Automatic main ingredient extraction from food recipe
Provided is a system ( 100 ) for determining the main ingredients of a food recipe. The system comprises a first input ( 102 ) for receiving the identity and mass fraction (w ri ) of each ingredient (i) in the food recipe (r), and a second input ( 104 ) for receiving the identity of ingredients of each known recipe of a population of known recipes. The system further includes a controller ( 106 ) configured to calculate, for each ingredient (i), the fraction (P(i)) of the population which uses the ingredient. The controller then calculates, using the fraction (P(i)) and the mass fraction (w ri ), a value (V ri ) which positively correlates with the mass fraction (w ri ) and negatively correlates with the fraction (P(i)). The main ingredients are then determined according to the ingredients which have values (V ri ) equal to or greater than a threshold value. Further provided is a cooking appliance ( 200 ) which includes the system, a computer implemented method for determining the main ingredients of a food recipe, and a computer program which implements the method.
1. A system for determining main ingredients of a food recipe, the system comprising:
a first input for receiving, via a user interface, an identity and mass fraction (w ri ) of each ingredient in the food recipe;
a second input for receiving, from a database, the identity of ingredients of each known food recipe of a population of known food recipes; and
a controller configured to:
calculate, for the each ingredient, fraction (P(i)) of said population of known food recipes which uses the ingredient;
calculate, using the fraction (P(i)) and the mass fraction (w ri ), values (V ri ) which positively correlates with the mass fraction (w ri ) and negatively correlates with the fraction (P(i)); and
determine the main ingredients according to the ingredients which have the values (V ri ) equal to or greater than a threshold value.
2. The system according to claim 1 , wherein the controller is further configured to:
sequence the values (V ri ) calculated for the each ingredient by size;
determine largest difference between consecutive values in the sequence; and
identify a pair of the consecutive values in the sequence corresponding to said largest difference, wherein the threshold value is equal to larger value of said pair.
3. The system according to claim 1 , wherein the controller is further configured to calculate said values (V ri ) from the mass fraction (w ri ) and the fraction (P(i)) using formula: V ri =w ri log(P(i)).
4. The system according to claim 1 , wherein the controller is further configured to determine a parameter (Sim(r1, r2)) relating to similarity of food recipes based on the determined main ingredients and/or based on determined non-main ingredients of the respective food recipes, wherein the determined non-main ingredients are ingredients which are not the main ingredients of the respective food recipes.
5. The system according to claim 4 , wherein said parameter (Sim(r1, r2)) is based on the determined main ingredients, and is defined by following formula:
Sim
(
r
1
,
r
2
)
=
(
Formula
1
)
∑
k
is
main
ingredient
i
n
r
1
or
r
2
min
(
w
r
1
j
,
w
r
2
j
)
·
log
(
P
(
j
)
)
∑
k
is
main
ingredient
i
n
r
1
or
r
2
max
(
w
r
1
j
,
w
r
2
j
)
·
log
(
P
(
j
)
)
wherein w r1j and w r2j are mass fractions of the main ingredient j in the respective food recipes r 1 and r 2 ; P(j) is fraction of said population of known food recipes which uses the main ingredient.
6. The system according to claim 4 , wherein said parameter (Sim(r1, r2)) is based on the determined non-main ingredients, and is defined by following formula:
Sim
(
r
1
,
r
2
)
=
(
Formula
2
)
∑
k
is
non
-
main
ingredient
i
n
r
1
or
r
2
min
(
w
r
1
k
,
w
r
2
k
)
·
log
(
1
-
P
(
k
)
)
∑
k
is
non
-
main
ingredient
i
n
r
1
or
r
2
max
(
w
r
1
k
,
w
r
2
k
)
·
log
(
1
-
P
(
k
)
)
wherein w r1k and w r2k are mass fractions of the non-main ingredient k in the respective food recipes r 1 and r 2 ; P(k) is fraction of said population of known food recipes which uses the main ingredient k.
7. The system according to claim 4 , wherein the controller is further configured to:
use a word similarity model to determine a similarity score for each pairwise combination of the determined main ingredients or the non-main ingredients of the respective recipes;
use the similarity score to identify pairs of main or non-main ingredients of the respective recipes; and
use a threshold of the similarity score to identify any unpaired main or non-main ingredients of the respective recipes.
