Customized content stream utilizing dwelltime-based machine learning

1. A computer-implemented method for selecting content items for presentation on a user device of a user, the method comprising:

measuring dwelltimes for a plurality of content items, the measured dwelltimes based on an amount of time that each of the plurality of content items is determined to have been displayed on the user device, each of the plurality of content items having a plurality of features associated therewith;

training a classifier for determining one or more ranking parameters for the plurality of content items based on the measured dwelltimes and based on the plurality of features associated with the plurality of content items, the training including calculating a relative importance of respective features for increasing measured dwelltimes, wherein the one or more ranking parameters are at least partially based upon the respective feature determined to have the higher relative importance than the other features of the plurality of features for increasing measured dwelltimes; and

sending, subsequent to the training, additional content items to be presented on the user device based on the one or more ranking parameters, wherein the additional content items are associated with said one or more ranking parameters that are expected to increase dwelltimes for the user of the user device.

2. The method as recited in claim 1 , wherein training the classifier further includes:

assigning labels to each content item from the plurality of content items; and

training the classifier based on the assigned labels.

3. The method as recited in claim 2 , wherein assigning labels further includes:

making a content item label equal to the dwelltime for the respective content item.

4. The method as recited in claim 2 , wherein assigning labels further includes:

making a content item label equal to an integer part of a log of a sum of the dwelltime and 1.

5. The method as recited in claim 2 , wherein assigning labels further includes:

making a content item label equal to an integer part of a log of a sum of the dwelltime and 1; and

truncating the content item label to a predetermined value if the content item label is greater than the predetermined value.

6. The method as recited in claim 1 , wherein the plurality of content items includes news articles, videos, slideshows, tweets, blogs, or photographs.

7. The method as recited in claim 1 , wherein features include any detectable property, marker, indicator, category, or attribute of a content item.

8. The method as recited in claim 1 , wherein the one or more ranking parameters are further based on a profile of the user, the profile including features of interest for the user based on items viewed by the user.

9. The method as recited in claim 1 , further including:

training the classifier of content items based on user characteristics, wherein the user characteristics include one or more of gender, or age, or address, or occupation, or online-activity history.

10. The method as recited in claim 1 , wherein the features include one or more of dates, or names, or keywords, or phrases, or people, or publisher, or location.

11. A computer-implemented method for delivering a customized content item stream to a user device of a user, comprising:

measuring dwelltimes for a plurality of content items presented to viewers, the dwelltimes based on an amount of time that each of the plurality of content items is determined to have been displayed on the user device;

obtaining one or more features associated with each of the plurality of content items;

training a classifier for determining one or more ranking parameters for ranking content items based on the measured dwelltimes and based on the one or more features associated with the plurality of content items, the training including calculating relative importance of respective one or more features associated with the plurality of content items for improving dwelltimes for the plurality of content items, wherein the one or more ranking parameters are at least partially based upon the respective feature determined to have the higher relative importance than the other features of the plurality of features for improving dwelltimes; and

sending, subsequent to the training, additional content items to be presented on the user device based on a ranking of the additional content items, wherein the ranking of the additional content items is at least partially based on ranking parameters that are associated with features that are expected to improve dwelltimes for the additional content items for the user of the user device.

12. The computer-implemented method of claim 11 , wherein a feature of the one or more features of the plurality of content items has a higher relative importance for improving dwelltime if a dwelltime is increased for a content item if the feature is present in the content item relative to if the feature is not present in the content item.

13. The computer-implemented method of claim 11 , wherein the ranking of the additional content items is further based on a profile of the user of the user device.

14. The computer-implemented method of claim 11 , wherein the ranking of the additional content items is further based on an expected interest of the user of the user device.

15. The computer-implemented method of claim 11 , wherein an importance of a feature for improving dwelltimes is at least partially based on whether the presence of the feature increases a probability that a dwelltime for a content item will be increased.

16. A non-transitory computer-readable storage medium storing a computer program for sending content items for presentation to a user device of a user, the non-transitory computer-readable storage medium comprising:

program instructions for measuring dwelltimes for a plurality of content items, the measured dwelltimes based on an amount of time that respective content items of the plurality of content items are determined to have been displayed on the user device, each of the plurality of content items having a plurality of features associated therewith;

program instructions for training a classifier for determining one or more ranking parameters for content items based on the measured dwelltimes and based on the plurality of features associated with the plurality of content items, the training including calculating impact factors of respective features for increasing measured dwelltimes, wherein the one or more ranking parameters are at least partially based upon the respective feature determined to have the greater impact factor than the other features of the plurality of features for increasing measured dwelltimes; and

program instructions for sending, subsequent to the training, additional content items to be presented on the user device based on the one or more ranking parameters, wherein the additional content items are associated with said one or more ranking parameters that are expected to increase dwelltimes for the user of the user device.

17. The computer-readable storage medium storing a computer program as recited in claim 16 , wherein training the classifier further includes:

assigning labels to each content item from the plurality of content items, wherein assigning labels includes making a content item label equal to the dwelltime for the respective content item; and

training the classifier based on the assigned labels.

18. The non-transitory computer-readable storage medium storing a computer program as recited in claim 17 , wherein assigning labels further includes:

making a content item label equal to an integer part of a log of a sum of the dwelltime and 1.

19. The non-transitory computer-readable storage medium storing a computer program as recited in claim 17 , wherein assigning labels further includes:

making a content item label equal to an integer part of a log of a sum of the dwelltime and 1; and

truncating the content item label to a predetermined value if the content item label is greater than the predetermined value.

20. The non-transitory computer-readable storage medium storing a computer program as recited in claim 16 , wherein measuring dwelltimes further includes:

measuring the dwelltimes utilizing one of web beacons or utilizing focus and blur of user browsing activities.