Identifying categorized misplacement

• US 8,812,420 B2
• Filed: 01/17/2012
• Issued: 08/19/2014
• Est. Priority Date: 01/25/2011
• Status: Active Grant

• Current Assignee: Alibaba Group Holding Ltd.
• Original Assignee: Alibaba Group Holding Ltd.
• Inventors: Wang, Yulin
• Primary Examiner(s): Holmes, Michael B
• Application Number:
  US13/513,547
• Publication Number:
  US 20130304687A1
• Time in Patent Office: 945 Days
• Field of Search: 706/46
• US Class Current: 706/46
• CPC Class Codes: G06F 16/353 into predefined classes
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation
• CPC Class Codes: G06N 5/04 Inference or reasoning models
• CPC Class Codes: G06Q 30/00 Commerce

Abstract

The present disclosure provides methods and devices for identifying category misplacement. In one embodiment, an example device obtains a word frequency of each respective word in a product title under a current category, calculates an overall word frequency of the product title under the current category based on the word frequency of each respective word under the current category, and compares the overall word frequency of the product title with a threshold of the current category to determine an existence of category misplacement. The techniques can accurately identify category misplacement and reduce the probability of missing identifying category misplacement. The techniques also require less system resources and improve calculation efficiency.

Claims

1. A method of identifying category misplacement, performed by one or more processors configured with computer-executable instructions, the method comprising:

obtaining a word frequency of each respective word in a product title under a current category;

calculating an overall word frequency of the product title under the current category based on the word frequency of each respective word under the current category; and

comparing the overall word frequency of the product title with a threshold of the current category to determine an existence of category misplacement.

2. The method as recited in claim 1, further comprising determining one or more stop words from the product title.

3. The method as recited in claim 2, wherein the determining is based on distribution entropy of each respective word under multiple categories.

4. The method as recited in claim 2, further comprising filtering the one or more stop words from one or more words in the product title prior to obtaining the word frequency of each respective word in the product title under the current category.

5. The method as recited in claim 1, wherein the threshold is different under different categories.

6. A method of identifying category misplacement, performed by one or more processors configured with computer-executable instructions, the method comprising:

obtaining a TOP value of each respective word in a product title under a current category by summing word frequencies of one or more words whose word frequencies are higher than that of respective word under the current category;

calculating a relevancy degree value of each respective word corresponding to the TOP value based on a predetermined mapping between the TOP value and the relevancy degree value;

adding relevancy degree values of each respective word to obtain an overall relevancy degree value of the product title under the current category; and

comparing the overall relevancy degree value of the product title under the current category with a threshold to determine an existence of a category misplacement.

7. The method as recited in claim 6, wherein the comparing the overall relevancy degree value of the product title under the current category with the threshold to determine the existence of the category misplacement comprises:

determining that the category misplacement exists in an event that the overall relevancy degree value of the product title under the current category is lower than the threshold; and

determining that the category misplacement does not exist in an event that the overall relevancy degree value of the product title under the current category is equal to or greater than the threshold.

8. The method as recited in claim 6, further comprising:

determining one or more stop words from the product title based on a distribution entropy of each respective word under multiple categories; and

filtering the one or more stop words from words in the product title prior to obtaining the TOP value of each word in the product title under the current category.

9. The method as recited in claim 6, wherein the threshold is different under different categories.

10. The method as recited in claim 6, wherein the comparing the overall relevancy degree value of the product title under the current category with the threshold to determine the existence of the category misplacement comprises:

comparing the overall relevancy degree value of the product title under the current category with the threshold;

determining that an abnormality in the category misplacement exists in an event that the overall relevancy degree value of the product title of the current category is lower than the threshold;

obtaining one or more recommended categories of the product title;

based on the product title and the one or more recommended categories, determining a target category that matches the product title;

obtaining a posterior probability of the product title under the current category and a posterior probability of the product title under the target category;

comparing the posterior probability of the product title under the current category with the posterior probability of the product title under the target category; and

determining that the category misplacement exists in an event that a difference or a ratio between the posterior probability of the product title under the current category with the posterior probability of the product title under the target category is greater than another threshold.

11. The method as recited in claim 10, wherein the obtaining the one or more recommended categories comprising:

obtaining a degree of matching of each respective word of the product title with multiple categories; and

comparing the degree of matching to find the one or more recommended categories.

12. The method as recited in claim 10, wherein the determining the target category comprises:

using a word frequency of each respective word under each of the one or more recommended categories as a weight of a respective recommended category;

calculating a sum of weights of all of the each respective word for each of the one or more recommended categories to obtain a sum of weights for each of the one or more recommended categories; and

choosing a recommended category with a highest sum of weights as the target category.

