Online adaptive filtering of messages

Abstract

In general, a two or more stage spam filtering system is used to filter spam in an e-mail system. One stage includes a global e-mail classifier that classifies e-mail as it enters the e-mail system. The parameters of the global e-mail classifier generally may be determined by the policies of e-mail system owner and generally are set to only classify as spam those e-mails that are likely to be considered spam by a significant number of users of the e-mail system. Another stage includes personal e-mail classifiers at the individual mailboxes of the e-mail system users. The parameters of the personal e-mail classifiers generally are set by the users through retraining, such that the personal e-mail classifiers are refined to track the subjective perceptions of their respective user as to what e-mails are spam e-mails. Retraining data fro the personal e-mail classifiers may be aggregated and a subset of the aggregate may be chosen for use in retraining the global e-mail classifier.

1 .- 58 . (canceled) 
     
     
         59 . A system for classifying electronic messages, comprising:
 a client device of a user comprising one or more hardware processors and a non-transitory computer-readable medium containing instructions that, when executed by the one or more hardware processors, cause the client device to perform operations comprising:
 receiving a plurality of electronic messages addressed to the user; 
 classifying the plurality of electronic messages according to a personal classifier of the user; 
 generating retraining data based on input received from the user regarding at least one of the electronic messages; and 
 providing the retraining data to a global classifier. 
   
     
     
         60 . The system of  claim 59 , wherein the global classifier is a probabilistic message classifier that updates, using the retraining data, an internal global model for determining whether a message is spam. 
     
     
         61 . The system of  claim 59 , wherein the personal classifier is a probabilistic message classifier that updates, using the retraining data, an internal personal model for determining whether a message is spam. 
     
     
         62 . The system of  claim 59 , wherein the input received from the user indicates that the user considers the message to be spam. 
     
     
         63 . The system of  claim 62 , wherein the input received from the user comprises one or more of the following instructions: keeping the at least one of the electronic messages as new after the at least one of the electronic messages has been read; forwarding the at least one of the electronic messages; replying to the at least one of the electronic messages; printing the at least one of the electronic messages; and adding a sender of the at least one of the electronic messages to at least one address book. 
     
     
         64 . The system of  claim 62 , wherein the input received from the user comprises at least one instruction for reporting the message as spam or moving the message to a spam folder. 
     
     
         65 . The system of  claim 59 , where the global classifier is less selective than the personal classifier. 
     
     
         66 . The system of  claim 59 , wherein the electronic messages comprise one or more of emails, instant messages, and SMS messages. 
     
     
         67 . A computer-implemented method for classifying electronic messages, the comprising the following operations performed by at least one processor:
 receiving, by a client device of a user, a plurality of electronic messages addressed to the user;   classifying, using an application executed by the client device, the plurality of electronic messages according to a personal classifier of the user;   generating, using the application executed by the client device, retraining data based on input received from the user regarding at least one of the electronic messages; and   providing, by the client device, the retraining data to a global classifier.   
     
     
         68 . The computer-implemented method of  claim 67 , wherein the global classifier is a probabilistic message classifier that updates, using the retraining data, an internal global model for determining whether a message is spam. 
     
     
         69 . The computer-implemented method of  claim 67 , wherein the personal classifier is a probabilistic message classifier that updates, using the retraining data, an internal personal model for determining whether a message is spam. 
     
     
         70 . The computer-implemented method of  claim 67 , wherein the input received from the user indicates that the user considers the message to be spam. 
     
     
         71 . The computer-implemented method of  claim 70 , wherein the input received from the user comprises one or more of the following instructions: keeping the at least one of the electronic messages as new after the at least one of the electronic messages has been read; forwarding the at least one of the electronic messages; replying to the at least one of the electronic messages; printing the at least one of the electronic messages; and adding a sender of the at least one of the electronic messages to at least one address book. 
     
     
         72 . The computer-implemented method of  claim 70 , wherein the input received from the user comprises at least one instruction for reporting the message as spam or moving the message to a spam folder. 
     
     
         73 . The computer-implemented method of  claim 67 , where the global classifier is less selective than the personal classifier. 
     
     
         74 . The computer-implemented method of  claim 67 , wherein the electronic messages comprise one or more of emails, instant messages, and SMS messages. 
     
     
         75 . A computer-readable non-transitory medium containing instructions that, when executed by one or more processors of a client device of a user, cause the client device to perform operations comprising:
 receiving, by the client device of the user, a plurality of electronic messages addressed to the user;   classifying the plurality of electronic messages according to a personal classifier of the user;   generating retraining data based on input received from the user regarding at least one of the electronic messages; and   providing, by the client device, the retraining data to a global classifier.   
     
     
         76 . The computer-readable non-transitory medium of  claim 75 , wherein the global classifier is a probabilistic message classifier that updates, using the retraining data, an internal global model for determining whether a message is spam, and wherein the personal classifier is a probabilistic message classifier that updates, using the retraining data, an internal personal model for determining whether a message is spam. 
     
     
         77 . The computer-readable non-transitory medium of  claim 75 , wherein the input received from the user indicates that the user considers the message to be spam, and wherein the global classifier is less selective than the personal classifier. 
     
     
         78 . The computer-readable non-transitory medium of  claim 75 , wherein the electronic messages comprise one or more of emails, instant messages, and SMS messages.