Facilitating machine learning in an online social network

1. A computer-implemented method for facilitating an automatic machine-learning process for an online social network, the method comprising:

during operation of the online social network, automatically collecting labeled training events;

snapshotting raw entity data associated with subjects from the collected training events;

generating features for each training event using the snapshotted entity data and the current entity data to produce training examples;

consolidating the training examples from one or more time periods to produce a consolidated set of training examples;

determining one or more contradictory user actions upon a set of entity data in the consolidated set of training examples, wherein a contradictory user action involves a same user responding differently to the same entity data over different time intervals;

resolving the contradictory user actions for the consolidated set of training examples;

splitting the training examples into a training set and a test set;

using a machine-learning technique to train a set of models and select the best model based on one or more evaluation metrics using the training set;

evaluating the performance of the best model on the test set; and

if the performance of the best model satisfies a performance criterion, using the best model to predict responses for the online social network.

2. The computer-implemented method of claim 1 , further comprising repeating the method to continually update a model that predicts responses for the online social network.

3. The computer-implemented method of claim 1 , wherein automatically collecting the labeled training events includes performing sampling operations to select training events for a response to be predicted.

4. The computer-implemented method of claim 1 , wherein before the training examples are split, the method further comprises:

performing sampling operations on the consolidated set of training examples to achieve a balanced ratio among different responses for the training examples.

5. The computer-implemented method of claim 1 ,

wherein evaluating the best model involves using quantitative and qualitative measures; and

wherein using a qualitative measure includes using a crowd-sourcing evaluation technique that comprises:

sending a selected set of training examples and associated responses that were produced by the best model to a crowd-sourcing module that solicits the opinions of an additional set of users about the accuracy of the responses; and

comparing the predictions of the best model with predictions of the additional set of users on the selected training examples.

6. The computer-implemented method of claim 1 , wherein the response to be predicted is associated with one or more of:

a classification of an item;

a prediction of a user preference; and

a prediction of a user action.

7. The computer-implemented method of claim 1 , wherein the user-defined features are associated with one or more of:

information from a member's profile;

a member's actions; and

information about an item.

8. The computer-implemented method of claim 1 , wherein the online social network is an online professional network.

9. A non-transitory computer-readable storage medium storing instructions that when executed by a computer cause the computer to perform a method for facilitating an automatic machine-learning process for an online social network, the method comprising:

during operation of the online social network, automatically collecting labeled training events;

snapshotting raw entity data associated with subjects from the collected training events;

generating features for each training event using the snapshotted entity data and the current entity data to produce training examples;

consolidating the training examples from one or more time periods to produce a consolidated set of training examples;

determining one or more contradictory user actions upon a set of entity data in the consolidated set of training examples, wherein a contradictory user action involves a same user responding differently to the same entity data over different time intervals;

resolving the contradictory user actions for the consolidated set of training examples;

splitting the training examples into a training set and a test set;

using a machine-learning technique to train a set of models and select the best model based on one or more evaluation metrics using the training set;

evaluating the performance of the best model on the test set; and

if the performance of the best model satisfies a performance criterion, using the best model to predict responses for the online social network.

10. The non-transitory computer-readable storage medium of claim 9 , further comprising repeating the method to continually update a model that predicts responses for the online social network.

11. The non-transitory computer-readable storage medium of claim 9 , wherein automatically collecting labeled training events includes performing sampling operations to select training events for a response to be predicted.

12. The non-transitory computer-readable storage medium of claim 9 , wherein before the training examples are split, the method further comprises:

performing sampling operations on the consolidated set of training examples to achieve a balanced ratio among different responses for the training examples.

13. The non-transitory computer-readable storage medium of claim 9 ,

wherein evaluating the best model involves using quantitative and qualitative measures; and

wherein using a qualitative measure includes using a crowd-sourcing evaluation technique that comprises:

sending a selected set of training examples and associated responses that were produced by the best model to a crowd-sourcing module that solicits the opinions of an additional set of users about the accuracy of the responses; and

comparing the predictions of the best model with predictions of the additional set of users on the selected training examples.

14. The non-transitory computer-readable storage medium of claim 9 , wherein the response to be predicted is associated with one or more of:

a classification of an item;

a prediction of a user preference; and

a prediction of a user action.

15. The non-transitory computer-readable storage medium of claim 9 , wherein the user-defined features are associated with one or more of:

information from a member's profile;

a member's actions; and

information about an item.

16. The non-transitory computer-readable storage medium of claim 9 , wherein the online social network is an online professional network.

17. A system that facilitates an automatic machine-learning process for an online social network, comprising:

a computing system including a processor and a memory;

wherein the computing system is configured to run an online social network; and

wherein the online social network is configured to,

automatically collect labeled training events;

snapshot raw entity data associated with subjects from the collected training events;

generate features for each training event using the snapshotted entity data and the current entity data to produce training examples;

consolidate the training examples from one or more time periods to produce a consolidated set of training examples;

determine one or more contradictory user actions upon a set of entity data in the consolidated set of training examples, wherein a contradictory user action involves a same user responding differently to the same entity data over different time intervals;

resolve the contradictory user actions for the consolidated set of training examples;

split the training examples into a training set and a test set;

use a machine-learning technique to train a set of models and select the best model based on one or more evaluation metrics using the training set;

evaluate the performance of the best model on the test set; and

if the performance of the best model satisfies a performance criterion, use the best model to predict responses for the online social network.

18. The system of claim 17 , wherein the online social network is configured to repeat the process of collecting, splitting, training, selecting and evaluating to continually update a model that predicts responses for the online social network.

19. The system of claim 17 , wherein while collecting labeled training events, the online social network is configured to perform sampling operations to select training events for the response to be predicted.

20. The system of claim 17 , wherein before the training examples are split, the system is configured to:

perform sampling operations on the consolidated set of training examples to achieve a balanced ratio among different responses for the training examples.

21. The system of claim 17 ,

wherein while evaluating the best model the online social network is configured to use quantitative and qualitative measures; and

wherein using a qualitative measure includes using a crowd-sourcing evaluation technique that comprises:

sending a selected set of training examples and associated responses that were produced by the best model to a crowd-sourcing module that solicits the opinions of an additional set of users about the accuracy of the responses; and

comparing the predictions of the best model with predictions of the additional set of users on the selected training examples.

22. The system of claim 17 , wherein the response to be predicted is associated with one or more of:

a classification of an item;

a prediction of a user preference; and

a prediction of a user action.

23. The system of claim 17 , wherein the user-defined features are associated with one or more of:

information from a member's profile;

a member's actions; and

information about an item.