ratings and reviews, which comment on
the quality of the recipe, and suggest changes and improvements. In addition to rating and commenting on recipes,
users are able to save them as favorites or recommend them
to others through a forum.
We downloaded 46,337 recipes including all information
listed from allrecipes.com, including several classifications,
such as a region (e.g. the midwest region of US or Europe), the course or meal the dish is appropriate for (e.g.:
appetizers or breakfast), and any holidays the dish may be
associated with. In order to understand users’ recipe preferences, we crawled 1,976,920 reviews which include reviewers’
ratings, review text, and the number of users who voted the
review as useful. We further downloaded information on the
530,609 users who reviewed and rated recipes. This information includes links to users’ pages, their interests, the date
they joined the site, their cooking experience level, home
town, and current city they live in.
3.1
Data preprocessing
The first step in processing the recipes is identifying the
ingredients listed. Matching on predefined lists of ingredients often missed or misidentified ingredients commonly
supplied by users. We therefore derived the list of ingredients from the recipes themselves through the following procedure. We removed quantifiers, such as e.g. “1 lb” or “2
cups”, words referring to consistency or temperature, e.g.
chopped or cold, along with a few other heuristics, such as
removing content in parentheses. For example “1 (28 ounce)
R
can baked beans (such as Bush’s Original)”
is identified as
“baked beans”. We erred on the side of not conflating potentially identical or highly similar ingredients, e.g. “cheddar
cheese”, used in 2450 recipes, was considered different from
“sharp cheddar cheese”, occurring in 394 recipes.
We then generated an ingredient list sorted by frequency
of ingredient occurrence and selected the top 1000 common
ingredient names as our finalized ingredient list. Each of the
top 1000 ingredients occurred in 23 or more recipes, with
plain salt making an appearance in 21,916 recipes. These
ingredients also accounted for 94.9% of ingredient entries in
the recipe dataset. The remaining ingredients were missed
either because of high specificity (e.g. yolk-free egg noodle),
referencing brand names (e.g. Planters almonds), rarity (e.g.
serviceberry), misspellings, or not being a food (e.g. “nylon
netting”).
The remaining processing task involved identifying cooking processes from the directions. We first identified all heating methods using a listing in the Wikipedia entry on cooking [18]. For example, baking, boiling, and steaming are
all ways of heating the food. We then identified mechanical
ways of processing the food such as chopping and grinding,
and other chemical techniques such as marinating and brining.
DESCRIPTIVE ANALYSIS
One of the interesting aspects about this dataset is that
it allows us to obtain a large-scale view of cooking methods.
Here we discuss how different cooking methods vary with
user and regional preference.
4.1
fry
bake
It has been suggested that cooking played a significant role
in human evolution by allowing us to extract more energy
value from food [19]. An experiment measuring energy expended by the Burmese python in digesting meat has shown
that processing the meat by grinding and cooking individually reduces the digestive cost to the snake, and combining both processing methods reduces energy cost more than
each method individually [1]. Interestingly, it appears that
average recipe ratings correlate with the ability of the processing method to reduce digestive cost. Table 4.1 shows
that recipes that call for cooking food have higher ratings
than ones that merely break it down mechanically, which in
turn are rated more highly than ones that simply “mix” or
“toss” ingredients together. Furthermore, we observe that
recipes with additional chemical processing, e.g. fermenting and marinating, tend to receive higher ratings than ones
preparing the food with only heating and mechanical methods. However, perhaps due to the additional time and planning they require, they occur in only about 8% of the recipes
in the dataset.
Table 1: Occurrence and average ratings of cooking
methods
occurrence
34759
40238
3686
average rating
3.60
4.11
4.14
The preference for multiple food processing methods might
at first be interpreted as a reflection of the sophistication
of the recipe, with more complex recipes rated more highly.
However, in general we find no correlation between the number of steps or the number of ingredients and the average
rating a recipe receives, making it more likely that the digestibility of the prepared food is a factor in how highly
rated it is.
4.2
grill
roast
Why we cook
Mechanical methods
Heating methods
Chemical methods
west-coast
south
northeast
mountain
midwest
marinate
method
4.
boil
simmer
Regional preferences
While cooking methods that make food more digestible
tend to be preferred, choosing one method over another
appears to be a question of regional taste. About 5.8%
(n=2693) of recipes were classified into one of 5 US regions:
Midwest, Northeast, South, West Coast (including Alaska
and Hawaii), and Mountain. Figure 1 shows significantly
(χ2 test p-value < 0.001) varying preferences in the different US regions among 6 of the most popular cooking methods. Boiling and simmering, both involving heating food in
hot liquids, are more common in the South and Midwest.
Marinating and grilling are relatively more popular in the
West and Mountain regions, but in the West more grilling
recipes involve seafood (18/42 = 42%) relative to other re-
0
10
20
30
40
% in recipes
Figure 1: The percentage of recipes by region that
apply a specific cooking recipe.
gions combined (7/106 = 6%). Frying is popular in the
South and Northeast. Baking is a universally popular and
versatile technique, which is often used for both sweet and
savory dishes, and is slightly more popular in the Northeast
and Midwest. Examination of individual recipes reflecting
these frequencies shows that these differences can be tied
to differences in demographics, immigrant culture and availability of local ingredients, e.g. seafood.
5.
INGREDIENT COMPLEMENT NETWORK
Can we learn how to combine ingredients from the data?
Here we employ the occurrences of ingredients across recipes
to learn users’ knowledge about combinations of ingredients.
We constructed an ingredient complement network based
on pointwise mutual information (PMI) defined on pairs of
ingredients (a, b):
pmi(a; b) = log
p(a, b)
,
p(a)p(b)
where
p(a, b) =
# of recipes containing a and b
,
# of recipes
p(a) =
# of recipes containing a
,
# of recipes
p(b) =
# of recipes containing b
.
# of recipes
The PMI gives the probability that two ingredients occur
together against the probability that they occur separately.
Complementary ingredients tend to occur together far more
often than would be expected by chance.
Figure 2 shows a visualization of ingredient complementarity. Two distinct subcommunities of recipes are immediately apparent: one corresponding to savory dishes, the
other to sweet ones. Some central ingredients, e.g. egg and
salt, actually are pushed to the periphery of the network.
They are so ubiquitous, that although they have many edges,
they are all weak, since they don’t show particular complementarity with any single group of ingredients.
