Graymail filtering-based on user preferences
A graymail detection and filtering system predicts whether a user will consider an email to be graymail using a classifier model based on features extracted from the email. The email is labelled as graymail or non-graymail based on the prediction. User actions are tracked on the email to determine whether the user actually considered the email to be graymail or non-graymail and the classifier model is trained using machine learning techniques to improve the prediction, without requiring explicit user feedback on whether the user considered the email to be graymail or non-graymail.
1. A method of detecting graymail without explicit user feedback, comprising:
receiving a non-spam email for a user;
extracting features of the email;
passing the extracted features to a classifier model;
predicting with the classifier model whether the user would consider the email as graymail;
modifying the email by inserting tracking information before delivering the email to the user;
storing the inserted tracking information in a tracking database;
tracking user actions on the email without explicit user feedback using the inserted tracking information;
updating the tracking database upon detection of tracked user actions on the email;
determining whether the user considered the email as graymail responsive to the tracked user actions without explicit user feedback; and
training the classifier model responsive to the tracked user actions.
2. The method of claim 1 , wherein the classifier model is a regression model.
3. The method of claim 1 , wherein predicting with the classifier model whether the user would consider the email as graymail comprises:
labelling the email as graymail or non-graymail based on the classifier model.
4. The method of claim 1 , wherein tracking in the user actions on the email without explicit user feedback using the inserted tracking information comprises:
detecting user actions on the email using the inserted tracking information without explicit user feedback; and
identifying the detected user actions,
wherein updating the tracking database upon detection of tracked user actions on the email comprises storing information about the identified detected user actions in the tracking database.
5. The method of claim 1 , wherein determining whether the user considered the email as graymail comprises:
detecting user actions on the email without receiving explicit user feedback;
analyzing the detected user actions; and
determining whether the detected user actions indicate the user considers the email as graymail or non-graymail.
6. The method of claim 1 , further comprising:
storing the extracted features in a database.
7. The method of claim 6 , wherein training the classifier model comprises:
storing information about whether the user considered the email as graymail in the database; and
performing supervised learning in the classifier model using the stored extracted features and stored information about whether the user considered the email as graymail.
8. The method of claim 1 , wherein tracking user actions on the email without explicit user feedback using the inserted tracking information comprises:
identifying the email with a unique identification before providing the email to the user; and
associating the user actions with the unique identification of the email.
9. A non-transitory computer readable medium, on which is stored software for detecting graymail without explicit user feedback, comprising instructions that when executed by a graymail detection system cause the graymail detection system to:
receive a non-spam email for a user;
extract features of the email;
predict with a classifier model based on the extracted features whether the user would consider the email as graymail;
modifying the email by inserting tracking information before delivering the email to the user;
storing the inserted tracking information in a tracking database;
track user actions on the email without explicit user feedback using the inserted tracking information;
updating the tracking database upon detection of tracked user actions on the email;
determine whether the user considered the email as graymail responsive to the tracked user actions without explicit user feedback; and
train the classifier model responsive to the tracked user actions.
10. The computer readable medium of claim 9 , wherein the classifier model is a support vector machine model.
11. The computer readable medium of claim 9 , wherein the instructions that when executed cause the graymail detection system to predict with the classifier model whether the user would consider the email as graymail comprise instructions that when executed cause the graymail detection system to:
label the email as graymail or non-graymail based on the classifier model.
12. The computer readable medium of claim 9 , wherein the instructions that when executed cause the graymail detection system to track user actions on the email without explicit user feedback using the inserted tracking information comprise instructions that when executed cause the graymail detection system to:
detect user actions on the email using the inserted tracking information without explicit user feedback; and
identify the detected user actions,
wherein the instructions that when executed cause the graymail detection system to update the tracking database upon detection of tracked user actions on the email comprise instructions that when executed cause the graymail detection system to store information about the identified detected user actions in the tracking database.
13. The computer readable medium of claim 9 , wherein the instructions that when executed cause the graymail detection system to determine whether the user considered the email as graymail comprise instructions that when executed cause the graymail detection system to:
detect user actions on the email without explicit user feedback;
analyze the detected user actions; and
determine whether the user considered the email as graymail or non-graymail responsive to the analysis.
