DISTRIBUTED PERSONAL SPAM FILTERING

• US 20090287618A1
• Filed: 05/19/2008
• Published: 11/19/2009
• Est. Priority Date: 05/19/2008
• Status: Active Grant

• Current Assignee: R2 Solutions LLC (Acacia Research Corporation)
• Original Assignee: Yahoo! Inc. (Apollo Global Management, Inc.)
• Inventors: Weinberger, Kilian Quirin, Langford, John
• Granted Patent: US 8,108,323 B2
• Time in Patent Office: Days
• US Class Current: 706/10
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: H04L 51/212 using filtering or selective blocking

Abstract

Embodiments are directed towards using a community of weighted results from local and global message classifiers to determine whether a message is spam. Each local classifier may receive a message that is to be evaluated to determine whether it is spam. A local classifier receives the message and performs a classification of the message. The local classifier may receive predictions of whether the message is spam from at least one global classifier. The local and global predictions are combined using, in one embodiment, a regression analysis to generate a single local message classification. Combining the local and global predictions is directed towards enabling a community of predictions to be used to classify messages. The user may then re-classify this output, which in turn is used as feedback to modify weights to the local and received global predictions for a next message.

Claims

1. A network device to manage a delivery of a message over a network, comprising:

a transceiver to send and receive data over a network; and

a processor that is operative to perform actions, comprising;

receiving a message over the network for at least a first user account;

providing the message to a plurality of global classifiers, wherein each of the global classifiers is associated with user accounts other than the first user account;

providing the message to a local classifier associated with the first user account;

receiving a message prediction from each of the plurality of global classifiers and the local classifier;

combining the message predictions from each of the plurality of global classifiers with the local classifier to determine a single message classification as to if the message is a spam message for the first user account; and

selectively directing the message to the first user account based on the determined single message classification.

2. The network device of claim 1, wherein the processor is operative to perform actions, farther comprising:

receiving feedback from a user of the first user account about the single message classification;

modifying a weighting associated with a message prediction from at least one a global classifier or the local classifier; and

modifying the combining of the message predictions based on the modified weighting.

3. The network device of claim 1, wherein the processor is operative to perform actions, further comprising:

receiving feedback from a user of the first user account about the single message classification; and

training the local classifier based on the received feedback from the user.

4. The network device of claim 1, wherein at least one of the local prediction or the single message classification is sent to at least one of the global classifiers for use in determining another single message classification for another message associated with the at least one of the global classifiers.

5. The network device of claim 1, wherein combining the message predictions further comprises employing at least one of a regression analysis.

6. A processor readable storage medium that includes data and instructions, wherein the execution of the instructions on a computing device provides for managing messages over a network by enabling actions, comprising:

receiving a message for a first user account over the network;

providing the message to a plurality of global classifiers, wherein each of the global classifiers are associated with user accounts other than the first user account, and wherein the message is provided to the plurality of global classifiers transparent to each user associated with a respective global classifier;

providing the message to a local classifier associated with the first user account;

receiving a message prediction from each of the plurality of global classifiers and the local classifier;

combining weighted message predictions from each of the plurality of global classifiers with the local classifier to determine a single message classification as to if the message is a spam message for the first user account;

selectively directing the message to the first user account based on the determined single message classification; and

modifying the weighting of the message predictions based on a user feedback of the determined single message classification.

7. The processor readable storage medium of claim 6, wherein at least one of a global classifier employ a different message classification mechanism from the local classifier, selected from at least one of a naï

ve Bayes algorithm, a support vector machine algorithm, a logistic regression, a linear regression, neural network, decision tree, or a perceptron.

8. The processor readable storage medium of claim 6, wherein the determined single message classification from the local classifier is provided to a global classifier of the plurality of global classifiers for use in classifying at least one other message directed towards a second account associated with the global classifier.

9. The processor readable storage medium of claim 6, wherein the user feedback is further employed to train the local classifier for use in classifying another message.

10. The processor readable storage medium of claim 6, wherein selectively directing the message further comprises directing the message to at least one of a bulk message folder, an inbox folder, a spam message folder, or a junk message folder.

11. The processor readable storage medium of claim 6, wherein the local classifier is configured and arranged to be trained based on a plurality of other messages and a plurality of other user feedback associated with the first user account.

12. A method for managing a message delivery over a network, comprising:

receiving a message for a first user account over the network;

providing the message to a local classifier associated with the first user account;

transparently providing the message to a plurality of global classifiers, wherein each of the global classifiers is associated with user accounts other than the first user account;

receiving a message prediction from each of the plurality of global classifiers and the local classifier;

weighting each message prediction from each of the plurality of global classifiers and the local classifier, wherein the weightings are determined in part based on feedback from a user associated with the first user account;

combining the weighted message predictions to determine a single message classification indicating if the message is a spam message for the first user account; and

selectively directing the message to a message folder for the first user account based on the determined single message classification.

13. The method of claim 12, wherein combining the weighted message predictions further comprises employing a regression analysis.

14. The method of claim 12, wherein the message prediction from the local classifier is sent to at least one global classifier in the plurality of classifiers.

15. The method of claim 12, wherein one global classifier employs a different message classification mechanism than the local classifier.

16. The method of claim 12, further comprising:

modifying at least one parameter for the local classifier based on feedback from the user associated with the first user account based on the single message classification, wherein the at least one parameter is useable by the local classifier to determine another message prediction.

17. A mobile device for enabling a communications over a network, comprising:

a memory arranged to store data and instructions;

an input interface for receiving requests and sending responses; and

a processor arranged to enable actions embodied by at least a portion of the stored instructions, the actions comprising;

receiving a message for a first user account over the network;

providing the message to a local classifier associated with the first user account;

transparently providing the message to a plurality of global classifiers, wherein each of the global classifiers is associated with user accounts other than the first user account;

receiving a message prediction from each of the plurality of global classifiers and the local classifier;

weighting each message prediction from each of the plurality of global classifiers and the local classifier, wherein the weightings are determined in part based on feedback from a user associated with the first user account;

combining the weighted message predictions to determine a single message classification indicating if the message is a spam message for the first user account; and

selectively directing the message to a message folder for the first user account based on the determined single message classification such that a user of the first user account is enabled to view the message at a display of the mobile device.

18. The mobile device of claim 17, wherein at least one of the plurality of global classifiers is associated with another user account other than the first user account that accessible through another mobile device.

19. The mobile device of claim 17, wherein the processor is arranged to enable further actions, comprising:

receiving at the mobile device a modification to the determined single message classification by a user of the mobile device associated with the first user account; and

employing the modification to modify a weighting to at least one message prediction from at least one of the local classifier or at least one global classifier.

20. The mobile device of claim 17, wherein the processor is arranged to enable further actions, comprising:

receiving at the mobile device a modification to the determined single message classification by a user of the mobile device associated with the first user account; and

using the modification to train the Local classifier.