We further probed the structure of the complementarity
network by applying a network clustering algorithm. The
algorithm confirmed the existence of two main clusters containing the vast majority of the ingredients. An interesting
satellite cluster is that of ingredients for mixed drinks, which
tiger prawn
lobster tail
sea salt black pepper
artichoke
greek yogurt
kosher salt black pepper
root beer
white mushroom
haddock
button mushroom
goat cheese
salt black pepper
port wine
watercres
sea scallop
triple sec
sour mix
sweet
white rum
club soda
butter
cranberry juice
ice
pomegranate juice
pink lemonade
banana liqueur
shallot
juiced
tequila
smoked ham
chocolate ice cream
asparagus
brie cheese
watermelon
hazelnut
orange juice
eggnog
maraschino cherry juice
lemon juice
juice
angel food cake mix
superfine sugar
plum
white chocolate
artificial sweetener
semisweet chocolate
chocolate coffee
cake flour
raspberry jam
hazelnut liqueur
cocoa powder
almond paste
creme de menthe liqueur
milk chocolate
vanilla wafer
peach
cantaloupe
pie shell
pistachio nut
bourbon whiskey
vanilla yogurt
blackberry
fig
golden syrup
kiwi
banana
pear
prune
chocolate wafer
candied cherry
red candied cherry
apricot jam
apple juice
raspberry gelatin mix
currant
orange gelatin
strawberry gelatin mix
tapioca
confectioners' sugar walnut coconut raisin
whipped topping
peppermint candy
turbinado sugar
pie crust
cream of tartar
german chocolate
flour
baking soda
strawberry preserve
yellow food coloring
green candied cherry
pistachio pudding mix
candied pineapple
coffee powder
vanilla
extract
semisweet chocolate chip
maple extract
white chip
chocolate chip
devil's food cake mix
vanilla frosting
low fat peanut butter
chocolate cookie crust
lemon gelatin mix
crisp rice cereal
marshmallow
fruit
brownie mix
unpie crust
unbleached flour
applesauce
solid pack pumpkin
flax seed
sugar free vanilla pudding mix
oat bran
butterscotch pudding mix
spice cake mix
skim milk
orange gelatin mix
teriyaki sauce
yeast
coleslaw mix
distilled white vinegar
sunflower kernel
matzo meal
pie filling
barley nugget cereal
wheat
cream cheese
wheat bran
beaten egg
sourdough starter
non fat milk powder
neufchatel cheese
pretzel
chocolate pudding
baker's semisweet chocolate
decorating gel
cook
low fat margarine
brick cream cheese
cornflakes cereal
crescent dinner roll
pancake mix
buttermilk baking mix
celery
white rice pork
imitation crab meat
beer
cornmeal
low fat cheddar cheese
ranch dressing
corn
green bean
spiral pasta
salt
italian dressing mix
whole wheat bread
cornflake
olive
kidney bean
white corn
vinegar
biscuit baking mix
ketchup
pickle relish
crescent roll
butter cooking spray
potato chip
dill pickle
bean
barbeque sauce
rye bread
butter cracker
green chile
baby pea
chili seasoning mix
spicy pork sausage
sausage
brown mustard
colby monterey jack cheese
stuffing
picante sauce
turkey gravy
cheese
lean beef
ranch bean
macaroni
taco seasoning
taco sauce
elbow macaroni
kernel corn
catalina dressing
ham
onion
whole wheat tortilla
tomato vegetable juice cocktail
corn tortilla chip
green chily
mexican cheese blend
butter bean
stuffed olive
egg noodle
mild cheddar cheese
colby cheese
beef gravy
cream of mushroom soup
corn bread mix
sour cream
vidalia onion
taco seasoning mix
french onion soup
processed cheese
stuffing mix
barbecue sauce
cream corn
biscuit mix
bread stuffing mix
buttermilk biscuit
onion salt
cream of chicken soup
sourdough bread
chili without bean
tuna
curd cottage cheese
monterey jack cheese
refried bean
enchilada sauce
lima bean
garlic salt
steak sauce
yellow mustard
mustard
pimento pepper
ranch dressing mix
french dressing
dill pickle relish
sauerkraut
corned beef
thousand island dressing
vegetable combination
corn chip
tortilla chip
pickled jalapeno pepper
guacamole
beef chuck
powder
chunk chicken breast
pepperjack cheese
kaiser roll
pimento
pickle
bacon grease
hoagie roll
corkscrew shaped pasta
tomato juice
flour tortilla salsa
english muffin
blue cheese dressing
pepperoni sausage
pizza sauce
chili bean
mixed vegetable
onion flake
seasoning salt
pimiento
onion soup mix
pepper
pizza crust
bread dough
chuck roast
wax bean
roast beef
beef consomme
wild rice mix
corn tortilla
brown gravy mix
cream of potato soup
dill pickle juice
saltine cracker
biscuit
bratwurst
round steak
golden mushroom soup
sandwich roll
white bread
apple jelly
baking mix
black olive
beef bouillon
pinto bean
parsley flake
meat tenderizer
vegetable soup mix
crescent roll dough
dressing
marinara sauce
spaghetti sauce
salami
pepperoni
tomato sauce
potato
green bell pepper
venison
broccoli floweret
cottage cheese
liquid smoke
cracker
zesty italian dressing
red kidney bean
smoked sausage
worcestershire
sauce
chicken
spicy brown mustard
part skim ricotta cheese
lasagna noodle
spaghetti
salt free seasoning blend
polish sausage
swiss cheese
provolone cheese
seashell pasta
bacon dripping
steak
onion separated
vegetable cooking spray
seasoning
horseradish
pork chop
vegetable
italian bread
italian salad dressing
great northern bean
long grain
salad green
buttery round cracker
saltine
adobo seasoning
noodle
toothpick
crouton
dill seed
bagel
grape jelly
part skim mozzarella cheese
louisiana hot sauce
black bean
navy bean
cornbread
beef brisket
mexican corn
pasta sauce
basil sauce
manicotti shell
red bean
seafood seasoning
browning sauce
chicken bouillon
iceberg lettuce
italian sauce
ziti pasta
meatless spaghetti sauce
turkey breast
lean turkey
lettucechili sauce
pizza crust dough
cheese ravioli
tomato
mozzarella cheese
white hominy
baby carrot
barley
beef broth
green pea
poultry seasoning
kielbasa sausage
monosodium glutamate
chile sauce
alfredo sauce
mild italian sausage
pasta shell
tube pasta
tomato paste
fajita seasoning
beef stew meat
paprika chili powder
garlic
turkey
hot pepper sauce
broccoli
mayonnaise
bacon bread
old bay seasoning tm
mustard powder
country pork rib
pasta
alfredo pasta sauce
italian sausage
white potato
nutritional yeast
rump roast
black eyed pea
veal
beef chuck roast
pepper jack cheese
celery salt
mixed nut
creole seasoning
okra
red potato
smoked paprika
long grain rice
lean pork
beef round steak
sugar based curing mixture
romano cheese
rotini pasta
banana pepper
pimento stuffed green olive
honey mustard
cabbage
cocktail rye bread
herb stuffing mix
popped popcorn
pork shoulder roast
chicken soup base
lump crab meat
hot sauce
onion powder
celery seed
herb bread stuffing mix
yellow summer squash
caesar dressing
lentil
marjoram
beef sirloin
bacon bit
cocktail sauce
fat free sour cream
pork sparerib
miracle whip ‚Ñ
potato flake
yellow cornmeal
milk
margarine
cereal
egg
candy
dill
ditalini pasta
