SMART ATTRIBUTE CLASSIFICATION (SAC) FOR ONLINE REVIEWS

• US 20120166370A1
• Filed: 03/06/2012
• Published: 06/28/2012
• Est. Priority Date: 01/19/2009
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Hu, Jian, Sun, Jian-Tao, Chen, Zheng
• Granted Patent: US 8,682,896 B2
• Time in Patent Office: Days
• US Class Current: 706/12
• CPC Class Codes: G06F 16/353 into predefined classes
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 20/10 using kernel methods, e.g. support vector machines [SVM]

Abstract

Techniques for identifying attributes in a sentence and determining a number of attributes to be associated with the sentence is described.

Claims

1. A method of identifying attributes in a sentence, the method comprising:

representing the sentence as a first feature vector;

training a binary classifier based upon the first feature vector;

obtaining a vector of probability for the sentence;

concatenating the first feature vector with the vector of probability to obtain a second feature vector; and

training the binary classifier based upon the second feature vector.

2. The method as recited in claim 1, wherein the sentence comprises a plurality of attributes.

3. The method as recited in claim 1, wherein the sentence comprises an online review for a product.

4. The method as recited in claim 1, wherein representing the sentence further comprises the norm of the first feature vector being equal to 1.

5. The method as recited in claim 1, wherein training the binary classifier based upon the first feature vector further comprises the binary classifier being trained with a support vector machine based classification model using a one-against-all mechanism.

6. The method as recited in claim 1, wherein obtaining the vector of probability further comprises the vector of probability storing the probability of the sentence belonging to an attribute category.

7. The method as recited in claim 1, wherein the norm of the second feature vector is equal to 1.

8. The method as recited in claim 1, wherein training the binary classifier based upon the second feature vector further comprises the binary classifier being trained with a support vector machine based classification model using a one-against-all mechanism.

9. A method of determining a number of attributes to be associated with each sentence of a plurality of sentences, the method comprising:

training i binary classifiers such that for each classifier;

a positive value is assigned to each sentence belonging to a category of the classifier; and

a negative value is assigned to each sentence not belonging to the category of the classifier; and

determining the number of attributes j of each sentence x by the equation;

j=arg max p(y=i|x)

with the first j attributes of the i categories being the predicted attributes for each sentence.

10. The method as recited in claim 9, further including determining a posterior probability for each sentence belonging to category i, with the posterior probability being defined as:

p(y=1|ƒ

_i)=1/1+exp(Aƒ

_i(x)+B)

wherein ƒ

_i(x) is the output of the decision function ƒ

_i and A and B are scalar values fitted by maximum likelihood estimation.

11. The method as recited in claim 10, wherein the sentence comprises a plurality of attributes.

12. The method as recited in claim 10, wherein the sentence is an online review for a product.

13. A computer storage media having recorded thereon computer-executable instructions that upon execution configure a processor to perform operations comprising:

identifying attributes in a sentence by;

representing the sentence as a first feature vector;

training a binary classifier based upon the first feature vector;

obtaining a vector of probability for the sentence;

concatenating the first feature vector with the vector of probability to obtain a second feature vector; and

training the binary classifier based upon the second feature vector.

14. A computer storage media as recited in claim 13, wherein the sentence comprises a plurality of attributes.

15. A computer storage media as recited in claim 13, wherein the sentence comprises an online review for a product.

16. A computer storage media as recited in claim 13, wherein representing the sentence further comprises the norm of the first feature vector being equal to 1.

17. A computer storage media as recited in claim 13, wherein training the binary classifier based upon the first feature vector further comprises the binary classifier being trained with a support vector machine based classification model using a one-against-all mechanism.

18. A computer storage media as recited in claim 13, wherein obtaining the vector of probability further comprises the vector of probability storing the probability of the sentence belonging to an attribute category.

19. A computer storage media as recited in claim 13, wherein training the binary classifier based upon the second feature vector further comprises the binary classifier being trained with a support vector machine based classification model using a one-against-all mechanism.

20. A computer storage media as recited in claim 13, the operations further comprising:

determining a number of attributes to be associated with each sentence of a plurality of sentences at least by;

training a number of binary classifiers such that for each binary classifier a value is assigned to each sentence based at least on the respective sentence belonging to a category of the binary classifier; and

calculating a predicted number of the attributes based at least on the number of categories associated with each sentence.