Online user space exploration for recommendation
To maximize the accuracy and efficiency of predicting users that will enjoy targeted content, a proposed content selection solution looks to combine a first strategy of utilizing selection rules with a second strategy of utilizing machine based learning models. By combining the selection rules-based approach and the machine learning model-based approach, the proposed content selection solution is able to consider and recommend a wider range of users for each available content.
1. A computing device comprising:
a communication interface configured to receive user information including user online interaction data;
a memory configured to store a selection rule describing parameters for including one or more users into a user set;
a processor configured to:
receive the user information through the communication interface;
retrieve the selection rule from the memory;
create a first predictor protocol grouping one or more users into a first user set based on the selection rule;
create a second predictor protocol grouping one or more users into a second user set based on a machine learning technique; and
generate a user set proposal including the first user set and the second user set.
2. The computing device of claim 1 , wherein the processor is further configured to:
monitor a first performance of a first user included in the first user set interacting with a targeted content, the first performance monitoring a number of conversions of the targeted content by the first user;
monitor a second performance of a second user included in the second user set interacting with the targeted content, the second performance monitoring a number of conversions of the targeted content by the second user;
compare the first performance to the second performance; and
select one of the first user set or the second user set based on the comparison of the first performance and the second performance according to a selection strategy.
3. The computing device of claim 2 , wherein a conversion of the targeted content includes either the first user or the second user clicking on the targeted content.
4. The computing device of claim 2 , wherein the selection strategy is a multi-armed bandit selection strategy.
5. The computing device of claim 1 , wherein the processor is further configured to:
create a third user set comprised of users included in both the first user set and the second user set.
6. The computing device of claim 1 , wherein the selection rule selects users within an overall user set that satisfies at least one of a user location requirement, user attribute requirement, or user web browser history requirement.
7. The computing device of claim 1 , wherein the first user set and the second user set include users from within an overall user set monitored by a common website publisher.
8. The computing device of claim 1 , wherein the machine learning technique outputs a respective conversion score based on conversion of targeted content presented to a user; and
wherein the processor is further configured to select a user to group into the second user set when the user has a conversion score greater than a predetermined threshold value.
9. The computing device of claim 1 , wherein the processor is further configured to:
detect a user visiting a website operated by a common website publisher;
determine whether to implement a cold start scenario based on an impression count of a targeted content being greater than a predetermined threshold; and
present the targeted content to the user when determined to implement the cold start scenario.
10. The computing device of claim 9 , wherein the processor is further configured to:
implement a warm start scenario when the impression count is less than the predetermined threshold;
select either the first user set or the second user set based on a Thompson Sampling selection strategy;
determine whether the user is included in either the first user set or the second user set; and
present the targeted content to the user when the user is included in the first user set or the second user set.
11. A method comprising:
receiving, by a processor, user information through a communication interface;
retrieving, by the processor, a selection rule stored in a memory;
creating, by the processor, a first predictor protocol grouping one or more users into a first user set based on the selection rule;
creating, by the processor, a second predictor protocol grouping one or more users into a second user set based on a machine learning technique; and
generating, by the processor, a user set proposal including the first user set and the second user set.
12. The method of claim 11 , further comprising:
monitoring, by the processor, a first performance of a first user included in the first user set interacting with a targeted content, the first performance monitoring a number of conversions of the targeted content by the first user;
monitoring, by the processor, a second performance of a second user included in the second user set interacting with the targeted content, the second performance monitoring a number of conversions of the targeted content by the second user;
comparing, by the processor, the first performance to the second performance; and
selecting, by the processor, one of the first user set or the second user set based on the comparison of the first performance and the second performance according to a selection strategy.
13. The method of claim 12 , converting the targeted content includes either the first user or the second user clicking on the targeted content.
14. The method of claim 12 , wherein the selection strategy is a multi-armed bandit selection strategy.
15. The method of claim 11 , further comprising:
creating, by the processor, a third user set comprised of users included in both the first user set and the second user set.
16. The method of claim 11 , wherein the selection rule selects users within an overall user set that satisfies at least one of a user location requirement, user attribute requirement, or user web browser history requirement.
17. The method of claim 11 , wherein the first user set and the second user set include users from within an overall user set monitored by a common website publisher.
18. The method of claim 11 , wherein the machine learning technique outputs a respective conversion score based on conversion of targeted content presented to a user; and
wherein the method further comprises selecting a user to group into the second user set when the user has a conversion score greater than a predetermined threshold value.
19. The method of claim 11 , further comprising:
detecting, by the processor, a user visiting a website operated by a common website publisher;
determining, by the processor, whether to implement a cold start scenario based on an impression count of a targeted content being greater than a predetermined threshold; and
presenting, by the processor, the targeted content to the user when determined to implement the cold start scenario.
20. The method of claim 19 , further comprising:
implementing, by the processor, a warm start scenario when the impression count is less than the predetermined threshold;
selecting, by the processor, either the first user set or the second user set based on a Thompson Sampling selection strategy;
determining, by the processor, whether the user is included in either the first user set or the second user set; and
presenting, by the processor, the targeted content to the user when the user is included in the first user set or the second user set.