Short message age classification

• US 9,063,927 B2
• Filed: 04/06/2012
• Issued: 06/23/2015
• Est. Priority Date: 04/06/2011
• Status: Active Grant

• Current Assignee: Citizennet Inc. (Advance Publications Incorporated)
• Original Assignee: Citizennet Inc. (Advance Publications Incorporated)
• Inventors: Hall, Michael Aaron
• Primary Examiner(s): ADESANYA, OLUJIMI A
• Application Number:
  US13/441,386
• Publication Number:
  US 20120259619A1
• Time in Patent Office: 1,173 Days
• Field of Search: 704/9
• US Class Current: 1/1
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

Systems and methods for short message age classification in accordance with embodiments of the invention are disclosed. In one embodiment of the invention, classifying messages using a classifier includes determining keyword feature information for a message using the classifier, classifying the determined feature information using the classifier, and estimating user age using the classifier.

Claims

1. A method for classifying messages using a classifier, comprising:

determining keyword feature information for a message using the classifier;

classifying the determined feature information using the classifier, wherein the classifier retrieves a set of verified user data utilized to classify the feature information from a classifier memory and the verified user data is generated by;

obtaining a set of training data comprising aggregated short message data;

obtaining determined feature data for the set of training data, wherein the determined feature data comprises identified features of the set of training data describing the age of the users providing the aggregated short message data in the set of training data;

determining training age data using the classifier by comparing the determined feature data to a set of pre-classified example data, wherein the pre-classified example data comprises a set of example short message data and a corresponding set of known age classification data, wherein each piece of example short message data has a known age described in the set of known age classification data; and

verifying the training age data based on the pre-classified example data; and

determining statistically relevant features in the verified training age data;

wherein the statistically relevant features comprises features of interests within a piece of feature information being classified using the classifier;

estimating user age data for the message based on the classified feature information using the classifier; and

providing the estimated user age data in response to a request for user age data.

2. The method of claim 1, wherein determining keyword feature information for a message includes feature information selected from the group consisting of:

vowel/consonant ratio, capitalization, number of emoticons, punctuation, usage of texting slang, and keyword frequency.

3. The method of claim 2, wherein determining vowel/consonant ratio keyword feature information further comprises:

counting the number of vowels in the message using the classifier;

counting the number of consonants in the message using the classifier; and

dividing the number of vowels in the message by the number of consonants in the message using the classifier.

4. The classifier of claim 3, wherein the letter ‘

y’

is excluded from the number of vowels in the message and from the number of consonants in the message.

5. The method of claim 2, wherein determining capitalization keyword feature information for a message further comprises counting the number of capital letter in the message using the classifier.

6. The method of claim 2, wherein determining emoticon keyword feature information for a message further comprises counting the number of emoticons in the message using the classifier.

7. The method of claim 2, wherein determining punctuation keyword feature information for a message further comprises:

removing emoticons from the message using the classifier; and

counting the number of punctuation marks in the message using the classifier.

8. The method of claim 2, wherein determining usage of texting slang keyword feature information for a message further comprises:

breaking the message into slang tokens using the classifier; and

counting the number of slang tokens using the classifier.

9. The method of claim 8, wherein slang tokens are determined using a dictionary of texting slang and the classifier.

10. The method of claim 2, wherein determining keyword frequency further comprises:

tokenizing a message using the classifier, where the message includes a set of words;

filtering stop words from the set of words using the tokenized message and the classifier;

keyword stemming the filtered tokenized message using the classifier; and

counting the stemmed keywords using the classifier.

11. The method of claim 10, wherein tokenizing a message further comprises:

breaking up the message into words using the classifier, where the words are contained in the set of words; and

creating a token for each word in the broken up message using the classifier.

12. The method of claim 10, wherein the stop words are words in a message that do not provide useful information to the classifier.

13. A classifier device, comprising:

memory storing aggregate user data and a classifier application, where the aggregate user data includes a set of messages; and

a processor, where the classifier application directs the processor to;

determines keyword feature information using aggregate user data;

classify the determined feature information, wherein the classifier comprises a set of verified user data utilized to classify feature information and the verified user data is generated by;

obtaining a set of training data comprising aggregated short message data;

obtaining determined feature data for the set of training data, wherein the determined feature data comprises identified features of the set of training data describing the age of the users providing the aggregated short message data in the set of training data;

determining training age data using the classifier by comparing the determined feature data to a set of pre-classified example data, wherein the pre-classified example data comprises a set of example short message data and a corresponding set of known age classification data, wherein each piece of example short message data has a known age described in the set of known age classification data; and

verifying the training age data based on the pre-classified example data; and

determining statistically relevant features in the verified training age data;

wherein the statistically relevant features comprises features of interests within a piece of feature information being classified using the classifier;

estimate user age data the aggregate user data based on the classified feature information using the classifier application; and

provides the estimated user age data in response to a request for the user age data.

14. The classifier of claim 13, wherein the classifier application utilizes a Naï

ve Bayesian classifier.

15. The classifier of claim 13, wherein the classifier application utilizes a maximum entropy classifier.

16. The classifier of claim 13, wherein the processor aggregates user data over a network.

17. The classifier of claim 13, wherein the processor determines the keyword feature information using the aggregate user data and predetermined feature information.

18. The classifier of claim 13, wherein processor determines the keyword feature information for a message contained in the set of messages using the classifier application.

19. The classifier of claim 13, wherein the determined keyword feature information includes feature information selected from the group consisting of:

vowel/consonant ratio, capitalization, number of emoticons, punctuation, usage of texting slang, and keyword frequency.