rigatoni pasta
ricotta cheese
pearl barley
ham hock
green salsa
chive
steak seasoning
cider vinegar
caraway seed
chow mein noodle
bread flour
crab meat
broiler fryer chicken up
herb stuffing
savory
meatball
jalapeno pepper
cauliflower
sirloin steak
low fat sour cream
oil
flat iron steak
pearl onion
chicken leg quarter
cod
cauliflower floret
lemon pepper seasoning
oyster
wild rice
catfish
apple cider vinegar
white vinegar
unpie shell
lemon pepper
pork loin chop
water chestnut
pickling spice
yellow squash
chorizo sausage
fat free italian dressing
beef stock
cajun seasoning
puff pastry shell
fat free mayonnaise
egg substitute
non fat yogurt
cooking spray
rapid rise yeast
vegetable bouillon
hungarian paprika
french bread
parmesan cheese
spinach
artichoke heart
russet potato
flounder
caulifloweret
pork loin roast
molasse
yellow onion
yellow pepper
poblano pepper
crawfish tail
radishe
low fat mayonnaise
fat free cream cheese
vital wheat gluten
italian cheese blend
cannellini bean
burgundy wine
pork shoulder
broccoli floret
beet
green beans snapped
chicken liver
white onion
beef sirloin steak
green chile pepper
cheese tortellini
fusilli pasta
fat free chicken broth
marinated artichoke heart
andouille sausage
white cheddar cheese
chicken wing
giblet
chicken bouillon powder
white wine vinegar
romaine
beef short rib
cumin
cayenne pepper sage
cooking oil
mustard seed
salmon steak
pumpkin seed
rye flour
bread machine yeast
oatmeal
nilla wafer
vanilla
blue cheese
half and half
unpastry shell
topping
garlic paste
egg roll wrapper
sunflower seed
whole wheat flour
powdered milk
sugar cookie mix
food coloring
ramen noodle
fat free evaporated milk
chutney
black pepper
french baguette
white bean
chicken breastmushroom
scallion
chicken thigh
sesame seed
cashew
softened butter
cherry gelatin
milk powder
brown sugar
crispy rice cereal
low fat yogurt
red grape
fat free yogurt
basil pesto
pre pizza crust
oregano
chicken broth
italian seasoning
bay
tomatillo
avocado
low fat
canola oil
dijon mustard
acorn squash
whole milk
red apple
pineapple chip
peanut poppy seed
lime gelatin mix
wheat germ
lemon gelatin
baking apple
peanut butter
vanilla pudding
german chocolate cake mix
maple syrup
tart apple
anise seed
turmeric
garam masala
lobster
tapioca flour
butterscotch chip
firmly brown sugar
ginger paste
chicken ramen noodle
pumpkin pie spice
caramel
chocolate mix
butter shortening
honey
asafoetida powder
green tomato
mixed fruit
caramel ice cream topping
candy coated milk chocolate
milk chocolate chip
mango chutney
pita bread
chipotle pepper
cucumber
pesto
escarole
white kidney bean
kale
clam
poblano chile pepper
clam juice
red pepper
brown rice
white pepper
ginger garlic paste
wonton wrapper
serrano pepper
green lettuce
baby corn
salad shrimp
curry powder
fruit gelatin mix
apple pie spice
whipped topping mix
marshmallow creme
orange marmalade
low fat cream cheese
cranberry sauce
lite whipped topping
low fat whipped topping
jellied cranberry sauce
individually wrapped caramel
candy coated chocolate
apple pie filling
maple flavoring
mixed berry
rice flour
lime gelatin
recipe pastry
whole wheat pastry flour
chocolate cake mix
cream of shrimp soup
green grape
ring
pastry
apple
peppercorn
green apple
apricot preserve
soy milk
potato starch
lemon cake mix
lemon pudding mix
nut
lard
peanut butter chip
vegetable shortening
toffee baking bit
lemon peel
lemon yogurt
berry cranberry sauce
sour milk
pumpkin
1% buttermilk
evaporated milk
baking cocoa
corn syrup
apple butter
milk chocolate candy kisse
powdered fruit pectin
cornstarch
cinnamon water
vegetable oil
oat sugar
persimmon pulp
blueberry pie filling
raspberry preserve
cinnamon sugar
raspberry gelatin
allspice
mace
gingersnap cooky
strawberry gelatin
white cake mix
cherry pie filling
strawberry jam
any fruit jam
black walnut
coconut extract
shortening
pecan
anise extract
orange peel
yellow cake mix
butterchocolate
extract cookie
bourbon
baking chocolate
rhubarb
self rising flour
graham cracker
buttermilk
date
powdered non dairy creamer
white chocolate chip
chocolate pudding
mix
chocolate frosting
candied citron
fruit cocktail
cinnamon
red candy
chocolate sandwich
cooky
lemon extract
golden delicious apple
chicken drum
red lentil
panko bread
habanero pepper
snow pea
bamboo shoot
low sodium soy sauce
cumin seed
fenugreek seed
curry
cooking sherry
romaine lettuce
fennel seed
oyster sauce
ghee
spaghetti squash
eggplant
bow tie pasta
plum tomato
bell pepper
thyme
garbanzo bean
farfalle pasta
brown lentil
bay scallop
carrot
gingerroot
coriander seed
bean sprout
smoked salmon
basmati rice
anchovy
angel hair pasta
green olive
chicken breast half
chickpea
rutabaga
cream cheese spread
vegetable stock
parsley
red wine
red wine vinegar
muenster cheese
red snapper
saffron thread
herb
prosciutto
collard green
green cabbage
fish stock
round
fontina cheese
basil
zucchini
vermicelli pasta
asiago cheese
linguine
low sodium beef broth
vegetable broth
shrimp
low sodium chicken broth
pork loin
beef flank steak
hoisin sauce
yogurt
soy sauce
cardamom
fettuccini pasta
parsnip
pita bread round
tarragon vinegar
turnip
cilantro
coriander
black peppercorn
red chile pepper
allspice berry
sugar pumpkin
cardamom pod
splenda
clove
mandarin orange
silken tofu
peppermint extract
hot
red lettuce
jalapeno chile pepper
adobo sauce
brussels sprout
seed
linguini pasta
orzo pasta
penne pasta
roma tomato
caper
cherry tomato
leek
phyllo dough
ears corn
halibut
sugar snap pea
chipotle chile powder
bok choy
chinese five spice powder
ginger
grape
walnut oil
granny smith apple
red delicious apple
candied mixed fruit peel
pastry shell
salt pepper
tofu
napa cabbage
rice wine vinegar
stuffed green olive
tarragon
red pepper flake
red cabbage
sherry
rice wine
short grain rice
raspberry vinegar
sweet potato
crystallized ginger
apricot nectar
golden raisin
mixed spice
food cake
orangeangel
extract
rum extract
mixed salad green
apple cider
apricot
egg white pineapplecranberry
vanilla pudding mix
miso paste
asian sesame oil
rice vinegar
white grape juice
rose water
balsamic vinaigrette dressing
chicken stock
pork tenderloin
sesame oil
fish sauce
beef tenderloin
saffron
flank steak
curry paste
jasmine rice
chile paste
rice noodle
grapefruit
low fat milk
orange zest
nectarine
pound cake
gruyere cheese
serrano chile pepper
low fat cottage cheese
red curry paste
lemon gras
peanut oil
fettuccine pasta
swiss chard
creme fraiche
pancetta bacon
debearded
squid
lamb
chile pepper
pork roast
kaffir lime
butternut squash
mirin
ginger root
coconut milk
tahini
spanish onion
scallop
mussel
arborio rice
rosemary
red onion
bulgur
salmon
portobello mushroom
new potato
red bell pepper
