Method and apparatus for predicting unwanted electronic messages for a user

An electronic messaging server comprising:

• a processor;
• a storage medium for tangibly storing thereon program logic for execution by the processor, the stored program logic comprising:
• generating logic executed by the processor for generating, for a user, training data, generating the training data comprising using actions of the user with respect to a plurality of the user's electronic mail messages;
• generating logic executed by the processor for generating, for the user, a personalized predictor, corresponding to an action, for use as a gray spam predictor for the user, the personalized predictor being generated using the user's training data;
• determining logic executed by the processor for determining an activeness level of the user corresponding to the action, the activeness level of the user corresponding to the action being based on a number of observed examples of the action and an activeness threshold;
• selecting logic executed by the processor for selecting, using the determined activeness level of the user, a horizontal model from a plurality of horizontal models trained using training data from a number of users, each horizontal model of the plurality of horizontal models corresponding to the action and a different activeness level;
• regularizing logic executed by the processor for regularizing, for the user, the personalized predictor, for use as the gray spam predictor, the regularizing using the selected horizontal model;
• prediction making logic executed by the processor for automatically making a prediction, for the user, whether or not a new electronic mail message of the user is unwanted by the user using the gray spam predictor personalized for the user and regularized using the selected horizontal model associated with the activeness level determined for the user;
• and
• preforming logic executed by the processor for automatically performing, for the user, at least one operation on the new electronic mail message if the prediction indicates that the new electronic mail message is unwanted by the user.

A computer readable non-transitory storage medium for tangibly storing thereon computer readable instructions that when executed by an electronic messaging server, perform functions comprising:

• generating, for a user, training data, generating the training data comprising using actions of the user with respect to a plurality of the user's electronic mail messages;
• generating, for the user, a personalized predictor, corresponding to the action, for use as a gray spam predictor for the user, the personalized predictor being generated using the user's training data;
• determining an activeness level of the user corresponding to an action, the activeness level of the user being based on a number of observed examples of the action and an activeness threshold;
• selecting, using the determined activeness level of the user, a horizontal model from a plurality of horizontal models trained using training data from a number of users, each horizontal model of the plurality of horizontal models corresponding to the action and a different activeness level;
• regularizing, for the user, the personalized predictor, for use as the gray spam predictor, the regularizing using the selected horizontal model;
• automatically making a prediction, for the user, whether or not a new electronic mail message of the user is unwanted by the user using the gray spam predictor personalized for the user and regularized using the selected horizontal model associated with the activeness level determined for the user;
• and
• automatically performing, for the user, at least one operation on the new electronic mail message if the prediction indicates that the new electronic mail message is unwanted by the user.