14. The computer readable medium of claim 9 , wherein the instructions further comprise instructions that when executed cause the graymail detection system to:
store the extracted features in a database.
15. The computer readable medium of claim 14 , wherein the instructions that when executed cause the graymail detection system to train the classifier model comprise instructions that when executed cause the graymail detection system to:
store information about whether the user considered the email as graymail in the database; and
perform machine learning in the classifier model using the stored extracted features and stored information about whether the user considered the email as graymail.
16. The computer readable medium of claim 9 , wherein the instructions that when executed cause the graymail detection system to track user actions on the email without explicit user feedback using the inserted tracking information comprise instructions that when executed cause the graymail detection system to:
identify the email with a unique identification before providing the email to the user; and
associate the user actions with the unique identification of the email.
17. A system for detecting graymail without explicit user feedback, comprising:
a data analytics computer system, comprising:
a processing element; and
a memory, coupled to the processing element, on which is stored software for predicting whether a user would consider an email graymail, comprising instructions that when executed cause the processing element of the data analytics computer system to:
receive a non-spam email for a user;
extract features of the email;
predict with a classifier model based on the extracted features whether the user would consider the email as graymail; and
train the classifier model responsive to tracked user actions; and
a tracking engine computer system, comprising:
a processing element;
a tracking database; and
a memory coupled to the processing element, on which is stored software for tracking user actions on the email, comprising instructions that when executed cause the processing element of the tracking engine computer system to:
modify the email by inserting tracking information before delivering the email to the user;
storing the inserted tracking information in a tracking database;
track user actions on the email without explicit user feedback using the inserted tracking information;
update the tracking database upon detection of tracked user actions on the email;
determine whether the user considered the email as graymail responsive to the tracked user actions without explicit user feedback; and
provide information regarding the determination to the data analytics computer system for training the classifier model.
18. The system of claim 17 , wherein the classifier model is a regression model.
19. The system of claim 17 , wherein the instructions that when executed cause the processing element of the data analytics computer system to predict with the classifier model based on the extracted features whether the user would consider the email as graymail comprise instructions that when executed cause the processing element of the data analytics computer system to:
label the email as graymail or non-graymail based on the classifier model.
20. The system of claim 17 , wherein the instructions that when executed cause the processing element of the tracking engine computer system to track user actions on the email without explicit user feedback comprise instructions that when executed cause the processing element of the tracking engine computer system to:
detect user actions on the email using the inserted tracking information without explicit user feedback; and
identify the detected user actions,
wherein the instructions that when executed cause the processing element of the tracking engine computer system to update the tracking database comprise instructions that when executed cause the processing element of the tracking engine computer system to store information about the identified detected user actions in the tracking database.
21. The system of claim 17 , wherein the instructions that when executed cause the processing element of the tracking engine computer system to determine whether the user considered the email as graymail comprise instructions that when executed cause the processing element of the tracking engine computer system to:
detect user actions on the email;
analyze the detected user actions; and
determine whether the user considered the email as graymail or non-graymail responsive to the analysis.
22. The system of claim 17 , wherein the instructions further comprise instructions that when executed cause the processing element of the data analytics computer system to:
store the extracted features in a database.
23. The system of claim 22 , wherein the instructions that when executed cause the processing element of the data analytics computer system to train the classifier model comprise instructions that when executed cause the processing element of the data analytics computer system to:
store information about whether the user considered the email as graymail in the database; and
perform machine learning in the classifier model using the stored extracted features and stored information about whether the user considered the email as graymail.
24. The system of claim 17 , wherein the instructions that when executed cause the processing element of the tracking engine computer system to modify the email by inserting tracking information before delivering the email to the user comprise instructions that when executed cause the processing element of the tracking engine computer system to:
identify the email with a unique identification before providing the email to the user, and
wherein the instructions that when executed cause the tracking engine computer system to track user actions on the email without explicit user feedback using the inserted tracking information comprise instructions that when executed cause the tracking engine computer system to associate the user actions with the unique identification of the email.