linguine pasta
tamari
sake
kalamata olive
feta cheese
asparagu
marsala wine
quinoa
corn oil
chili oil
whipping cream
graham cracker crust
almond
extract
macadamia nut
puff pastry
cream
lime peel
baking powder nutmeg almond
cocoa
red food coloring
green food coloring
lime zest
whiskey
star anise pod
strawberry
mandarin orange segment
cherry
blueberry
yam
tea bag
semolina flour
raspberry
key lime juice
lemon zest brandy
orange sherbet
cola carbonated beverage
heavy whipping cream
gelatin
grape juice
cream of coconut
amaretto liqueur
vanilla ice cream
chocolate syrup
coffee liqueur
ladyfinger
lime juice
mango
skewer
wooden skewer
chicken leg
portobello mushroom cap
crimushroom
yukon gold potato
cracked black pepper
shiitake mushroom
jicama
couscou
heavy cream
honeydew melon
orange liqueur
vanilla bean
white sugar
mascarpone cheese
pine nut
tilapia
cornish game hen
kosher salt
papaya
zested
orange
greek seasoning
english cucumber
coconut oil
malt vinegar
brandy based orange liqueur
pineapple ring
coconut cream
egg yolk
chocolate hazelnut spread
irish cream liqueur
bittersweet chocolate
balsamic vinegar
orange roughy
lime mint
sauce
lemon lime carbonated beverage
champagne
sour cherry
garlic
gorgonzola cheese
sea salt
baby spinach
grapefruit juice
vodka
lemonade
pineapple juice
spiced rum
rum
maraschino cherry
vanilla vodka
lemon
simple syrup
carbonated water
fat free half and half
white balsamic vinegar
fennel
irish stout beer
italian parsley
tuna steak
vermouth
gin
limeade
triple sec liqueur
ginger ale
butterscotch schnapp
olive oil
grape tomato
chestnut
leg of lamb
melon liqueur
coconut rum
grenadine syrup
lemon lime soda
peach schnapp
arugula
white wine
trout
process cheese sauce
cream of celery soup
hamburger bun
processed cheese food
baking potato
process american cheese
pork sausage
creamed corn
canadian bacon
chili
sharp cheddar cheese
french green bean
cheddar cheese soup
beef
tomato soup
processed american cheese
cheddar cheese
biscuit dough
process cheese
american cheese
hash brown potato
chunk chicken
corn muffin mix
tomato based chili sauce
hot dog
tater tot
grit
hot dog bun
dinner roll
Figure 2: Ingredient complement network. Two ingredients share an edge if they occur together more than
would be expected by chance and if their pointwise mutual information exceeds a threshold.
6.
RECIPE MODIFICATIONS
Co-occurrence of ingredients aggregated over individual
recipes reveals the structure of cooking, but tells us little
about how flexible the ingredient proportions are, or whether
some ingredients could easily be left out or substituted. An
experienced cook may know that apple sauce is a low-fat alternative to oil, or may know that nutmeg is often optional,
but a novice cook may implement recipes literally, afraid
that deviating from the instructions may produce poor results. While a traditional hardcopy cookbook would provide
few such hints, they are plentiful in the reviews submitted
by users who implemented the recipes, e.g. “This is a great
recipe, but using fresh tomatoes only adds a few minutes to
the prep time and makes it taste so much better”, or another
comment about the same salsa recipe “This is by far the best
recipe we have ever come across. We did however change it
just a little bit by adding extra onion.”
As the examples illustrate, modifications are reported even
when the user likes the recipe. In fact, we found that 60.1%
of recipe reviews contain words signaling modification, such
as “add”,“omit”,“instead”,“extra” and 14 others. Further-
0.6
0.1
0.2
0.3
0.4
0.5
no modification
with modification
0.0
evident as a constellation of small nodes located near the
top of the sweet cluster in the visualization of Figure 2. The
cluster includes the following ingredients: lime, rum, ice,
orange, pineapple juice, vodka, cranberry juice, lemonade,
tequila, etc.
For each recipe we examined the minimum, average, and
maximum pairwise pointwise mutual information between
ingredients. The intuition is that complementary ingredients would yield higher ratings, while ingredients that don’t
go together would lower the average rating. We found that
while the average and minimum pointwise mutual information between ingredients is uncorrelated with ratings, the
maximum is very slightly positively correlated with the average rating for the recipe (ρ = 0.09, p-value < 10−10 ). This
suggests that having at least two complementary ingredients
very slightly boosts a recipe’s prospects, but having clashing
or unrelated ingredients does not seem to do harm.
proportion of reviews with given rating
luncheon meat
1
2
3
4
5
rating
Figure 3: The modifiability of ingredients. The line
represents equal number of occurrences where the
reviews suggested to increase as opposed to increase
the amount of the ingredient in the dish.
more, it is the reviews that include changes that have a
statistically higher average rating (4.49 vs. 4.39, t-test pvalue < 10−10 ), and lower rating variance (0.82 vs. 1.05,
Bartlett test p-value < 10−10 ), as is evident in the distribution of ratings, shown in Fig. 3. This suggests that flexibility
in recipes is not necessarily a bad thing, and that reviewers
who don’t mention modifications are more likely to think of
the recipe as perfect, or to dislike it entirely.
In the following, we describe the recipe modifications extracted from user reviews, including adjustment, deletion
and addition. We then present how we constructed an ingredient substitute network based on the extracted information.
6.1
Adjustments
Some modifications involve increasing or decreasing the
1.00
soy sauce
0.20
chicken breast
milk
carrot
sour cream
tomato
flour
brown sugar
basil
butterwhite sugar
onionnutmeg water
celery
mayonnaise
sugar
oregano
cs’. sugar black pepper
egg
salt
walnut
baking powder
pepper
olive oil
green bell pepper
baking soda
0.02
0.05
0.10
vanilla extract
pecan
parsley
shortening
vegetable oil
0.01
(# reviews adjusting up)/(# recipes)
0.50
garlic
chicken broth
cheddar cinnamon
bacon
chocolate chip honey
mushroom
parmesan cornstarch cream cheese
worcestershire s.
potato
lemon juice
garlic powder
margarine
0.01
0.02
0.05
0.10
0.20
0.50
1.00
(# reviews adjusting down)/(# recipes)
Figure 4: Modifications to the 50 most common ingredients, derived from recipe reviews. The line
denotes equal numbers of suggested increases and
decreasess.
amount of an ingredient in the recipe. In this and the following analyses, we split the review on punctuation such
as commas and periods. We used simple heuristics to detect when a review suggested a modification: adding/using
more/less of an ingredient counted as an increase/decrease.