13. The method as recited in claim 10, wherein the determining the target category comprises:

obtaining a product title vector and a category vector for a respective recommended category, one or more elements of the product title vector including a probability of appearance of each respective word in the product title, one or more elements of the category vector including a word frequency of each respective word under the respective category;

obtaining a product of the product title vector and the category vector for the respective recommended category; and

choosing a recommended category with a highest product as the target category.

14. The method as recited in claim 10, wherein the other threshold is different under different categories.

15. The method as recited in claim 6, wherein the comparing the overall relevancy degree value of the product title under the current category with the threshold to determine the existence of the category misplacement comprises:

comparing the overall relevancy degree value of the product title under the current category with the threshold;

determining that an abnormality in the category misplacement exists in an event that the overall relevancy degree value of the product title of the current category is lower than the threshold;

obtaining one or more recommended categories of the product title;

based on the product title and the one or more recommended categories, determining a target category that matches the product title;

comparing the overall relevancy degree value of the product title under the current category with an overall relevancy degree value of the product title under the target category; and

determining that the category misplacement exists in an event that a difference or a ratio between the overall relevancy degree value of the product title under the current category and the overall relevancy degree value of the product title under the target category is greater than another threshold.

16. A device for identifying category misplacement comprising:

one or more processors;

memory storing multiple modules, which are executable by the one or more processor, comprising;

a word frequency module that obtains a word frequency of each respective word in a product title under a current category;

an overall word frequency calculation module that calculates an overall word frequency of the product title under the current category based on the word frequency of each respective word under the current category; and

a category misplacement determination module that compares the overall word frequency of the product title with a threshold of the current category to determine an existence of category misplacement.

17. A device for identifying category misplacement comprising:

one or more processors;

memory storing multiple modules, which are executable by the one or more processor, comprising;

a TOP value obtaining module that obtains a TOP value of each respective word in a product title under a current category, the TOP value being a sum of word frequencies of one or more words whose word frequencies are higher than that of respective word;

an overall relevancy degree obtaining module that calculates a relevancy degree value of each respective word corresponding to the TOP value based on a predetermined mapping between the TOP value and the relevancy degree value, adds up relevancy degree values of each respective word, and obtains an overall relevancy degree value of the product title under the current category; and

a category misplacement determination module that compares the overall relevancy degree value of the product title under the current category with a threshold to determine an existence of a category misplacement.

18. The device as recited in claim 17, wherein the category misplacement determination module comprises:

an abnormality in product placement determination module that compares an overall relevancy degree value of the product title under the current category with a threshold, and determines that there is abnormality in product placement in an event that the overall relevancy degree value of the product title under the current category is less than the threshold;

a target category obtaining module that obtains one or more recommended categories for the product title after a determination that there is abnormality in product placement, and obtains a target category that matches the product title based on the product title and the one or more recommended categories;

a posterior probability obtaining module that obtains a posterior probability of the product title under the target category and a posterior probability of the product title under the current category; and

a determination of category misplacement module that compares the posterior probability of the product title under the target category and the posterior probability of the product title under the current category, and determines that the category misplacement exists in an event that a difference or ratio between the posterior probability of the product title under the target category and the posterior probability of the product title under the current category is greater than another threshold.

19. The device as recited in claim 17, wherein the category misplacement determination module comprises:

an abnormality in product placement determination module that compares an overall relevancy degree value of the product title under the current category with a threshold, and determines that there is abnormality in product placement in an event that the overall relevancy degree value of the product title under the current category is less than the threshold;

a target category obtaining module that obtains one or more recommended categories for the product title after a determination that there is abnormality in product placement, and obtains a target category that matches the product title based on the product title and the one or more recommended categories;

an overall relevancy degree value comparison module that compares an overall relevancy degree value of the product title under the target category with the overall relevancy degree value of the product title under the current category; and

a determination of category misplacement module that determines that category misplacement exists if a difference or ratio between the overall relevancy degree value of the product title under the target category and the overall relevancy degree value of the product title under the current category is greater than another threshold.

20. The device as recited in claim 17, wherein the TOP value obtaining module comprises:

a stop word searching module that determines if the product title contains one or more stop words based on distribution entropy of words in multiple categories;

a stop word filtering module that filters the one or more stop words; and

an obtaining of TOP value module that obtains the TOP value of each respective word under the current category after the one or more stop words are filtered.