Identifying categorized misplacement

• US 9,104,968 B2
• Filed: 06/11/2014
• Issued: 08/11/2015
• 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:
  US14/302,332
• Publication Number:
  US 20140297577A1
• Time in Patent Office: 426 Days
• Field of Search: None
• US Class Current: 1/1
• 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 comprising:

obtaining a word frequency of one or more words in a product title under a first category and another word frequency of the one or more words in the product title under a second category;

calculating a first overall word frequency of the product title under the first category based on the word frequency of the one or more words in the product title under the first category and a second overall word frequency of the product title under the second category based on the word frequency of the one or more words in the product title under the second category;

setting a first threshold for the first category and a second threshold for the second category;

storing the first threshold and the second threshold in a storage device; and

comparing the first overall word frequency of the product title with the first threshold and the second overall word frequency of the product title with the second threshold to determine a category of the product title.

2. The method as recited in claim 1, further comprising in an event that the first overall word frequency of the product title is higher than the first threshold, determining that the category of the product title is the first category.

3. The method as recited in claim 1, further comprising:

determining one or more stop words from the product title; and

filtering the one or more stop words from the product title prior to obtaining the word frequency of the one or more words in the product title under the first category and another word frequency of the one or more words in the product title under the second category, the one or more words in the product title including none of the one or more stop words.

4. The method as recited in claim 3, wherein the determining the one or more stop words from the product title comprises presetting a table of stop words.

5. The method as recited in claim 4, further comprising:

determining a distribution entropy of a word under multiple categories;

determining that the word is a stop word when the distribution of the word is higher than a threshold for stop word; and

adding the word into the table of stop words.

6. The method as recited in claim 5, further comprising:

ranking multiple words based on their respective distribution entropy under the multiple categories; and

determining the threshold for stop word based on an order of the ranking.

7. The method as recited in claim 1, wherein the first threshold is different from the second threshold.

8. The method as recited in claim 1, wherein the obtaining the word frequency of the one or more words in the product title under the first category and another word frequency of the one or more words in the product title under the second category comprises:

in an event that a respective word does not appear in the first category, counting that the respective word appears once in the first category.

9. A method comprising:

obtaining one or more recommended categories for each of multiple words in a product title;

combining the one or more recommended categories for each of the multiple words to obtain a plurality of recommended categories for the product title;

using a word frequency of a respective word in the multiple words under each of the plurality of recommended categories for the product title as a weight of the respective word under a respective recommended category for the product title;

calculating a sum of weights of the multiple words for each of the plurality of recommended categories for the product title; and

choosing one or more recommended categories for the product title based on a result of calculating the sum of weights of the multiple words for each of the plurality of recommended categories.

10. The method of claim 9, wherein the obtaining the one or more recommended categories for each of the multiple words in the product title comprises:

selecting one or more recommended categories for the respective word in the multiple words based on the word frequency of the respective word under different categories.

11. The method of claim 9, wherein the choosing one or more recommended categories for the product title based on the result of calculating the sum of weights of the multiple words for each of the plurality of recommended categories comprises:

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

12. The method of claim 9, wherein the choosing one or more recommended categories for the product title based on the result of calculating the sum of weights of the multiple words for each of the plurality of recommended categories comprises:

ranking the plurality of recommended categories according to respective sum of weights of the multiple words for each of the plurality of recommend categories; and

choosing the one or more recommended categories with top rankings as candidates of a target category.

13. The method as recited in claim 9, further comprising:

determining one or more stop words from the product title; and

filtering the one or more stop words from the product title prior to selecting the one or more recommended categories for each of the multiple words in a product title, the multiple words in the product title including none of the one or more stop words.

14. The method as recited in claim 13, wherein the determining the one or more stop words from the product title comprises presetting a table of stop words.

15. The method as recited in claim 14, further comprising:

determining a distribution entropy of a word under multiple categories;

determining that the word is a stop word when the distribution of the word is higher than a threshold for stop word; and

adding the word into the table of stop words.

16. The method as recited in claim 15, further comprising:

ranking multiple words based on their respective distribution entropy under the multiple categories; and

determining the threshold for stop word based on an order of the ranking.

17. A method comprising:

obtaining one or more recommended categories for a product title;

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

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

choosing a target category from the one or more recommended categories based on a result of the respective product for the respective recommended category.

18. The method as recited in claim 17, wherein the choosing the target category from the one or more recommended categories based on the result of the respective product for the respective recommended category comprises choosing a recommended category with a highest product as the target category.

19. The method as recited in claim 17, further comprising

determining one or more stop words from the product title;

and

filtering the one or more stop words from the product title, the one or more words in the product title including none of the one or more stop words.

20. The method as recited in claim 19, wherein the determining one or more stop words from the product title comprises:

determining a distribution entropy of a word under multiple categories;

determining that the word is a stop word when the distribution of the word is higher than a threshold for stop word; and

adding the word into a table of stop words.