Doubling or increasing counted as an increase, while reducing, cutting, or decreasing counted as a decrease. While it is
likely that there are other expressions signaling the adjustment of ingredient quantities, using this set of terms allowed
us to compare the relative rate of modification, as well as
the frequency of increase vs. decrease between ingredients.
The ingredients themselves were extracted by performing a
maximal character match within a window following an adjustment term.
Figure 4 shows the ratios of the number of reviews suggesting modifications, either increases or decreases, to the
number of recipes that contain the ingredient. Two patterns
are immediately apparent. Ingredients that may be perceived as being unhealthy, such as fats and sugars, are, with
the exception of vegetable oil and margarine, more likely
to be modified, and to be decreased. On the other hand,
flavor enhancers such as soy sauce, lemon juice, cinnamon,
Worcestershire sauce, and toppings such as cheeses, bacon
and mushrooms, are also likely to be modified; however, they
tend to be added in greater, rather than lesser quantities.
Combined, the patterns suggest that good-tasting but “unhealthy” ingredients can be reduced, if desired, while spices,
extracts, and toppings can be increased to taste.
6.2
Deletions and additions
Recipes are also frequently modified such that ingredients
are omitted entirely. We looked for words indicating that
the reviewer did not have an ingredient (and hence did not
use it), e.g. “had no” and “didn’t have”. We further used
“omit/left out/left off/bother with” as indication that the
reviewer had omitted the ingredients, potentially for other
reasons. Because reviewers often used simplified terms, e.g.
“vanilla” instead of “vanilla extract”, we compared words in
proximity to the action words by constructing 4-character-
grams and calculating the cosine similarity between the ngrams in the review and the list of ingredients for the recipe.
To identify additions, we simply looked for the word “add”,
but omitted possible substitutions. For example, we would
use “added cucumber”, but not “added cucumber instead of
green pepper”, the latter of which we analyze in the following section. We then compared the addition to the list of
ingredients in the recipe, consider the addition valid only if
the ingredient does not already belong in the recipe.
Table 6.2 shows the correlation of the ingredient modifications. As might be expected, the more frequently an ingredient occurs in a recipe, the more times its quantity has the
opportunity to be modified, as is evident in the strong correlation between the recipe frequency and both increases and
decreases recommended in reviews. However, if we take the
proportion of modifications to the number of recipes the ingredient appears in, these are typically negatively correlated
with the frequency of the ingredient, e.g. deletions/recipe
with ρ = −0.22, additions ρ = −0.25, increases ρ = −0.26.
For example, salt is so essential, appearing in over 21,000
recipes, that we detected only 18 reviews where it was explicitly dropped. In contrast, Worcheshire sauce, appearing
in 1,542 recipes, is dropped explicitly in 148 reviews.
As might also be expected, additions are positively correlated with increases, and deletions with decreases. However,
additions and deletions are very weakly negatively correlated, indicating that an ingredient that is added frequently
is not necessarily omitted more frequently as well.
Table 2: Correlations between ingredient modifications
recipes
addition
deletion
increase
6.3
addition
0.41
deletion
0.22
-0.15
increase
0.61
0.79
0.09
decrease
0.68
0.11
0.58
0.39
Ingredient substitute network
Replacement relationships show whether one ingredient
is preferable to another. The preference could be based
on taste, availability, or price. Some ingredient substitution tables can be found online 1 , but are neither extensive
nor contain information about relative frequencies of each
substitutuion. Thus, we found an alternative source for extracting replacement relationships – users’ comments, e.g.
“I replaced the butter in the frosting by sour cream, just to
soothe my conscience about all the fatty calories”.
To extract such knowledge, we first parsed the reviews
as follows: we considered several phrases to signal replacement relationships: “replace a with b”, “substitute a with
b”, “a instead of b”, etc, and matched a and b to our list of
ingredients.
We constructed an ingredient substitute network to capture users’ knowledge about ingredient replacement. This
weighted, directed network consists of ingredients as nodes.
We thresholded and eliminated any suggested substitutions
that occurred fewer than 5 times. We then determined the
weight of each edge by p(b|a), the proportion of substitu1
e.g.,
http://allrecipes.com/HowTo/common-ingredientsubstitutions/detail.aspx
Table 3: Clusters of ingredients that can be substituted for one another. A maximum of 5 additional
ingredients for each cluster are listed, ordered by
PageRank.
main
chicken
olive oil
sweet
potato
baking
powder
almond
apple
egg
tilapia
spinach
italian
seasoning
cabbage
Figure 5: The network of ingredient substitution.
Nodes are sized according to the number of times
they have been recommended as a substitute for another ingredient in reviews, and colored according to
their indegree.
other ingredients
turkey, beef, sausage, chicken breast, bacon
butter, apple sauce, oil, banana, margarine
yam, potato, pumpkin, butternut squash,
parsnip
baking soda, cream of tartar
pecan, walnut, cashew, peanut, sunflower s.
peach, pineapple, pear, mango, pie filling
egg white, egg substitute, egg yolk
cod, catfish, flounder, halibut, orange roughy
mushroom, broccoli, kale, carrot, zucchini
basil, cilantro, oregano, parsley, dill
coleslaw mix, sauerkraut, bok choy
napa cabbage
vegetable shortening,..
pumpkin seed,..
lemon cake mix,..
baking powder,..
dijon mustard,..
black olive,..
golden syrup,..
lemonade,..
graham cracker,..
coconut milk,..
almond extract,..
vanilla,..
peach schnapp,..
cranberry,..
strawberry,..
almond,..
milk,..
lemon juice,..
cinnamon,..
apple juice,..
bread,..
corn chip,..
chocolate chip,..
olive oil,..
apple,..
sour cream,..
white wine,.. champagne,..
flour,..
cottage cheese,..
egg,..
chicken broth,..
garlic,.. sauce,..
hoagie roll,..
tions of ingredient a that suggest ingredient b. For example,
68% of substitutions for white sugar were to splenda, an
artificial sweetener, and hence the assigned weight for the
sugar → splenda edge is 0.68. The resulting network is
shown in Figure 5.
