Machine learning techniques that identify attribution of small signal stimulus in noisy response channels
View Patent ↗A method, system, and computer program product identifies attribution of small signal stimulus in noisy response channels. Using machine-learning techniques in a computer, a small signal correlation engine correlates time series stimuli data vectors to time series response data vectors, and generates correlation coefficients that identify contributions of event notifications, including small signal attributes, to aggregated response data. Also using machine-learning techniques in a computer, a learning model simulates variations of stimuli data to predict user responses using the correlation coefficients, including computing a contribution of the small signal attributes of an event notification.
1. A computer-implemented method for using machine-learning techniques to present a user interface with campaign spend allocations based on attribution of small signal stimulus in noisy response channels, the method comprising:
generating, in a computer, from stimuli data associated with at least one channel, a plurality of time series stimuli data vectors from event notifications for an event propagated through a channel over a period of time to a plurality of users, the event notifications being associated with a plurality of small signal attributes specified at a sub-channel level of the channel;
generating, from aggregated response data associated with the event and including a plurality of user responses to the event, a plurality of time series response data vectors from responses elicited over time;
generating, using machine-learning techniques in a computer, a small signal correlation engine, to correlate at least one of the time series stimuli data vectors to one of the time series response data vectors, and generate a plurality of correlation coefficients that correlate the event notifications to the aggregated response data;
generating, using machine-learning techniques in a computer, a learning model to simulate variations of the stimuli data based on stimuli and aggregated response data to predict user responses using the correlation coefficients, including computing a contribution value of at least one of the plurality of small signal attributes of at least one of the event notifications;
updating the learning model based on a comparison of the predicted user responses to actual user responses to the event;
generating a media spend allocation for the channel and at least one of a sub-channel associated with the channel or another channel based on the contributions of the event notifications associated with the channel and the at least one of the sub-channel associated with the channel or the other channel and a budget received via a user interface;
displaying the media spend allocation to a user via the user interface;
simulating an effect of an adjustment to the media spend allocation, received via the user interface, based on the predicted user responses simulated by the learning model; and
presenting, via the user interface, the adjusted media spend allocation and predicted user response activity based on the adjustment determined by the simulating.
2. The computer implemented method as set forth in claim 1 , further including filtering the aggregated response data with a transfer function, h(n), that includes a first portion of the response data vectors and excludes a second portion of the response data vectors.
3. The computer implemented method as set forth in claim 2 , wherein the transfer function, h(n), is based on a geographical origin of the responses, the IP address of the users, a response time, or a response channel.
4. The computer-implemented method as set forth in claim 1 , wherein generating, using machine-learning techniques in a computer, a small signal correlation engine in accordance with the expression:
r
=
n
∑
xy
-
(
∑
x
)
(
∑
y
)
n
(
∑
x
2
)
-
(
∑
x
)
2
n
(
∑
y
2
)
-
(
∑
y
)
2
where:
x represents components of the time series stimuli data vectors,
y represents components of the time series response data vectors, and
n is a number of {x, y} pairs used.
5. The computer-implemented method as set forth in claim 1 , wherein the small signal attributes include sequences of attributes, values and measurements of the event notification.
6. The computer-implemented method as set forth in claim 1 , wherein the event notifications include marketing messages deployed across television channels.
7. The computer-implemented method as set forth in claim 6 , wherein the small signal attributes include attributes that define a television spot.
8. The computer-implemented method as set forth in claim 1 , wherein the channels include channels associated with an Internet of Things System.
