Machine learning based electronic messaging system

• US 6,424,997 B1
• Filed: 01/27/1999
• Issued: 07/23/2002
• Est. Priority Date: 01/27/1999
• Status: Expired due to Term

• Current Assignee: LinkedIn Corporation (Microsoft Corporation)
• Original Assignee: International Business Machines Corporation
• Inventors: Johnson, David H., Raaen, Marguerite, Damerau, Frederick J., Buskirk, Martin C. Jr.
• Primary Examiner(s): Sheikh, Ayaz
• Assistant Examiner(s): DINH, KHANH Q
• Application Number:
  US09/237,886
• Time in Patent Office: 1,273 Days
• Field of Search: 709/201, 709/220, 709/228, 709/206, 709/204, 709/226, 709/230, 717/1, 717/3, 364/401
• US Class Current: 709/206
• CPC Class Codes: G06F 40/237 Lexical tools
• CPC Class Codes: G06Q 10/107 Computer-aided management of electronic mailing [e-mailing]
• CPC Class Codes: H04L 51/214 using selective forwarding
• CPC Class Codes: H04L 51/226 Delivery according to priorities

Abstract

A machine learning based electronic mail system. A classifier and action selection module analyzes the incoming message and classifies the messages with associated confidence levels, which may include analyzing the electronic message by tokenization of the text, morphological analysis of the text, and other well known processes. The classifier and action selection module then determines the appropriate action or actions to effect on the message.

Claims

1. A method of effecting one or more actions on an electronic message, comprising the steps of:

receiving the electronic message including raw text data;

providing a dictionary including words or phrases;

analyzing the electronic message including the raw text data and the words or phrases;

assigning one or more class labels and confidence levels to analyzed electronic message based on the words or phrases; and

performing one or more actions on the electronic message based on the one or more class labels and confidence labels,

wherein the confidence levels for classes are obtained from training data associated with those classes assigned to the electronic message, and

wherein training on the training data includes the steps of;

breaking the raw data of the training data into tokens;

counting frequencies of the tokens in the training data;

analyzing the tokens by at least one of filtering and weighting the tokens of the training data; and

transforming the at least one of filtered and weighted tokens of the training data into vector format.

2. The method of claim 1, wherein the step of analyzing the electronic message includes analyzing header and body information of the electronic message.

3. The method of claim 1, further comprising the step of routing the electronic message to an action module for performing one or more actions on the electronic message based on the one or more class labels and confidence levels.

4. The method of claim 3, further comprising the step of assigning confidence levels to the one or more class labels prior to performing the one or more actions on the electronic message, wherein each of the actions is based on the one or more class labels and the confidence levels.

5. The method of claim 1, wherein the electronic message includes a plurality of electronic messages.

6. The method of claim 5, wherein an action includes prioritizing the electronic messages in a predetermined order such that the actions are performed in the predetermined order.

7. The method of claim 1, wherein an action includes automatically generating a response to the electronic message.

8. The method of claim 7, wherein the class labels and confidence levels directly or indirectly determine the action to be performed on the electronic message.

9. The method of claim 1, further comprising providing a dictionary used in assigning the class labels.

10. The method of claim 9, wherein the dictionary includes words or phrases used as features such that the class labels are based on the features.

11. An apparatus for effecting one or more actions on an electronic message, comprising:

means for receiving the electronic message including raw text data;

means for analyzing the electronic message including the raw text data;

means for assigning one or more class labels and confidence levels to the electronic message; and

means for performing an action on the electronic message based on the one or more class labels and confidence levels,

wherein the means for assigning the one or more confidence levels to the analyzed electronic message includes means for training on training data,

the training means includes;

means for breaking the raw text data of the training data into tokens;

means for counting frequencies of the tokens in the training data;

means for analyzing the tokens by at least one of filtering and weighting the tokens of the training data; and

means for providing the at least one of filtered and weighted tokens of the training data into vector format.