The substitution network shown in Fig. 5 exhibits strong
clustering. We examined this structure by applying the map
generator tool by Rosvall et al. [13], which uses a random
walk approach to identify clusters in weighted, directed networks. The resulting clusters, and their relationships to one
another, are shown in Fig. 6. The derived clusters could be
used when following a relatively new recipe which may not
have many reviews, and therefore many suggestions for ingredient substitutions. If one does not have all ingredients
at hand, one could examine the content of one’s fridge and
panty and match it with other ingredients found in the same
cluster as the ingredient called for by the recipe. Table 6.3
lists the contents of a few such sample ingredient clusters,
and Fig. 7 shows two example clusters extracted from the
substitute network.
Finally, we examined whether the substitution network
encodes preferences for one ingredient over another, as evidenced by the relative ratings of similar recipes, one which
contains an original ingredient, and another which implements a substitution. To test this hypothesis, we construct
a “preference network”, where one ingredient is preferred to
another in terms of received ratings, and is constructed by
creating an edge (a, b) between a pair of ingredients, where
a and b are listed in two recipes X and Y respectively, if
recipe ratings RX > RY . For example, if recipe X includes
beef, ketchup and cheese, and recipe Y contains beef and
pickles, then this recipe pair contributes to two edges: one
from pickles to ketchup, and the other from pickles to cheese.
The aggregate edge weights are defined based on PMI. Because PMI is a symmetric quantity (pmi(a; b) = pmi(b; a)),
honey,..
pie crust,..
sweet potato,..
onion,..
tomato,..
brown rice,..
celery,..
pepper,..
spaghetti sauce,..
hot,..
cheese,.. chicken,..
spinach,..
seasoning,..
black bean,..
red potato,..
italian seasoning,..
cream of mushroom soup,..
sugar snap pea,..iceberg lettuce,.. curry powder,..
imitation crab meat,..
pickle,..
quinoa,..
tilapia,.. cabbage,..sea scallop,.. smoked paprika,..
Figure 6: Ingredient substitution clusters. Nodes
represent clusters and edges indicate the presence of
recommended substitutions that span clusters. Each
cluster represents a set of related ingredients which
are frequently substituted for one another
we introduce a directed PMI measure to cope with the directionality of the preference network:
pmi(a → b) = log
p(a → b)
,
p(a)p(b)
where
p(a → b) =
# of recipe pairs f rom a to b
,
# of recipe pairs
and p(a), p(b) are defined as in the previous section.
Comparing the substitution network with this preference
network, we found high correlations between the two net-
whipping cream
ginger root
evaporated milk
half and half
cream
buttermilk
heavy cream
cardamom
pumpkin pie spice
cinnamon
heavy whipping cream
milk
clove
whole milk
soy milk
skim milk
(a) milk substitutes
ginger
nutmeg
allspice
mace
(b) cinammon substitutes
Figure 7: Relationships between ingredients located
within two of the clusters from Fig. 6.
works (ρ = 0.72, p < 0.001). This observation suggests that
the substitute network encodes users’ ingredient preference,
which we will use in the recipe prediction task described in
the next section.
7.
RECIPE RECOMMENDATION
We use the above insights to uncover novel recommendation algorithms suitable for recipe recommendations. We
use ingredients and the relationships encoded between them
in ingredient networks as our main feature sets to predict
recipe ratings, and compare them against features encoding nutrition information, as well as other baseline features
such as cooking methods, and preparation and cook time.
Then we apply a discriminative machine learning method,
stochastic gradient boosting tree [7], to predict recipe ratings.
In the experiments, we seek to answer three questions useful for recipe recommendation: (1) Can we predict users’
preference for a new recipe given the information present in
the recipe? (2) What are the key aspects that determine
users’ preference? (3) Does the structure of ingredient networks help in recipe recommendation, and how?
We shall answer these questions through a prediction task.
7.1
Recipe Pair Prediction
The goal of our prediction task is: given a pair of similar
recipes, determine which one has higher average rating than
the other. This task is designed particularly to help users
with a specific dish or meal in mind, and who are trying to
decide between several recipe options for that dish.
Recipe pair data. The data for this prediction task
consists of pairs of similar recipes. The reason for selecting similar recipes, with high ingredient overlap, is that
while apples may be quite comparable to oranges in the
context of recipes, especially if one is evaluating salads or
desserts, lasagna may not be comparable to a mixed drink.
To derive pairs of related recipes, we computed similarity
with a cosine similarity between the ingredient lists for the
two recipes, weighted by the inverse document frequency,
log(# of recipes/# of recipes containing the ingredient).
We considered only those pairs of recipes whose cosine similarity exceeded 0.2. The weighting is intended to identify
higher similarity among recipes sharing more distinguishing
ingredients, such as Brussels sprouts, as opposed to recipes
sharing very common ones, such as butter.
A further challenge to obtaining reliable relative rankings
of recipes is variance introduced by having different users
choose to rate different recipes. In addition, some users
might not have a sufficient number of reviews under their
belt to have calibrated their own rating scheme. To control for variation introduced by users, we examined recipe
pairs where the same users are rating both recipes and are
collectively expressing a preference for one recipe over another. Specifically, we generated 62,031 recipe pairs (a, b)
where (ratingi (a) > ratingi (b), for at least 10 users i, and
over 50% of users who rated both recipe a and recipe b. Furthermore, each user i should be an active enough reviewer
to have rated at least 8 other recipes.
Features. In the prediction dataset, each observation
consists of a set of predictor variables or features that represent information about two recipes, and the response variable is a binary indicator of which gets the higher rating on
average. To study the key aspects of recipe information, we
constructed different set of features, including:
• Baseline: This includes cooking methods, such as chopping, marinating, or grilling, and cooking effort descriptors, such as preparation time in minutes, as well
as the number of servings produced, etc. These features are considered as primary information about a
recipe and will be included in all other feature sets
described below.
• Full ingredients: We selected up to 1000 popular ingredients to build a “full ingredient list”. In this feature
set, each observed recipe pair contains a vector with
entries indicating whether an ingredient in the full list
is present in either recipe in the pair.
• Nutrition: This feature set does not include any ingredients but only nutrition information such the total
caloric content, as well as amount of fats, carbohydrates, etc.
• Ingredient networks: In this set, we replaced the full
ingredient list by structural information extracted from
different ingredient networks, as described in section 5
and 6.3. Co-occurrence is treated separately as a raw
count, and a complementarity, captured by the PMI.
• Combined set: Finally, a combined feature set is constructed to test the performance of a combination of
features, including baseline, nutrition and ingredient
networks.
To build the ingredient network feature set, we extracted
the following two types of structural information from the
co-occurrence and substitution networks, as well as the complement network derived from the co-occurrence information:
Network positions are calculated to represent how a recipe’s
ingredients occupy positions within the networks. Such position measures are likely to inform if a recipe contains any
“popular” or “unusual” ingredients. To calculate the position measures, we first calculated various network centrality
measures, including degree centrality, betweenness centrality, etc., from the ingredient networks. A centrality measure
can be represented as a vector ~g where each entry indicates
the centrality of an ingredient. The network position of a
recipe, with its full ingredient list represented as a binary
vector f~, can be summarized by ~g T · f~, i.e., an aggregated
centrality measure based on the centrality of its ingredients.