8. The system according to claim 7 , wherein said parameter (Sim(r 1 , r 2 )) is based on the determined main ingredients, and is defined by following formula:
Sim
(
r
1
,
r
2
)
=
(
Formula
4
)
∑
l
,
m
are
paired
main
ingredients
min
(
w
r
1
l
,
w
r
2
m
)
·
log
(
P
(
l
)
·
P
(
m
)
)
·
Sim
(
l
,
m
)
∑
l
,
m
are
paired
main
ingredients
max
(
w
r
1
l
,
w
r
2
m
)
·
log
(
P
(
l
)
·
P
(
m
)
)
+
2
∑
n
is
unpaired
main
ingredient
i
n
r
1
or
r
2
max
(
w
r
1
n
,
w
r
2
n
)
·
log
(
P
(
n
)
)
wherein w r 1 l is mass fraction of main ingredient l, paired with m, in recipe r 1 ; w r 2 m is mass fraction of main ingredient m, paired with c, in recipe r 2 ; P(l) is fraction of said population of known food recipes which uses the ingredient l; P(m) is fraction of said population of known food recipes which uses the ingredient m; Sim(l, m) is the similarity score for the paired main ingredients (l, m); w r 1 n and w r 2 n are mass fractions of unpaired main ingredient n in the respective recipes r 1 and r 2 ; P(n) is fraction of said population of known food recipes which uses the unpaired main ingredient n.
9. The system according to claim 7 , wherein said parameter (Sim(r 1 , r 2 )) is based on the determined non-main ingredients, and is defined by following formula:
Sim
(
r
1
,
r
2
)
=
(
Formula
5
)
∑
o
,
p
are
paired
non
-
main
ingredients
min
(
w
r
1
l
,
w
r
2
m
)
·
log
(
P
(
l
)
·
P
(
m
)
)
·
Sim
(
l
,
m
)
∑
o
,
p
are
paired
non
-
main
ingredients
max
(
w
r
1
p
,
w
r
2
p
)
·
log
(
(
1
-
P
(
o
)
)
·
P
(
1
-
(
p
)
)
)
+
2
∑
q
is
unpaired
non
-
main
ingredient
i
n
r
1
or
r
2
max
(
w
r
1
q
,
w
r
2
q
)
·
log
(
1
-
P
(
q
)
)
wherein w r1o is mass fraction of main ingredient o, paired with p, in recipe r 1 ; w r2p is mass fraction of non-main ingredient p, paired with o, in recipe r 2 ; P(o) is the fraction of said population of known food recipes which uses the ingredient o; P(p) is fraction of said population of known food recipes which uses the ingredient p; Sim(o, p) is the similarity score for the paired non-main ingredients (o, p); w r1q and w r2q are mass fractions of unpaired non-main ingredient q in the respective recipes r 1 and r 2 ; P(q) is fraction of said population of known food recipes which uses the unpaired non-main ingredient q.
10. A cooking appliance, comprising:
a heating element; and
the system according to claim 1 , wherein the controller is configured to control the heating element based on the determined main ingredients and optionally based on masses of the determined main ingredients in the food recipe.
11. A computer implemented method for determining main ingredients of a food recipe, the method comprising:
receiving, via a user interface, an identity and mass fraction (w ri ) of each ingredient (i) in the food recipe (r);
receiving, from a database, the identity of ingredients of each known food recipe of a population of known food recipes;
calculating, for the each ingredient (i), fraction (P(i)) of said population of known food recipes which uses the ingredient;
calculating, using the fraction (P(i)) and the mass fraction (w ri ), values (V ri ) which positively correlates with the mass fraction (w ri ) and negatively correlates with the fraction (P(i)); and
determining the main ingredients according to the ingredients which have the values (V ri ) equal to or greater than a threshold value.
12. The method according to claim 11 , further comprising: sequencing the values (V ri ) calculated for the each ingredient by size;
determining largest difference between consecutive values in the sequence; and identifying a pair of the consecutive values in the sequence corresponding to said largest difference, wherein the threshold value is equal to larger value of said pair.
13. The method according to claim 11 , wherein the calculating said values (V ri ) from the mass fraction (w ri ) and the fraction (P(i)) uses formula:
V ri =w ri log( P ( i )).
14. The method according to claim 11 , further comprising determining a parameter relating to similarity of recipes based on the determined main ingredients of the respective recipes and/or based on determined non-main ingredients of the respective food recipes, wherein the determined non-main ingredients are ingredients which are not the main ingredients of the respective food recipes.
15. A non-transitory computer readable medium comprising a computer program which is adapted, when said computer program is run on a computer, to implement the method of claim 11 .