Lightweight content filtering system for mobile phones

• US 7,756,535 B1
• Filed: 07/07/2006
• Issued: 07/13/2010
• Est. Priority Date: 07/07/2006
• Status: Expired due to Fees

• Current Assignee: Trend Micro Inc.
• Original Assignee: Trend Micro Inc.
• Inventors: Cao, Jackie, Diao, Lili, Chan, Vincent
• Primary Examiner(s): Harper; Vincent P
• Assistant Examiner(s): Figueroa; Marisol
• Application Number:
  US11/483,073
• Time in Patent Office: 1,467 Days
• Field of Search: 455/466, 455/412.1, 709/206
• US Class Current: 455/466
• CPC Class Codes: H04L 51/212 using filtering or selective blocking
• CPC Class Codes: H04L 51/58 Message adaptation for wireless communication
• CPC Class Codes: H04W 4/12 Messaging; Mailboxes; Announcements

Abstract

In one embodiment, a content filtering system includes a feature list and a learning model. The feature list may be a subset of a dictionary that was used to train the content filtering system to identify classification (e.g., spam, phishing, porn, legitimate text messages, etc.) of text messages during a training stage. The learning model may include representative vectors, each of which represents a particular class of text messages. The learning model and the feature list may be generated in a server computer during the training stage and then subsequently provided to the mobile phone. An incoming text message in the mobile phone may be parsed for occurrences of feature words included in the feature list and then converted to an input vector. The input vector may be compared to the learning model to determine the classification of the incoming text message.

Claims

1. A method of determining a classification of an incoming text message in a mobile phone, the method comprising:

loading a feature list into a main memory of the mobile phone, the feature list being a subset of a dictionary used to parse training text messages that were used to train a content filtering system to classify text messages in a computer separate from the mobile phone;

parsing the incoming text message in the mobile phone for occurrence of feature words included in the feature list;

converting the incoming text message into an input vector in the mobile phone, the input vector comprising numeric representations of feature words identified as occurring in the incoming text message, the feature words identified as occurring in the incoming text message having associated statistical values that are used as a parameter in converting the feature words into the input vector in the mobile phone, the statistical values being pre-computed in a computer separate from the mobile phone;

unloading the feature list from the main memory of the mobile phone;

loading a learning model into the main memory of the mobile phone, the learning model being generated during training of the content filtering system;

determining a classification of the incoming text message in the mobile phone by determining a similarity of the input vector to a first representative vector of the learning model.

2. The method of claim 1 further comprising:

unloading the learning model from the main memory of the mobile phone after computing a similarity of the input vector to representative vectors of the learning model.

3. The method of claim 1 wherein the incoming text message comprises a Short Message Service (SMS) text message.

4. The method of claim 1 wherein the feature words included in the feature list comprise words from the dictionary and selected by rank.

5. The method of claim 1 wherein the classification is selected from a list of classifications that includes spam.

6. The method of claim 1 wherein the first representative vector represents a spam text message.

7. The method of claim 1 wherein the learning model comprises a plurality of representative vectors, each of the representative vectors representing a class of text messages.

8. A content filtering system comprising:

a server computer comprising a dictionary that comprises a plurality of dictionary words, a word segmentation module configured to detect occurrences of dictionary words in a plurality of training text messages, a feature selection module configured to select feature words from among dictionary words occurring in the plurality of training text messages and to generate a feature list that is a subset of the dictionary and comprising the feature words, a first conversion module configured to convert the plurality of training text messages into vectors having numeric representations of selected words as items, and a training module configured generate a learning model; and

a mobile phone configured to receive the feature list and the learning model, the mobile phone comprising a second conversion module configured to detect occurrences of feature words from the feature list in an incoming text message and to convert the incoming text message into an input vector having numeric representations of feature words occurring in the incoming text message as items, the feature words occurring in the incoming text message having associated statistical values that are used as a parameter in converting the incoming text message into the input vector, the statistical values being pre-computed prior to being received in the mobile phone, and a prediction module configured to determine a classification of the incoming text message based on a comparison of the input vector to the learning model.

9. The system of claim 8 wherein the feature list and the learning model do not coexist in a main memory of the mobile phone at any given time.

10. The system of claim 8 wherein the incoming text message comprises an SMS text message.

11. The system of claim 8 wherein the feature list and the learning model are wirelessly transmitted from the server computer to the mobile phone.

12. The system of claim 8 wherein the learning model comprises a plurality of representative vectors to which the input vector is compared to determine the classification of the incoming text message.

13. The system of claim 8 wherein the classification is selected from a group comprising spam, phishing, and legitimate text messages.

14. A method of determining a classification of a text message, the method comprising:

wirelessly receiving an incoming text message in a resource limited device;

parsing the incoming text message in the resource limited device for occurrence of words included in a listing of words, the listing of words being a subset of a dictionary used to generate a learning model in a computer separate from the resource limited device, the listing of words but not the dictionary being available in the resource limited device;

converting the incoming text message into an input vector in the resource limited device, the input vector comprising numeric representations of words occurring in the incoming text message, the words occurring in the incoming message having associated statistical values that are used as a parameter in converting the incoming text message into the input vector in the resource limited device; and

determining a classification of the incoming text message in the resource limited device by comparing the input vector to the learning model.

15. The method of claim 14 wherein determining the classification of the text message by comparing the input vector to the learning model comprises:

computing a similarity value of the input vector to each representative vector of the learning model to generate a plurality of similarity values; and

deeming the incoming text message of a same classification as a representative vector of the learning model to which the input vector is most similar.

16. The method of claim 14 wherein the text message comprises an SMS text message.