Systems and methods for automatic item classification

• US 8,510,307 B1
• Filed: 12/22/2009
• Issued: 08/13/2013
• Est. Priority Date: 12/22/2009
• Status: Active Grant

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Touloumtzis, Michael Paul, Kansal, Abhishek, Khoshnevisan, Cyrus, Nguyen, Lam D.
• Primary Examiner(s): TIMBLIN, ROBERT M
• Application Number:
  US12/645,405
• Time in Patent Office: 1,330 Days
• Field of Search: 707/737, 707/740, 707/778
• US Class Current: 707/740
• CPC Class Codes: G06F 16/353 into predefined classes

Abstract

An item categorization service is described that automatically categorizes items of interest to a user. The user may possess an item that they wish to offer for sale using a network-based service. The user may submit item information to the categorization service to categorize the item of interest. Upon receipt, the categorization service may assess the relevance of the item information to hierarchically organized categories maintained by the network-based service. Categories having the highest relevance may be identified as first category candidates. The deepest common ancestor of the first category candidates may be identified the first category. One or more categories related to the first category may also be identified and subjected to relevance assessment. Those related categories having the highest relevance may be identified as second category candidates. The deepest common ancestor of the second category candidates may be identified as a second category for the item of interest.

Claims

1. A computer-implemented method for categorizing items, the method comprising:

under control of one or more configured computer systems;

obtaining item information regarding an item of interest;

obtaining one or more hierarchically organized categories for items;

determining a relevance of the item information to each of the categories;

identifying one or more first category candidates and one or more second category candidates for the item of interest from the categories based at least upon the determined relevance of the item information to each of the categories;

determining a first likelihood value for each pairing of an identified first category candidate and an identified second category candidate, the first likelihood value representing a statistical likelihood that the pairing of the identified first category candidate and the identified second category candidate are paired categories for the item of interest;

assigning, as categories for the item of interest, the identified first category candidate and the identified second category candidate of a pairing having a first likelihood value that is greater than a threshold value;

determining a second likelihood value that represents a statistical likelihood that the identified first category candidate and at least two identified second category candidates are paired categories to the item of interest; and

if the second likelihood value is greater than the threshold value, assigning, as categories for the item of interest, each of the at least two identified second category candidates and the first identified category candidate.

2. The computer-implemented method of claim 1, wherein the item information regarding the item of interest is obtained from a request to categorize the item of interest.

3. The computer-implemented method of claim 1,

wherein the item information comprises at least one item attribute and a corresponding attribute value;

wherein each of the categories is described by at least one category attribute and a corresponding category attribute value; and

wherein determining a relevance of the item information to each of each of the categories comprises assessing the relevance of at least one of the item attributes and item attribute values of the item of interest to one or more of the category attributes and category attribute values of the respective categories.

4. The computer implemented method of claim 3, wherein identifying the second category candidates comprises:

determining a common ancestor of the hierarchically organized first category candidates; and

identifying a selected number of categories having a category attribute value corresponding to a selected category attribute of the determined common ancestor of the first category candidates.

5. The method of claim 4, wherein identifying the selected number of categories having a category attribute value corresponding to a selected category attribute of the determined common ancestor of the first category candidates comprises obtaining at least one rule that identifies at least one category attribute of the determined common ancestor of the first category candidates.

6. The computer-implemented method of claim 1, wherein determining a first likelihood value comprises obtaining one or more likelihood values for the pairing of an identified first category candidate and an identified second category candidate from a data store.

7. The computer implemented method of claim 1, wherein identifying the first category candidates comprises identifying a selected number of categories having the highest determined relevance of the item information to each of the categories.

