Inferring latency sensitivity of user activity

1. A method for identifying latency preferences of users, comprising:

determining a biased distribution of latency of a plurality of user actions for an application over a timeframe, wherein the biased distribution of latency includes a latency for each user action of the plurality of user actions;

inferring an unbiased distribution of latency based on the biased distribution of latency by selecting random times within the timeframe of the biased distribution of latency and using the latency for a user action at or close to the chosen random times for the unbiased distribution of latency;

computing a latency preference of the plurality of user actions as a ratio of a probability density function of the biased distribution of latency and a probability density function of the unbiased distribution of latency;

computing a normalized latency preference by dividing the latency preference by the latency preference corresponding to a reference latency; and

outputting the normalized latency preference as a function of latency.

2. The method of claim 1 , wherein the latency of the plurality of user actions is obtained from a latency log from a server.

3. The method of claim 2 , wherein the latency log aggregates the plurality of user actions from a plurality of users interacting with the application.

4. The method of claim 2 , wherein the latency log indicates a time when the user actions occurred.

5. The method of claim 2 , wherein the latency log further includes:

metadata of a plurality of users or metadata for the plurality of user actions, wherein the metadata is obtained from one or more of user profile information, a context of the plurality of users, or an action type of the plurality of user actions.

6. The method of claim 5 , wherein the action type identifies a type of the plurality of user actions and the user profile information identifies different groups of the plurality of users.

7. The method of claim 6 , wherein different normalized latency preferences are generated for the different groups of the plurality of users.

8. The method of claim 5 , wherein different normalized latency preferences are generated for different action types of the plurality of user actions.

9. The method of claim 5 , wherein the context of the plurality of users is used to identify one or more of a location where the plurality of user actions occurred, a network connectivity of the users, or a type of user, and

wherein different normalized latency preferences are generated based on one or more of the location where the plurality of user actions occurred, the network connectivity of the users, or the type of user.

10. The method of claim 1 , further comprising:

adjusting the normalized latency preference based on at least one confounding factor by modifying the normalized latency preference to mitigate an effect of the at least one confounding factor on the normalized latency preference,

wherein the at least one confounding factor includes a time-based activity factor, content driven user activity preference, or previous user conditioning.

11. The method of claim 1 , further comprising:

generating one or more recommendations for the application functionality to be prioritized for latency improvement, based on analyzing the normalized latency preference.

12. The method of claim 11 , wherein the one or more recommendations identify one or more areas of the application to modify or change to reduce an amount of latency for the application.

13. The method of claim 11 , wherein analyzing the normalized latency preference is based on analyzing different normalized latency preferences of different groups of users for an action type of the plurality of user actions.

14. The method of claim 11 , wherein analyzing the normalized latency preference is based on analyzing different normalized latency preferences for a combination of different groups of users and different action types of the plurality of user actions.

15. The method of claim 10 , wherein analyzing the normalized latency preference is based on analyzing the normalized latency preference of a plurality of users for the plurality of user actions during different times of day.

16. A method for determining an unbiased latency, comprising:

obtaining a biased distribution of latency that includes a plurality of user actions with an associated latency over a timeframe for an application;

selecting random points in time of the timeframe;

identifying latency samples from the plurality of user actions in the biased distribution of latency based on the points in time selected; and

inferring an unbiased distribution of latency for the application using the latency samples.

17. The method of claim 16 , wherein the biased distribution of latency is based on logs of user actions from a plurality of users interacting with the application.

18. The method of claim 16 , wherein identifying the latency samples includes:

identifying a measurement from the plurality of user actions that is closest in time to a selected point in time and using the latency of a nearest user action as a latency sample for the selected point in time.

19. The method of claim 16 , wherein identifying the latency sample includes:

identifying one or more user actions of the plurality of user actions that are close in time to a selected point in time and taking an average latency of the one or more user actions as a latency sample for the selected point in time.

20. The method of claim 16 , wherein identifying the latency sample includes:

identifying one or more user actions of the plurality of user actions that are close in time to a selected point in time and randomly selecting a latency of the one or more user actions as a latency sample for the selected point in time.