Network communities provide information about which
ingredient is more likely to co-occur with a group of other
ingredients in the network. A recipe consisting of ingredients
that are frequently used with, complemented by or substituted by certain groups may be predictive of the ratings
network
0.7
combined
substitution (39.8%)
co−occurrence (30.9%)
0.6
complement (29.2%)
ing. networks
importance
0.5
nutrition
full ingredients
0.4
0.3
0.2
baseline
0.1
0.80
0.75
0.70
0.65
0.60
0.0
Accuracy
Figure 8: Prediction performance. The nutrition
information and ingredient networks are more effective features than full ingredients. The ingredient network features lead to impressive performance
close to the best performance, indicating the power
of network structures in recipe recommendation.
40
60
80
100
feature
Figure 10: Relative importance of features represented the network structure. The substitution network has stronger contribution (39.8%) to the total
importance of ingredient network features than the
other two networks, and it also has more influential features in the top 100 list, which suggests the
information about substitution network is complementary to other features.
group
1.0
nutrition (6.5%)
cook effort (5.0%)
0.8
ing. networks (84%)
cook methods (3.9%)
importance
20
0.6
0.4
0.2
0.0
20
40
60
80
100
feature
Figure 9: Relative importance of features in the
combined set. The individual items from nutrition information are very indicative in differentiating high-rated recipes, while most of the prediction
power comes from ingredient networks.
the recipe will receive. To obtain the network community
information, we applied latent semantic analysis (LSA) on
recipes. We first factorized each ingredient network, represented by matrix W , using singular value decomposition
(SVD). In the matrix W , each entry Wij indicated whether
ingredient i co-occurrs, complements or substitues ingredient j.
Suppose Wk = Uk Σk VkT is a rank-k approximation of W ,
we can then transform each recipe’s full ingredient list using
T ~
the low-dimensional representation, Σ−1
k Vk f , as community
information within a network. These low-dimensional vectors, together with the vectors of network positions, constitute the ingredient network features.
Learning method. We applied discriminative machine
learning methods such as support vector machines (SVM) [2]
and stochastic gradient boosting trees [6] to our prediction
problem. Here we report and discuss the detailed results
based on the gradient boosting tree model. Like SVM, the
gradient boosting tree model seeks a parameterized classifier, but unlike SVM that considers all the features at one
time, the boosting tree model considers a set of features
at a time and iteratively combines them according to their
empirical errors. In practice, it not only has competitive
performance comparable to SVM, but can serve as a feature
ranking procedure [11].
In this work, we fitted a stochastic gradient boosting tree
model with 8 terminal nodes under an exponential loss function. The dataset is roughly balanced in terms of which
recipe is the higher-rated one within a pair. We randomly
divided the dataset into a training set (2/3) and a testing
set (1/3). The prediction performance is evaluated based on
accuracy, and the feature performance is evaluated in terms
of relative importance [9]. For each single decision tree, one
of the input variables, xj , is used to partition the region associated with that node into two subregions in order to fit
to the response values. The squared relative importance of
variable xj is the sum of such squared improvements over
all internal nodes for which it was chosen as the splitting
variable, as:
X 2
imp(j) =
îk I(splits on xj )
k
where î2k is the empirical improvement by the k-th node
splitting on xj at that point.
7.2
Results
The overall prediction performance is shown in Fig. 8.
Surprisingly, even with a full list of ingredients, the prediction accuracy is only improved from .712 (baseline) to
.746. In contrast, the nutrition information and ingredient
networks are more effective (with accuracy .753 and .786, respectively). Both of them have much lower dimensions (from
tens to several hundreds), compared with the full ingredients
that are represented by more than 2000 dimensions (1000
ingredients per recipe in the pair). The ingredient network
features lead to impressive performance close to the best
performance given by the combined set (.792), indicating
the power of network structures in recipe recommendation.
Figure 9 shows the influence of different features in the
0.80
1.0
nutrition
carbs (20.9%)
cholesterol (17.7%)
0.8
calories (19.7%)
●
●
0.79
fiber (12.3%)
●
fat (12.4%)
●
0.4
Accuracy
importance
sodium (16.8%)
0.6
0.2
●
●
●
0.78
0.0
2
4
6
8
10
12
feature
Figure 11: Relative importance of features from
nutrition information. The carbs item is the most
influential feature in predicting higher-rated recipes.
network
0.77
● combined
substitution
complement
co−occurrence
0.76
10
20
30
40
50
60
70
Dimensions
Figure 12: Prediction performance over reduced
dimensionality. The dimensionality of network features can be determined by cross-validation. The
best performance is given by reduced dimension
k = 50 when combining all three networks. In addition, using the information about the complement
network alone is more effective in prediction than
using other two networks.
Color Key
82
−0.5
433
0.5
Value
splenda
olive oil
applesauce
honey
butter
brown sugar
milk
half and half
chicken broth
buttermilk
sour cream
evaporated milk
vanilla extract
vanilla
kale
almond
beef
cream of chicken soup
almond extract
chocolate pudding
lemon extract
lime juice
walnut
coconut extract
turkey
chicken
sausage
italian sausage
pork
chicken breast
194
65
splenda
olive oil
applesauce
honey
butter
brown sugar
milk
half and half
chicken broth
buttermilk
sour cream
evaporated milk
vanilla extract
vanilla
kale
almond
beef
cream of chicken soup
almond extract
chocolate pudding
lemon extract
lime juice
walnut
coconut extract
turkey
chicken
sausage
italian sausage
pork
chicken breast
6
19
43
8
4
Figure 13:
Influential substitution communities.
The matrix shows the most influential feature dimensions extracted from the substitution network.
For each dimension, the six representative ingredients with the highest intensity values in the decomposed matrix are shown, with colors indicating their
intensity. These features suggest that the communities of ingredient substitutes, such as the sweet and
oil in the first dimension, are particularly informative in prediction.
In Fig. 13 we show the most representative ingredients in
the decomposed matrix derived from the substitution network. We display the top five influential dimensions, evaluated based on the relative importance, from the SVD resultant matrix Vk , and in each of these dimensions we extracted
six representative ingredients based on their intensities in
Color Key
−0.5
Value
ingredient
41
svd dimension
combined feature set. Up to 100 features with the highest
relative importance are shown. The importance of a feature
group is summarized by how much the total importance is
contributed by all features in the set. For example, the baseline consisting of cooking effort and cooking methods have
contributed 8.9% to the overall performance. The individual items from nutrition information are very indicative in
differentiating highly-rated recipes, while most of the prediction power comes from ingredient networks (84%).
Figure 10 shows the top 100 features from the three networks. In terms of the total importance of ingredient network features, the substitution network has slightly stronger
contribution (39.8%) than the other two networks, and it
also has more influential features in the top 100 list. This
suggests that the structural information extracted from the
substitution network is not only important but also complementary to information from other aspects.