8. A computer-implemented method for assessing a confidence level in an item categorization for an item of interest, the method comprising:

under control of one or more configured computer systems;

obtaining item information regarding an item of interest;

obtaining categories for items;

determining one or more relevance values representing a similarity of the item information to each of the obtained categories;

selecting one or more of the obtained categories based upon their determined relevance value;

assigning an obtained category as a first category for the item of interest based, at least in part, upon an analysis of the selected categories;

determining a relevance clustering parameter based at least in part upon a pairwise difference between each of the determined relevance values of the selected categories;

determining a category distance parameter representing a similarity of selected categories to one another;

determining a relevance clustering classification characterizing the confidence level in the assigned first category, wherein;

the relevance clustering classification is determined to be tight if the relevance clustering parameter is less than a first threshold value; and

the relevance clustering classification is determined to be loose if the relevance clustering parameter is greater than the first threshold value;

determining a category distance classification further characterizing the confidence level in the assigned first category, wherein;

the category distance classification is determined to be close if the category distance parameter based upon the determined category distances of the selected categories is less than a second threshold value; and

the category distance classification is determined to be far if the category distance parameter is greater than the second threshold value; and

determining a confidence level in the assigned first category based upon the relevance clustering parameter, the category distance parameter, the relevance clustering classification, and the category distance classifications.

9. The computer-implemented method of claim 8, wherein the relevance clustering parameter comprises at least one of a maximum pairwise difference between determined relevance values, an average pairwise difference between determined relevance values, and a sum of the pairwise differences between respective determined relevance values.

10. The computer-implemented method of claim 8, wherein the category distance parameter comprises at least one of a maximum pairwise difference between determined category distances, an average pairwise difference between determined category distances, and a sum of the pairwise differences between respective determined category distances.

11. The computer implemented method of claim 8, wherein the item information comprises at least one item attribute and a corresponding attribute value and the obtained categories are described by at least one category attribute and a corresponding category attribute value and wherein the one or more obtained categories are hierarchically organized with respect to one another.

12. The computer-implemented method of claim 11, wherein determining one or more relevance values representing the similarity of the item information to each of the obtained categories comprises assessing the relevance of at least one of the item attributes and item attribute values of the item of interest to one or more of the category attributes and category attribute values of the obtained categories.

13. The computer-implemented method of claim 11, wherein assigning an obtained category to the item of interest comprises selecting a deepest common ancestor of the selected categories.

14. The computer-implemented method of claim 13, wherein the selected categories comprise a selected number of the obtained categories having the highest determined relevance values.

15. The computer-implemented method of claim 14, further comprising assigning a second category for the item of interest, the second category assignment comprising:

identifying a category attribute of the first category;

designating a selected number of the obtained categories that possess the identified category attribute of the first category as second category candidates; and

assigning, as the second category for the item of interest, a common ancestor of the second category candidates.

16. The computer-implemented method of claim 15, wherein the second category candidates comprise the selected number of highest relevance categories that possess the identified attribute of the first category.

17. The computer-implemented method of claim 16, wherein the common ancestor of the first and second category candidates occurs in a majority of the first and second category candidates.

18. The computer-implemented method of claim 8, wherein the relevance clustering is classified as tight and the category distance is classified as close.

19. The computer-implemented method of claim 18 further comprising identifying the confidence in the item categorization of the item of interest as high.

20. The computer-implemented method of claim 8, wherein the relevance clustering is classified as loose and the category distance is classified as close.

21. The computer-implemented method of claim 20, further comprising identifying the confidence in the item categorization of the item of interest as medium-high.

22. The computer-implemented method of claim 8, wherein the relevance clustering is classified as loose and the category distance is classified as far.

23. The computer-implemented method of claim 22, further comprising identifying the confidence in the item categorization of the item of interest as medium.

24. The computer-implemented method of claim 8, wherein the relevance clustering is classified as tight and the category distance is classified as far.

25. The computer-implemented method of claim 24, further comprising identifying the confidence in the item categorization of the item of interest as low.

26. The computer-implemented method of claim 8, wherein the relevance clustering classification is determined to be tight and the category distance classification is determined to be close and further comprising:

identifying the selected category having the highest relevance to the item information; and

assigning one or more of the attribute values of the identified selected category as a category for the item of interest.

27. The computer-implemented method of claim 8, wherein the relevance clustering classification is determined to be tight and the category distance classification is determined to be far and further comprising:

determining a common ancestor occurring in each of the selected categories; and

assigning the common ancestor of the selected categories as a category of the item of interest.