Dynamic message filtering

• US 20050076084A1
• Filed: 10/03/2003
• Published: 04/07/2005
• Est. Priority Date: 10/03/2003
• Status: Active Grant

• Current Assignee: Mailgate, LLC
• Original Assignee: Corvigo, L.L.C.
• Inventors: Olson, Mike, Maedge, Ehren, Loughmiller, Scott, Collier, Jason, White, Phil, Ready, Jeff
• Granted Patent: US 7,257,564 B2
• Time in Patent Office: Days
• US Class Current: 709/206
• CPC Class Codes: G06Q 10/107 Computer-aided management of electronic mailing [e-mailing]
• CPC Class Codes: H04L 51/212 using filtering or selective blocking

Abstract

Dynamically filtering and classifying messages, as good messages, bulk periodicals, or spam. A regular expression recognizer, and pre-trained neural networks. The neural networks distinguish “likely good” from “likely spam,” and also operate at a more discriminating level to distinguish among the three categories above. A dynamic whitelist and blacklist; sending addresses are collected when the number of their messages indicates the sender is good or a spammer. A dynamically selected set of regular expressions input to the neural networks.

Claims

1. A method for classifying messages, comprising the steps of:

recognizing patterns including words and groups of words in a messages;

applying a plurality of machine learning techniques responsive to the recognized patterns in order to classify the message.

2. A method as in claim 1, wherein the machine learning techniques include neural networks.

3. A method as in claim 2, wherein the neural networks are pretrained to classify the message as a good message, a bulk message, or a spam message.

4. A method as in claim 2, wherein the neural networks further comprise at least two levels of neural networks.

5. A method as in claim 4, wherein the two levels of neural networks include a first level that determines if the message is likely good or likely spam, and a second level that determines if a likely good message is good and if a likely spam message is spam.

6. A method as in claim 4, wherein the two levels of neural networks include a first level that determines if the message is likely good or likely spam, and a second level that determines if a likely good message is good or bulk and if a likely spam message is spam or bulk.

7. A method as in claim 4, wherein for at least one of the classifications the neural networks classify the message in one of three classifications, wherein more than one path through the neural networks exists for the message to arrive at that classification.

8. A method as in claim 1, further comprising the step of dynamically maintaining the neural networks responsive to classification of the message.

9. A method as in claim 1, further comprising the step of applying rules to the message to help classify the message.

10. A method as in claim 9, wherein if the message is classified by the rules, the step of applying the neural networks is skipped.

11. A method as in claim 9, wherein the rules utilize a whitelist, a blacklist, or both the whitelist and the blacklist.

12. A method as in claim 11, further comprising the step of dynamically maintaining the whitelist, the blacklist, or both the whitelist and the blacklist responsive to classification of the message.

13. A method as in claim 11, wherein the step of recognizing expressions further includes the step of applying a genetic algorithm to select a set of regular expressions to be recognized.