Bayesian surety check to reduce false positives in filtering of content in non-trained languages
View Patent ↗A Bayesian spam filter determines an amount of content in incoming email messages that it knows from training. If the filter is familiar with a threshold amount of the content, then the filter proceeds to classify the email message as being spam or legitimate. On the other hand, if not enough of the words in the email are known to the filter from training, then the filter cannot accurately determine whether or not the message is spam. In this case, the filter classifies the message as being of type unknown. Different threshold metrics can be used, such as the percentage of known words, and the percentage of maximum correction value used during processing. This greatly improves the processing of emails in languages on which the filter was not trained.
1. A computer implemented method for reducing false classifications during Bayesian filtering, the method comprising the steps of:
within a context of Bayesian filtering, analyzing content of at least one document to be classified;
determining whether more than a threshold amount of the content of the at least one document is known wherein determining whether more than a threshold amount of the content of the at least one document is known further comprises performing at least one step from a group of steps consisting of:
determining a percentage of words of the at least one document that are known from Bayesian filter training; and
determining a percentage of a maximum correction value utilized during Bayesian filtering of the at least one document; and
classifying the document responsive to results of the determination.
2. The method of claim 1 wherein classifying the document responsive to results of the determination further comprises performing a step from a group of steps consisting of:
responsive to determining that more than a threshold amount of the content of the document is not known, classifying the document as being of an unknown type; and
responsive to determining that less than a threshold amount of the content of the document is not known, classifying the document according to results of standard Bayesian filtering.
3. The method of claim 1 wherein determining a percentage of a maximum correction value utilized during Bayesian filtering of the at least one document further comprises:
calculating the maximum correction value for the at least one document as a function of a zero count value and a number of words in the at least one document; and
calculating a percentage of the maximum correction value utilized during Bayesian filtering of the at least one document as a function of a correction value utilized during Bayesian filtering of the at least one document and the maximum correction value for the at least one document.
4. The method of claim 3 wherein the zero count value further comprises a special value to use during Bayesian filtering for words not encountered during Bayesian filter training, the method further comprising:
calculating the zero count value as a natural logarithm of one divided by the number of words in the training data set multiplied by a constant.
5. The method of claim 3 wherein the correction value utilized during Bayesian filtering of the at least one document further comprises:
a special value calculated to prevent any individual word from unduly influencing classification of the at least one document.
6. The method of claim 1 wherein the at least one document comprises at least one email message and the Bayesian filtering classifies the at least one email message as belonging to a single category from a group of categories consisting of:
spam;
legitimate email; and
unknown.
7. At least one non-transitory computer readable medium containing a computer program product for reducing false classifications during Bayesian filtering, the computer program product comprising:
program code for, within a context of Bayesian filtering, analyzing content of at least one document to be classified;
program code for determining whether more than a threshold amount of the content of the at least one document is known, wherein the program code for determining whether more than a threshold amount of the content of the at least one document is known further comprises program code for performing at least one step from a group of steps consisting of:
determining a percentage of words of the at least one document that are known from Bayesian filter training; and
determining a percentage of a maximum correction value utilized during Bayesian filtering of the at least one document; and
program code for classifying the document responsive to results of the determination.
8. The computer program product of claim 7 wherein the program code for classifying the document responsive to results of the determination further comprises program code for performing a step from a group of steps consisting of:
responsive to determining that more than a threshold amount of the content of the document is not known, classifying the document as being of an unknown type; and
responsive to determining that less than a threshold amount of the content of the document is not known, classifying the document according to results of standard Bayesian filtering.
9. The computer program product of claim 7 wherein the program code for determining a percentage of a maximum correction value utilized during Bayesian filtering of the at least one document further comprises:
program code for calculating the maximum correction value for the at least one document as a function of a zero count value and a number of words in the at least one document; and
program code for calculating a percentage of the maximum correction value utilized during Bayesian filtering of the at least one document as a function of a correction value utilized during Bayesian filtering of the at least one document and the maximum correction value for the at least one document.
10. The computer program product of claim 9 wherein the zero count value further comprises a special value to use during Bayesian filtering for words not encountered during Bayesian filter training, the computer program product further comprising:
program code for calculating the zero count value as a natural logarithm of one divided by the number of words in the training data set multiplied by a constant.
11. The computer program product of claim 9 wherein the correction value utilized during Bayesian filtering of the at least one document further comprises:
a special value calculated to prevent any individual word from unduly influencing classification of the at least one document.
12. The computer program product of claim 7 further comprising program code for classifying at least one email message as belonging to a single category from a group of categories consisting of:
spam;
legitimate email; and
unknown.
13. A computer system for reducing false classifications during Bayesian filtering, the computer system comprising:
a portion for, within a context of Bayesian filtering, analyzing content of at least one document to be classified;
a portion for determining whether more than a threshold amount of the content of the at least one document is known, wherein the portion for determining whether more than a threshold amount of the content of the at least one document is known further comprises a portion for performing at least one step from a group of steps consisting of:
determining a percentage of words of the at least one document that are known from Bayesian filter training; and
determining a percentage of a maximum correction value utilized during Bayesian filtering of the at least one document; and
a portion for classifying the document responsive to results of the determination.
14. The computer system of claim 13 wherein the portion for classifying the document responsive to results of the determination further comprises a portion for performing a step from a group of steps consisting of:
responsive to determining that more than a threshold amount of the content of the document is not known, classifying the document as being of an unknown type; and
responsive to determining that less than a threshold amount of the content of the document is not known, classifying the document according to results of standard Bayesian filtering.
15. The computer system of claim 13 wherein the portion for determining a percentage of a maximum correction value utilized during Bayesian filtering of the at least one document further comprises:
a portion for calculating the maximum correction value for the at least one document as a function of a zero count value and a number of words in the at least one document; and
a portion for calculating a percentage of the maximum correction value utilized during Bayesian filtering of the at least one document as a function of a correction value utilized during Bayesian filtering of the at least one document and the maximum correction value for the at least one document.
16. The computer system of claim 15 wherein the zero count value further comprises a special value to use during Bayesian filtering for words not encountered during Bayesian filter training, the computer system further comprising:
a portion for calculating the zero count value as a natural logarithm of one divided by the number of words in the training data set multiplied by a constant.
17. The computer system of claim 13 further comprising a portion for classifying at least one email message as belonging to a single category from a group of categories consisting of:
spam;
legitimate email; and
unknown.