Looking into the nutrition information (Fig. 11), we found
that carbohydrates are the most influential feature in predicting higher-rated recipes. Since carbohydrates comprise
around 50% or more of total calories, the high importance
of this feature interestingly suggests that a recipe’s rating
can be influenced by users’ concerns about nutrition and
diet. Another interesting observation is that, while individual nutrition items are powerful predictors, a higher prediction accuracy can be reached by using ingredient networks
alone, as shown in Fig. 8. This implies the information
about nutrition may have been encoded in the ingredient
network structure, e.g. substitutions of less healthful ingredients with “healthier” alternatives.
Constructing the ingredient network feature involves reducing high-dimensional network information through SVD,
as described in the previous section. The dimensionality can
be determined by cross-validation. As shown in Fig. 12, features with a very large dimension tend to overfit the training
data. Hence we chose k = 50 for the reduced dimension of
all three networks. The figure also shows that using the
information about the complement network alone is more
effective in prediction than using other the co-occurrence
and substitute networks, even in the case of low dimensions.
However, as shown in terms of relative importance (Fig. 10),
the substitution network alone is not the most effective, but
it provides more complementary information in the combined feature set.
the dimension (the squared entry values). These reprsentative ingredients suggest that the communities of ingredient
substitutes, such as the sweet and oil substitutes in the first
dimension or the milk substitutes in the second dimesion
(which is similar to the cluster shown in Fig. 6), are particularly informative in predicting recipe ratings.
To summarize our observations from the experiments, we
found we were able to effectively predict users’ preference for
a recipe, but the prediction is not through using a full list
of ingredients. Instead, by using the structural information
extracted from the relationships among ingredients, we can
better uncover users’ preference about recipes.
8.
CONCLUSION
Recipes are little more than instructions for combining
and processing sets of ingredients. Individual cookbooks,
even the most expansive ones, contain single recipes for each
dish. The web, however, permits collaborative recipe generation and modification, with tens of thousands of recipes
contributed in individual websites. We have shown how this
data can be used to glean insights about regional preferences
and modifiability of individual ingredients, and also how it
can be used to construct two kinds of networks, one of ingredient complements, the other of ingredient substitutes.
These networks encode which ingredients go well together,
and which can be substituted to obtain superior results, and
permit one to predict, given a pair of related recipes, which
one will be more highly rated by users.
In future work, we plan to extend ingredient networks to
incorporate the cooking methods as well. It would also be
of interest to generate region-specific and diet-specific ratings, depending on the users’ background and preferences.
A whole host of user-interface features could be added for
users who are interacting with recipes, whether the recipe
is newly submitted, and hence unrated, or whether they are
browsing a cookbook. In addition to automatically predicting a rating for the recipe, one could flag ingredients that
can be omitted, ones whose quantity could be tweaked, as
well as suggested additions and substitutions.
9.
ACKNOWLEDGMENTS
This work was supported by MURI award FA9550-08-10265 from the Air Force Office of Scientific Research. The
methodology used in this paper was developed with support from funding from the Army Research Office, MultiUniversity Research Initiative on Measuring, Understanding, and Responding to Covert Social Networks: Passive
and Active Tomography.
10.
REFERENCES
[1] S. M. Boback, C. L. Cox, B. D. Ott, R. Carmody,
R. W. Wrangham, and S. M. Secor. Cooking and
grinding reduces the cost of meat digestion.
Comparative Biochemistry and Physiology - Part A:
Molecular and Integrative Physiology, 148(3):651 –
656, 2007.
[2] C. Cortes and V. Vapnik. Support-vector networks.
Machine learning, 20(3):273–297, 1995.
[3] P. Forbes and M. Zhu. Content-boosted matrix
factorization for recommender systems: Experiments
with recipe recommendation. Proceedings of
Recommender Systems, 2011.
[4] J. Freyne and S. Berkovsky. Intelligent food planning:
personalized recipe recommendation. In IUI, pages
321–324. ACM, 2010.
[5] J. Freyne and S. Berkovsky. Recommending food:
Reasoning on recipes and ingredients. User Modeling,
Adaptation, and Personalization, pages 381–386, 2010.
[6] J. Friedman. Stochastic gradient boosting.
Computational Statistics & Data Analysis,
38(4):367–378, 2002.
[7] J. Friedman, T. Hastie, and R. Tibshirani. Additive
logistic regression: a statistical view of boosting.
Annals of Statistics, 28:2000, 1998.
[8] G. Geleijnse, P. Nachtigall, P. van Kaam, and
L. Wijgergangs. A personalized recipe advice system
to promote healthful choices. In IUI, pages 437–438.
ACM, 2011.
[9] T. Hastie, R. Tibshirani, J. Friedman, and J. Franklin.
The elements of statistical learning: data mining,
inference and prediction. The Mathematical
Intelligencer, 27(2), 2005.
[10] F. Kamieth, A. Braun, and C. Schlehuber. Adaptive
implicit interaction for healthy nutrition and food
intake supervision. Human-Computer Interaction.
Towards Mobile and Intelligent Interaction
Environments, pages 205–212, 2011.
[11] Y. Lu, F. Peng, X. Li, and N. Ahmed. Coupling
feature selection and machine learning methods for
navigational query identification. In CIKM, pages
682–689. ACM, 2006.
[12] I. Rombauer, M. Becker, E. Becker, and L. Maestro.
Joy of cooking. Scribner Book Company, 1997.
[13] M. Rosvall and C. Bergstrom. Maps of random walks
on complex networks reveal community structure.
PNAS, 105(4):1118, 2008.
[14] Y. Shidochi, T. Takahashi, I. Ide, and H. Murase.
Finding replaceable materials in cooking recipe texts
considering characteristic cooking actions. In Proc. of
the ACM multimedia 2009 workshop on Multimedia for
cooking and eating activities, pages 9–14. ACM, 2009.
[15] M. Svensson, K. Höök, and R. Cöster. Designing and
evaluating kalas: A social navigation system for food
recipes. ACM Transactions on Computer-Human
Interaction (TOCHI), 12(3):374–400, 2005.
[16] M. Ueda, M. Takahata, and S. Nakajima. User’s food
preference extraction for personalized cooking recipe
recommendation. Proc. of the Second Workshop on
Semantic Personalized Information Management:
Retrieval and Recommendation, 2011.
[17] L. Wang, Q. Li, N. Li, G. Dong, and Y. Yang.
Substructure similarity measurement in chinese
recipes. In WWW, pages 979–988. ACM, 2008.
[18] Wikipedia. Outline of food preparation, 2011. [Online;
accessed 22-Oct-2011].
[19] R. Wrangham. Catching fire: how cooking made us
human. Profile Books, 2010.
[20] Q. Zhang, R. Hu, B. Mac Namee, and S. Delany. Back
to the future: Knowledge light case base cookery. In
Proc. of The 9th European Conference on Case-Based
Reasoning Workshop, page 15, 2008.