Rule based product classification

• US 9,201,967 B1
• Filed: 05/10/2012
• Issued: 12/01/2015
• Est. Priority Date: 05/10/2012
• Status: Active Grant

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Khoshnevisan, Cyrus, Zhao, Shaojun
• Primary Examiner(s): BROWN, SHEREE N
• Application Number:
  US13/468,400
• Time in Patent Office: 1,300 Days
• Field of Search: 707/710
• US Class Current: 1/1
• CPC Class Codes: G06F 16/35 Clustering; Classification
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G06Q 30/06 Buying, selling or leasing transactions

Abstract

Disclosed are various embodiments for item categorizer. The item categorizer is configured to parse, in at least one computing device, at least a plurality of product descriptions of a like item category for identifying a plurality of keywords with regard to the like item category. Furthermore, the item categorizer selects a plurality of subsets of keywords from the plurality of keywords, each subset of keywords comprises a unique combination of the plurality of keywords with respect to each other. Moreover, the item categorizer is configured to generate a respective rule for each corresponding subset of keywords, wherein an application of each respective rule specifies a respective binary result depending upon whether the corresponding subset of keywords is included in a seller product description.

Claims

1. A non-transitory computer-readable medium embodying a program executable in a computing device, the program comprising a set of instructions that, when executed by a processor of the computing device, cause the computing device to at least:

facilitate access to a catalog of items, the catalog of items comprising an amount of unstructured text for describing each item, wherein each item corresponds to at least one category;

identify a plurality of keywords expressed within the amount of unstructured text with respect to a selected category;

select a plurality of subsets from the plurality of keywords, wherein each of the subsets comprises a unique combination of the plurality of keywords with respect to each other and wherein each of the subsets includes a predetermined number of search terms;

generate a respective rule for each corresponding subset, wherein an application of each respective rule specifies a respective binary result depending upon whether the corresponding subset is included in a prospective seller product description;

apply each respective rule to the prospective seller product description to determine whether a predefined number of the rules indicate a like binary result for indicating whether the prospective seller product description expresses an item within the selected category;

calculate a defect rate corresponding to each respective rule, the defect rate representing a percentage of a plurality of product descriptions correctly classified with each respective rule; and

update each respective rule based at least in part on the corresponding defect rate.

2. The computer-readable medium of claim 1, wherein at least one of the plurality of keywords comprises at least one of a negative keyword, a bi-gram, or a tri-gram.

3. The computer-readable medium of claim 1, wherein the program further comprises instructions that cause the computing device to at least apply each respective rule to at least the amount of unstructured text to determine an accuracy of a set of rules comprising each of the respective rules.

4. The computer-readable medium of claim 1, wherein causing the computing device to identify the plurality of keywords further comprises causing the computing device to use a term frequency inverse document frequency analysis.

5. A system, comprising:

at least one computing device comprising a processor and a memory; and

an application stored in the memory and comprising instructions executable by the processor of the at least one computing device, wherein the application, when executed, causes the at least one computing device to at least;

access a catalog that organizes a plurality of items into a categorical hierarchy, wherein each item is expressed in terms of a product description;

parse at least a portion of the product descriptions associated with a selected category to generate a plurality of keywords for the selected category;

select a subset of the plurality of keywords;

generate at least one rule based on the subset, wherein an application of the at least one rule specifies a respective binary result depending upon whether the subset of the plurality of keywords is included in a prospective seller product description;

apply the at least one rule to the prospective seller product description to determine that the prospective seller product description expresses an item within the selected category;

calculate a percentage of a plurality of product descriptions correctly categorized with the at least one rule; and

update the at least one rule based at least in part on the calculated percentage.

6. The system of claim 5, wherein causing the at least one computing device to select the subset of the plurality of keywords further comprises causing the at least one computing device to select a predetermined number of search terms.

7. The system of claim 6, wherein causing the at least one computing device to generate the at least one rule further comprises causing the at least one computing device to generate a plurality of rules, wherein each of the plurality of rules comprises a respective subset of the plurality of keywords, each respective subset of the plurality of keywords comprises a unique combination of the plurality of keywords with respect to each other, wherein each respective subset includes the predetermined number of search terms.

