Determination of user perspicaciousness during a content stream

1. A method comprising:

receiving a request comprising a join timestamp representing when a target user joined a live stream and a request time point representing a break in the live stream;

applying a predictive model to a join duration that represents a difference between the join timestamp and the request time point to generate a predicted attention score for the target user, the predicted attention score representing a perspicaciousness of a predicted user at the request time point;

determining whether the predicted attention score is greater than a threshold score; and

responsive to the determination that the predicted attention score is greater than the threshold score, retrieving a content item from a first content category of a plurality of content categories; and

providing the retrieved content item to the target user during the break in the live stream.

2. The method of claim 1 , further comprising:

determining an updated attention score for the target user corresponding to the request time point;

comparing the updated attention score to the predicted attention score; and

determining whether to update the predictive model based on the comparison.

3. The method of claim 2 , wherein determining the updated attention score comprises:

receiving one or more user actions performed by a user on a content item presented on a client device to the user during the break in the live stream; and

generating the updated attention score based on the one or more user actions.

4. The method of claim 3 , wherein each of the one or more user actions is one of a click on the content item, a conversion, a skip of the content item, a change of volume on the client device, a change in orientation of the client device.

5. The method of claim 3 , wherein generating the updated attention score based on the one or more user actions comprises:

for each of the one or more user actions:

assigning a weight to the user action; and

adjusting the updated attention score based on the weight assigned to the user action.

6. The method of claim 2 , wherein determining whether to update the predictive model based on the comparison comprises:

determining a difference between the updated attention score and the predicted attention score; and

comparing the determined difference to a threshold value.

7. The method of claim 2 , wherein determining an updated attention score for the target user comprises:

receiving a listening pattern of the target user;

evaluating the join timestamp and the request time point received in the request based on the listening pattern; and

adjusting the updated attention score based on the evaluation.

8. The method of claim 1 , wherein the predictive model describes an exponential decay relationship between user attention score and a user join duration.

9. The method of claim 1 further comprising:

subsequent to receiving the request, identifying a characteristic of the target user; and

retrieving a predictive model generated for a target audience with the identified characteristic of the target user.

10. The method of claim 1 , wherein the predictive model is generated by:

receiving a plurality of experimental training examples, each training example comprising a join duration representing a duration of time between when a user joined a prior live stream and a break in the prior live stream;

generating, for each experimental training example, an experimental attention score for the user, the experimental attention score associated with the join duration; and

generating the predictive model based on a plurality of the generated experimental attention scores.

11. The method of claim 10 , wherein each training example further comprises one or more user actions performed by the user on a content item presented on a client device to the user during the break in the prior live stream.

12. The method of claim 11 , wherein generating the experimental attention score for the user comprises:

for each of the one or more user actions in the training example:

assigning a weight to the user action; and

adjusting the experimental attention score based on the weight assigned to the user action.

13. The method of claim 10 , wherein generating the predictive model comprises fitting an exponentially decaying curve describing a relationship between the plurality of experimental attention scores and the join duration associated with a training example that each experimental attention score was calculated from.

14. A non-transitory computer-readable medium comprising computer code that, when executed by a processor, causes the processor to:

receive a request comprising a join timestamp representing when a target user joined a live stream and a request time point representing a break in the live stream;

apply a predictive model to a join duration that represents a difference between the join timestamp and the request time point to generate a predicted attention score for the target user, the predicted attention score representing a perspicaciousness of a predicted user at the request time point;

determine whether the predicted attention score is greater than a threshold score; and

responsive to the determination that the predicted attention score is greater than the threshold score, retrieve a content item from a first content category of a plurality of content categories; and

provide the retrieved content item to the target user during the break in the live stream.

15. The non-transitory computer-readable medium of claim 14 further comprising computer code that, when executed by the processor, causes the processor to:

determine an updated attention score for the target user corresponding to the request time point;

compare the updated attention score to the predicted attention score; and

determine whether to update the predictive model based on the comparison.

16. The non-transitory computer-readable medium of claim 15 , wherein the computer code that causes the processor to determine the updated attention score further comprises computer code that, when executed by the processor, causes the processor to:

receive one or more user actions performed by a user on a content item presented on a client device to the user during the break in the live stream; and

generate the updated attention score based on the one or more user actions.

17. The non-transitory computer-readable medium of claim 16 , wherein each of the one or more user actions is one of a click on the content item, a conversion, a skip of the content item, a change of volume on the client device, a change in orientation of the client device.

18. The non-transitory computer-readable medium of claim 15 , wherein the computer code that causes the processor to determine whether to update the predictive model based on the comparison further comprises computer code that, when executed by the processor, causes the processor to:

determine a difference between the updated attention score and the predicted attention score; and

compare the determined difference to a threshold value.

19. The non-transitory computer-readable medium of claim 14 , wherein the predictive model describes an exponential decay relationship between user attention score and a user join duration.

20. The non-transitory computer-readable medium of claim 14 , further comprising computer code that, when executed by the processor, causes the processor to generate the predictive model, wherein the computer code that causes the processor to generate the predictive model further comprises computer code that, when executed by the processor, causes the processor to:

receive a plurality of experimental training examples, each training example comprising a join duration representing a duration of time between when a user joined a prior live stream and a break in the prior live stream;

generate, for each experimental training example, an experimental attention score for the user, the experimental attention score associated with the join duration; and

generate the predictive model based on a plurality of the generated experimental attention scores.