BUDGET DISTRIBUTION IN ONLINE ADVERTISING
A method for budget distribution in online advertising, the method comprising using at least one hardware processor for: receiving a definition of a single advertiser budget to be spent on advertising multiple ad entities in an online advertising platform; receiving historical performance data associated with the multiple ad entities, wherein the historical performance data comprises multiple proportional performance metrics for each of the multiple ad entities; computing a health index for each of the multiple ad entities, the health index being a weighted average of multiple components comprising the multiple proportional performance metrics, wherein the multiple components are each monotonic with respect to spend; and proportionally distributing the single advertiser budget between the multiple ad entities, based on the health indices of the multiple ad entities.
1 . A method for budget distribution in online advertising, the method comprising using at least one hardware processor for:
receiving a definition of a single advertiser budget to be spent on advertising multiple ad entities in an online advertising platform;
receiving historical performance data associated with the multiple ad entities, wherein the historical performance data comprises multiple proportional performance metrics for each of the multiple ad entities;
computing a health index for each of the multiple ad entities, the health index being a weighted average of multiple components comprising the multiple proportional performance metrics, wherein the multiple components are each monotonic with respect to spend; and
proportionally distributing the single advertiser budget between the multiple ad entities, based on the health indices of the multiple ad entities.
2 . The method according to claim 1 , wherein the multiple components further comprise an absolute parameter.
3 . The method according to claim 2 , wherein the absolute parameter is a current reach value for each of the multiple ad entities.
4 . The method according to claim 1 , wherein the computing of the health index further comprises a Bayesian estimation of at least some of the multiple proportional performance metrics.
5 . The method according to claim 4 , wherein the computing of the health index further comprises standardizing the multiple components.
6 . The method according to claim 5 , wherein the health indices are sigmoid-transformed health indices.
7 . The method according to claim 1 , wherein the proportionally distributing comprises solving the optimization program:
Minimize
∑
(
B
*
h
k
∑
(
h
k
)
-
x
k
)
2
subject
to
∑
(
x
k
)
=
B
lb
k
<
x
k
<
u
b
k
where:
k is one ad entity of the multiple ad entities,
h k is one health index of the of the k th ad entity,
B is the single advertiser budget,
x k represents an unknown budget ration per the k th ad entity, and
x k is constrained by lb k and ub k , which are lower and upper bounds, respectively, imposed on the k th ad entity of the budget ration x k .
8 . The method according to claim 1 , wherein the computing of the health index further comprises endowing different ones of the multiple ad entities with different weights, based on business logic provided by an advertiser.
9 . The method according to claim 8 , wherein the multiple components comprise the click-through rate, the conversion rate, the potential reach, the spend rate and the reach.
10 . The method according to claim 1 , wherein:
the single advertiser budget is defined for a period of time over which the single advertiser budget is to be spent; and
the receiving of the historical performance data, the computing of the health index and the proportionally distributing are performed multiple times during the period of time for which the single advertiser budget is defined, thereby optimizing a spend rate of the single advertiser budget during the period of time.
11 . A computer program product for budget distribution in online advertising, the computer program product comprising a non-transitory computer-readable storage medium having program code embodied therewith, the program code executable by at least one hardware processor for:
receiving a definition of a single advertiser budget to be spent on advertising multiple ad entities in an online advertising platform;
receiving historical performance data associated with the multiple ad entities, wherein the historical performance data comprises multiple proportional performance metrics for each of the multiple ad entities;
computing a health index for each of the multiple ad entities, the health index being a weighted average of multiple components comprising the multiple proportional performance metrics, wherein the multiple components are each monotonic with respect to spend; and
proportionally distributing the single advertiser budget between the multiple ad entities, based on the health indices of the multiple ad entities.
12 . The computer program product according to claim 11 , wherein the multiple components further comprise an absolute parameter.
13 . The computer program product according to claim 12 , wherein the absolute parameter is a current reach value for each of the multiple ad entities.
14 . The computer program product according to claim 12 , wherein the computing of the health index further comprises a Bayesian estimation of at least some of the multiple proportional performance metrics.
15 . The computer program product according to claim 14 , wherein the computing of the health index further comprises standardizing the multiple components.
16 . The computer program product according to claim 15 , wherein the health indices are sigmoid-transformed health indices.
17 . The computer program product according to claim 11 , wherein the proportionally distributing comprises solving the optimization program:
Minimize
∑
(
B
*
h
k
∑
(
h
k
)
-
x
k
)
2
subject
to
∑
(
x
k
)
=
B
lb
k
<
x
k
<
u
b
k
where:
k is one ad entity of the multiple ad entities,
h k is one health index of the of the k th ad entity,
B is the single advertiser budget,
x k represents an unknown budget ration per the k th ad entity, and
x k is constrained by lb k and ub k , which are lower and upper bounds, respectively, imposed on the k th ad entity of the budget ration x k .
18 . The computer program product according to claim 11 , wherein the computing of the health index further comprises endowing different ones of the multiple ad entities with different weights, based on business logic provided by an advertiser.
19 . The computer program product according to claim 11 , wherein the multiple components comprise the click-through rate, the conversion rate, the potential reach, the spend rate and the reach.
20 . The computer program product according to claim 11 , wherein:
the single advertiser budget is defined for a period of time over which the single advertiser budget is to be spent; and
the receiving of the historical performance data, the computing of the health index and the proportionally distributing are performed multiple times during the period of time for which the single advertiser budget is defined, thereby optimizing a spend rate of the single advertiser budget during the period of time.