9. A computer program product embodied in a non-transitory computer readable medium, the computer readable medium having stored thereon a sequence of instructions which, when executed by a processor causes the processor to execute a process to present a user interface with campaign spend allocations based on attribution of small signal stimulus in noisy response channels, the process comprising:
generating, in a computer, from stimuli data associated with at least one channel, a plurality of time series stimuli data vectors from event notifications for an event propagated through a channel over a period of time to a plurality of users, the event notifications being associated with a plurality of small signal attributes specified at a sub-channel level of the channel;
generating, from aggregated response data associated with the event and including a plurality of user responses to the event, a plurality of time series response data vectors from responses elicited over time;
generating, using machine-learning techniques in a computer, a small signal correlation engine, to correlate at least one of the time series stimuli data vectors to one of the time series response data vectors, and generate a plurality of correlation coefficients that correlate the event notifications to the aggregated response data;
generating, using machine-learning techniques in a computer, a learning model to simulate variations of the stimuli data based on stimuli and aggregated response data to predict user responses using the correlation coefficients, including computing a contribution value of at least one of the plurality of small signal attributes of at least one of the event notifications;
updating the learning model based on a comparison of the predicted user responses to actual user responses to the event;
generating a media spend allocation for the channel and at least one of a sub-channel associated with the channel or another channel based on the contributions of the event notifications associated with the channel and the at least one of the sub-channel associated with the channel or the other channel and a budget received via a user interface;
displaying the media spend allocation to a user via the user interface;
simulating an effect of an adjustment to the media spend allocation, received via the user interface, based on the predicted user responses simulated by the learning model; and
presenting, via the user interface, the adjusted media spend allocation and predicted user response activity based on the adjustment determined by the simulating.
10. The computer program product as set forth in claim 9 , further including filtering the aggregated response data with a transfer function, h(n), that includes a first portion of the response data vectors and excludes a second portion of the response data vectors.
11. The computer program product as set forth in claim 10 , wherein the transfer function, h(n), is based on a geographical origin of the responses, the IP address of the users, a response time, or a response channel.
12. The computer program product as set forth in claim 9 , wherein generating, using machine-learning techniques in a computer, a small signal correlation engine in accordance with the expression:
r
=
n
∑
xy
-
(
∑
x
)
(
∑
y
)
n
(
∑
x
2
)
-
(
∑
x
)
2
n
(
∑
y
2
)
-
(
∑
y
)
2
where:
x represents components of the time series stimuli data vectors,
y represents components of the time series response data vectors, and
n is a number of {x, y} pairs used.
13. The computer program product as set forth in claim 9 , wherein the small signal attributes include sequences of attributes, values and measurements of the event notification.
14. The computer program product as set forth in claim 9 , wherein the event notifications include marketing messages deployed across television channels.
15. The computer program product as set forth in claim 14 , wherein the small signal attributes include attributes that define a television spot.
16. The computer program product as set forth in claim 9 , wherein the channels include channels associated with an Internet of Things System.
17. A computer system comprising:
a computer processor to execute a set of program code instructions to present a user interface with campaign spend allocations based on attribution of small signal stimulus in noisy response channels; and
a memory to hold the program code instructions, in which the program code instructions includes program code to perform:
generating, in a computer, from stimuli data associated with at least one channel, a plurality of time series stimuli data vectors from event notifications for an event propagated through a channel over a period of time to a plurality of users, the event notifications being associated with a plurality of small signal attributes specified at a sub-channel level of the channel;
generating, from aggregated response data associated with the event and including a plurality of user responses to the event, a plurality of time series response data vectors from responses elicited over time;
generating, using machine-learning techniques in a computer, a small signal correlation engine, to correlate at least one of the time series stimuli data vectors to one of the time series response data vectors, and generate a plurality of correlation coefficients that correlate of the event notifications, to the aggregated response data;
generating, using machine-learning techniques in a computer, a learning model to simulate variations of the stimuli data based on stimuli and aggregated response data to predict user responses using the correlation coefficients, including computing a contribution value of at least one of the plurality of small signal attributes of at least one of the event notifications;
updating the learning model based on a comparison of the predicted user responses to actual user responses to the event;
generating a media spend allocation for the channel and at least one of a sub-channel associated with the channel or another channel based on the contributions of the event notifications associated with the channel and the at least one of the sub-channel associated with the channel or the other channel and a budget received via a user interface;
displaying the media spend allocation to a user via the user interface;
simulating an effect of an adjustment to the media spend allocation, received via the user interface, based on the predicted user responses simulated by the learning model; and
presenting, via the user interface, the adjusted media spend allocation and predicted user response activity based on the adjustment determined by the simulating.
18. The computer system as set forth in claim 17 , further including filtering the aggregated response data with a transfer function, h(n), based on a geographical origin of the responses, the IP address of the users, a response time, or a response channel.
19. The computer system as set forth in claim 17 , wherein the event notifications include marketing messages deployed across television channels.
20. The computer system as set forth in claim 19 , wherein the small signal attributes include attributes that define a television spot.