8. The system of claim 7, wherein the application further causes the at least one computing device to apply the plurality of rules to the prospective seller product description for determining a number of rules that indicate a like binary result, wherein application of the at least one rule to the prospective seller product description comprises a determination that the prospective seller product description expresses an item within the selected category when the number of rules that indicate the like binary result exceeds a threshold amount.

9. The system of claim 5, wherein causing the at least one computing device to parse at least the portion of the product descriptions associated with the selected category to generate the plurality of keywords relating to the selected category further causes the at least one computing device to use a term frequency inverse document frequency analysis to parse the portion of the product descriptions associated with the selected category.

10. The system of claim 5, wherein causing the at least one computing device to parse at least the portion of the product descriptions associated with the selected category to generate the plurality of keywords relating to the selected category further causes the at least one computing device to use a term frequency squared inverse document frequency analysis to parse the portion of the product descriptions associated with the selected category, wherein the term frequency squared inverse document frequency analysis comprises squaring a frequency score of a target term.

11. The system of claim 5, wherein at least one of the plurality of keywords comprises at least one of a negative keyword, a bi-gram, or a tri-gram.

12. The system of claim 5, wherein at least one rule comprises a weight representing the calculated percentage of a plurality of product descriptions correctly categorized with the at least one rule.

13. The system of claim 12, wherein a value of the weight is inversely proportional to the calculated percentage of the plurality of product descriptions correctly categorized with the at least one rule.

14. A method, comprising:

parsing, in at least one computing device, at least a plurality of product descriptions of a like item category for identifying a plurality of keywords associated with the like item category;

selecting, in the at least one computing device, a plurality of subsets of keywords from the plurality of keywords, wherein each of the subsets of keywords comprises a unique combination of the plurality of keywords with respect to each other;

generating, in the at least one computing device, a respective rule for each corresponding subset of keywords, wherein an application of each respective rule specifies a respective binary result depending upon whether the corresponding subset of keywords is included in a seller product description;

calculating, in the at least one computing device, a percentage of the plurality of product descriptions correctly classified by the respective rule to generate a defect rate for the respective rule; and

updating, in the at least one computing device, the respective rule based at least in part on the defect rate.

15. The method of claim 14, further comprising encoding, in the at least one computing device, a user interface for display at a client, wherein the user interface is configured to facilitate receiving a prospective seller catalog, wherein the prospective seller catalog comprises at least one prospective seller product description that expresses at least one seller item.

16. The method of claim 15, further comprising:

identifying, in the at least one computing device, at least one recommended category for the at least one item based on applying at least one generated rule to the at least one prospective seller product description; and

presenting the at least one recommended category in the user interface.

17. The method of claim 14, further:

receiving, in the at least one computing device, a prospective seller product description that expresses an item for sale by the prospective seller;

identifying, in the at least one computing device, a recommended category for the item based on applying at least one generated rule to the prospective seller product description; and

automatically designating, in the at least one computing device, a category for the item.

18. The method of claim 16, wherein the user interface is configured to receive an input from the client indicating whether a seller using the client has rejected the at least one recommended category, wherein the method further comprises flagging the at least one generated rule for review in response to the input from the client.

19. The method of claim 14, further comprising determining, in the at least one computing device, whether a prospective seller product description that expresses a seller item falls within the like item category by applying the respective rule to the prospective seller product description and counting a number of rules that generate a like binary result.

20. The method of claim 14, wherein parsing at least the plurality of product descriptions of the like item category for identifying the plurality of keywords with regard to the like item category further comprises using at least one of term frequency inverse document frequency or term frequency squared inverse document frequency.

21. The method of claim 14, wherein each subset of keywords is selected according to a predetermined number of search terms.

22. The method of claim 14, further comprising applying, in the at least one computing device, the respective rule to a prospective seller product description to determine a recommended category for an item expressed by the prospective seller product description.

23. The method of claim 22, wherein applying the respective rule further comprises applying, in the at least one computing device, the respective rule according to a respective weight, each respective weight being determined based on the corresponding defect rate.