Utilizing a bayesian approach and multi-armed bandit algorithms to improve distribution timing of electronic communications

Patent metadata

Field	Value
Publication number	US-11710065-B2
Application number	US-201916371460-A
Country	US
Kind code	B2
Filing date	Apr 1, 2019
Priority date	Apr 1, 2019
Publication date	Jul 25, 2023
Grant date	Jul 25, 2023

Abstract

The present disclosure relates to systems, methods, and non-transitory computer readable media for determining send times to provide electronic communications based on predicted response rates by utilizing a Bayesian approach and multi-armed bandit algorithms. For example, the disclosed systems can generate predicted response rates by training and utilizing one or more response rate prediction models to generate a weighted combination of user-specific response information and population-specific response information. The disclosed systems can further utilize a Bayes upper-confidence-bound send time model to determine send times that are more likely to elicit user responses based on the predicted response rates and further based on exploration and exploitation considerations. In addition, the disclosed systems can update the response rate prediction models and/or the Bayes upper-confidence-bound send time model based on providing additional electronic communications and receiving additional responses to modify model weights.

First claim

What is claimed is:

1. In a digital medium environment for distributing electronic communications, a computer-implemented method of determining electronic communication send times using a Bayesian approach, the computer-implemented method comprising:
identifying one or more responses to a plurality of electronic communications transmitted to a user;
determining a response rate at a target time granularity via a response rate prediction model by:
identifying, based on responses received from a first attribute group that includes the user, a first label rate indicating a ratio between responses received from, and electronic communications delivered to, the first attribute group over a first coarse time granularity coarser than the target time granularity;
identifying, based on responses received from a second attribute group that includes the user, a second label rate indicating a ratio between responses received from, and electronic communications delivered to, the second attribute group over a second coarse time granularity coarser than the target time granularity and the first coarse time granularity; and
determining the response rate at the target time granularity by applying a user weight to the one or more responses, a first weight to the first label rate, and a second weight to the second label rate;

determining a send time based on the response rate utilizing a Bayes upper-confidence-bound send time model; and
providing an electronic communication to the user based on the send time.

2. The computer-implemented method of claim 1 , wherein identifying the first label rate comprises determining a relationship between a number of responses associated with the first attribute group and a number of electronic communications provided to the first attribute group.

3. The computer-implemented method of claim 1 , further comprising:
identifying one or more additional responses to additional electronic communications provided to the user; and
based on the one or more additional responses, modifying one or more of the first weight corresponding to the first coarse time granularity and the first attribute group or the second weight corresponding to the second coarse time granularity and the second attribute group.

4. The computer-implemented method of claim 1 , wherein determining the response rate comprises determining one or more of an open rate, a click rate, or a conversion rate.

5. The computer-implemented method of claim 4 , further comprising receiving an input from an administrator device to define the response rate as one or more of the open rate, the click rate, or the conversion rate.

6. A non-transitory computer readable medium comprising instructions that, when executed by at least one processor, cause a computer device to:
identify one or more responses to a plurality of electronic communications transmitted to a user;
determine a response rate at a target time granularity via a response rate prediction model by:
identifying, based on responses received from a first attribute group that includes the user, a first label rate indicating a ratio between responses received from, and electronic communications delivered to, the first attribute group over a first coarse time granularity coarser than the target time granularity;
identifying, based on responses received from a second attribute group that includes the user, a second label rate indicating a ratio between responses received from, and electronic communications delivered to, the second attribute group over a second coarse time granularity coarser than the target time granularity and the first coarse time granularity; and
determining the response rate at the target time granularity by applying a user weight to the one or more responses, a first weight to the first label rate, and a second weight to the second label rate;

determine a send time based on the response rate utilizing a Bayes upper-confidence-bound send time model; and
provide an electronic communication to the user based on the send time.

7. The non-transitory computer readable medium of claim 6 , further comprising instructions that, when executed by the at least one processor, cause the computer device to:
determine, utilizing the Bayes upper-confidence-bound send time model, a first quantile score for a first send time based on a predicted response rate for the first send time and a number of messages the user has received at the first send time;
determine, utilizing the Bayes upper-confidence-bound send time model, a second quantile score for a second send time based on a predicted response rate for the second send time and a number of messages the user has received at the second send time; and
select the send time by comparing the first quantile score and the second quantile score.

8. The non-transitory computer readable medium of claim 7 , further comprising instructions that, when executed by the at least one processor, cause the computer device to:
identify one or more additional responses to additional electronic communications provided to the user; and
based on the one or more additional responses, reduce the first weight corresponding to the first coarse time granularity and the first attribute group and the second weight corresponding to the second coarse time granularity and the second attribute group.

9. The non-transitory computer readable medium of claim 7 , further comprising instructions that, when executed by the at least one processor, cause the computer device to:
identify one or more additional responses to additional electronic communications provided to the user; and
based on the one or more additional responses, increase the user weight applied to the one or more responses.

10. The non-transitory computer readable medium of claim 6 , wherein identifying the first label rate comprises determining a relationship between a number of responses associated with the first attribute group and a number of electronic communications provided to the first attribute group.

11. The non-transitory computer readable medium of claim 6 , further comprising instructions that, when executed by the at least one processor, cause the computer device to determine, via the response rate prediction model, predicted response rates associated with a plurality of time values within the target time granularity.

12. The non-transitory computer readable medium of claim 11 , wherein the send time comprises a time to provide the electronic communication to the user and that has a highest response rate from among the plurality of time values within the target time granularity.

13. The non-transitory computer readable medium of claim 6 , wherein the user weight comprises a quantity of the plurality of electronic communications transmitted to the user.

14. The non-transitory computer readable medium of claim 6 , wherein identifying the first label rate corresponding to the first coarse time granularity and the first attribute group comprises:
determining an initial first label rate that indicates an initial ratio between a number of responses received from the first attribute group for the first coarse time granularity and a number of electronic communications sent to the first attribute group for the first coarse time granularity; and
applying a Bayesian smoothing factor to the initial first label rate to generate a smoothed label rate.

15. The non-transitory computer readable medium of claim 6 , wherein identifying the first label rate corresponding to the first coarse time granularity and the first attribute group comprises:
determining an initial first label rate that indicat

Assignees

• Adobe Inc

Inventors

• He Jun
• Gu Shiyuan
• Yan Zhenyu
• Dai Wuyang
• Kuo Yi-Hong
• Pani Abhishek

Classifications

• G06N20/00 Primary

  Machine learning · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06F18/24155

  Bayesian classification · CPC title

• G06F40/279

  Recognition of textual entities · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title