Statistical language-model based system for detection of missing attachments

• US 7,574,349 B2
• Filed: 03/29/2006
• Issued: 08/11/2009
• Est. Priority Date: 03/29/2006
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Perronnin, Florent C.
• Primary Examiner(s): Hudspeth; David R
• Assistant Examiner(s): Spooner; Lamont M
• Application Number:
  US11/392,194
• Publication Number:
  US 20070233459A1
• Time in Patent Office: 1,231 Days
• Field of Search: 704/9, 704/1, 704/10, 707 1- 6, 726/22, 709/206, 715/205, 715/255, 715/256
• US Class Current: 704/9
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G06Q 10/107 Computer-aided management of electronic mailing [e-mailing]

Abstract

A method for processing electronic mail includes computing a probability that a text string in an electronic mail message refers to an attachment as a function of a stored probability value for each of a plurality of sequences of words within the text string. Where the email message lacks an attachment, the method includes prompting a user if the computed probability indicates that the text string refers to an attachment.

Claims

1. A method for processing electronic mail comprising:

computing a probability that a text string in an electronic mail message refers to an attachment as a function of a stored probability value for each of a plurality of sequences of words within the text string, the computing of the probability including computing a first probability that the text string refers to an attachment using a first set of stored probability values and, where the first probability exceeds a predetermined value, computing a second probability that the text string does not refer to an attachment using a second set of stored probability values, the probability that a text string in an electronic mail message refers to an attachment being computed as a function of the first and second probabilities; and

where the electronic mail message lacks an attachment, prompting a user if the computed probability indicates that the text string refers to an attachment.

2. The method of claim 1, further comprising:

prior to computing the probability, identifying whether the electronic message lacks an attachment.

3. The method of claim 1 wherein at least some of the stored probabilities are derived from a corpus of email message text which refers to an attachment.

4. The method of claim 1, wherein the first set of probability values are derived from a first corpus of email message text strings determined to refer to an attachment and the second set of probability values are derived from a second corpus of email message text strings determined not to refer to an attachment.

5. The method of claim 1, wherein the computing of the probability includes summing log probability values for each of the plurality of sequences of words.

6. The method of claim 1, wherein the method further comprises:

for each of a plurality of text strings in the electronic mail message, computing a probability that the text string refers to an attachment as a function of a stored probability value for each of a plurality of sequences of words within the text string.

7. The method of claim 6, wherein when none of the plurality of text strings has a computed probability which exceeds a threshold value, sending the electronic mail message without prompting the user.

8. The method of claim 1, wherein the computing of the probability that the text string refers to an attachment includes identifying a preselected number of sequences of words in the text string, retrieving stored probability values for each of the identified sequences, and computing the probability as a function of the retrieved probability values.

9. The method of claim 1, wherein the text string comprises a predetermined fixed number of words.

10. The method of claim 1, wherein each sequence of words comprises the same number of words.

11. The method of claim 1, further comprising, prior to computing the probability, identifying a language of the electronic mail message.

12. The method of claim 1, further comprising, prior to computing the probability, identifying a portion of the electronic mail message prepared by the user and computing the probability for a text string in the identified portion.

13. The method of claim 1, wherein the prompting includes displaying a query to the user on a screen.

14. The method of claim 1, further comprising adapting at least one stored probability value to the writing patterns of the user.

15. A system for processing electronic mail comprising:

a statistical language model which stores probability values for a plurality of text sequences, the statistical language model including a first statistical language model component which models the probability that a text sequence of an electronic mail message refers to an attachment and a second statistical language model component which models the probability that a text sequence of an electronic mail message does not refer to an attachment; and

a processor which executes instructions for retrieving stored probability values from the statistical language model for a plurality of text sequences in the electronic mail message and computing a probability that that the electronic mail message refers to an attachment as a function of the retrieved probability values, the instructions including instructions for computing a first probability for each of the text sequences using the first statistical language model component and instructions for computing a second probability for each text sequence using the second statistical language model component, the instructions for computing the probability that the electronic mail message refers to an attachment including instructions for computing a function of the first and second probabilities.

16. The system of claim 15, wherein the processor executes instructions for prompting a user when the computed probability that the electronic mail message refers to an attachment exceeds a threshold value.

17. A method for processing electronic mail comprising:

analyzing words of text strings of an electronic mail message with a statistical language model to determine whether one of the string has at least a threshold probability of referring to an attachment, each text string comprising an N-gram having a fixed number N of words, the probability of the text string being computed as a function of the individual probability of each word in the text string, the individual probability of said each word being the probability of said word in combination with the words history, the history comprising previous words to said word in the electronic mail message, where present; and

where the electronic mail message lacks an attachment and the text string is determined to have at least the threshold probability that the text string refers to an attachment, prompting a user.

18. A method for processing electronic mail comprising:

stepwise moving a sliding window of a string K of words along a text string of an electronic mail message and at each sliding step;

computing a probability for the window including using a first statistical language model component which models the probability that a text sequence of an electronic mail message refers to an attachment,

optionally, if the probability exceeds a threshold, computing a probability for the window using a second statistical language model component which models the probability that a text sequence of an electronic mail message does not refer to an attachment; and

where the electronic mail message lacks an attachment, prompting a user if a computed probability for at least one window indicates that the text string refers to an attachment, the computed probability being based on the probability computed using the first statistical language model component and optionally on the probability computed using the second statistical language model component.