System for detecting associations between items

• US 7,779,040 B2
• Filed: 09/27/2007
• Issued: 08/17/2010
• Est. Priority Date: 09/27/2007
• Status: Active Grant

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Hicks, Cory
• Primary Examiner(s): Lu; Kuen S
• Assistant Examiner(s): PHAM, TUAN A
• Application Number:
  US11/863,108
• Publication Number:
  US 20090089273A1
• Time in Patent Office: 1,055 Days
• Field of Search: None
• US Class Current: 707/795
• CPC Class Codes: G06F 16/00 Information retrieval; Database structures therefor; File system structures therefor
• CPC Class Codes: G06F 7/00 Methods or arrangements for processing data by operating upon the order or content of the data handled logic circuits H03K19/00
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/0631 Item recommendations

Abstract

A method of detecting associations between items can include identifying a plurality of items represented in a data repository from which to select items to recommend to a target user, each item including one or more attributes. A degree of fit between an item'"'"'s attributes and other items is calculated. The degree of fit can indicate the relevance of the attributes of one item to the other item. A degree of association between the two items is calculated based at least in part on the calculated degree of fit. The degree of association between the two items can indicate the relatedness of the two items. Based on this degree of association, an association between the items can be stored in a data repository.

Claims

1. A computer-implemented method of detecting associations between items represented in a data repository, the method comprising:

by a computer system comprising computer hardware;

identifying a plurality of items represented in a data repository from which to select items to recommend to a target user, each item having one or more attributes;

calculating degrees of fit of the attributes of a first item with a second item, wherein calculating the degrees of fit comprises;

forming an initial matrix of values, each of the values reflecting whether the first and second items possess a selected one of the attributes;

calculating a singular value decomposition of the initial matrix, the singular value decomposition comprising a left matrix, a singular value matrix, and a right transpose matrix;

reducing the dimension of one or more of the left, singular value, and right transpose matrices to create a reduced singular value decomposition; and

multiplying the matrices of the reduced singular value decomposition to create a reduced-dimension matrix approximating the initial matrix, wherein selected values of the reduced-dimension matrix reflect the degrees of fit of the attributes of the first item with the second item;

calculating a correlation value between the values in the reduced-dimension matrix corresponding to the first and second items, the correlation value representing a degree of association between the first and second items, the degree of association between the first and second items reflecting the relatedness of the first and second items;

programmatically using the calculated correlation value between the first and second items to create, in computer storage, an association between the first and second items; and

programmatically using the association to generate a recommendation for the target user.

2. The method of claim 1, wherein the items include at least one of movies, apparel, music, books, and television shows.

3. The method of claim 2, wherein the items comprise movies and the attributes comprise contributors to the movies, the contributors including at least one of actors, directors, producers, narrators, and writers.

4. The method of claim 3, wherein calculating a degree of fit further comprises predicting which actors would likely contribute to a movie with other actors.

5. The method of claim 1, wherein at least some of the values of the initial matrix are weighted based at least partly on a type of the attributes.

6. A computer-implemented method of detecting associations between items represented in a data repository, the method comprising:

by a computer system comprising computer hardware;

identifying a plurality of items represented in a data repository from which to select items to recommend to a target user, each item comprising one or more attributes;

calculating degrees of fit of the attributes of a first item with a second item, said calculating the degrees of fit comprising;

providing an initial matrix of values, each of the values reflecting whether the first and second items possess a selected attribute,

calculating a singular value decomposition of the initial matrix, the singular value decomposition comprising a left matrix, a singular value matrix, and a right transpose matrix,

reducing the dimension of one or more of the left, singular value, and right transpose matrices to create a reduced singular value decomposition, and

multiplying the matrices of the reduced singular value decomposition to create a reduced-dimension matrix approximating the initial matrix, wherein selected values of the reduced-dimension matrix reflect the degrees of fit of the attributes of the first item with the second item;

calculating a degree of association between the first and second items based at least in part on the calculated degrees of fit, the degree of association between the first and second items reflecting relatedness of the first and second items;

storing the calculated degree of association in computer storage; and

programmatically using the association to generate a recommendation for the target user.

7. The method of claim 6, wherein calculating a degree of association between the first and second items based on the calculated degree of fit comprises calculating correlation values between values in the reduced-dimension matrix.

8. The method of claim 6, wherein the items include at least one of movies, apparel, music, books, e-books, television shows, and electronics.

9. The method of claim 8, wherein the items are movies and the attributes comprise contributors to the movies, the contributors including at least one of actors, directors, producers, narrators, writers, and film studios.

10. The method of claim 9, wherein calculating a degree of fit further comprises assigning values to the attributes based at least in part on popularity of the contributors.

11. The method of claim 9, wherein at least some of the values of the initial matrix are given greater weight for contributors that are actors than contributors that are producers.

12. The method of claim 9, wherein at least some of the values of the initial matrix are given greater weight for contributors that are producers than contributors that are directors.

13. The method of claim 9, wherein at least some of the values of the initial matrix are given greater weight for contributors that are supporting actors than contributors that are lead actors.

14. The method of claim 9, wherein at least some of the values of the initial matrix are given greater weight for contributors that produce or direct fewer movies than other contributors.

15. The method of claim 6, wherein calculating a degree of fit further comprises assigning values to the attributes based at least in part on attribute type.

16. The method of claim 6, wherein at least some of the values of the initial matrix are weighted based at least partly on a type of the attributes.

17. A system for detecting associations between items represented in a data repository, the system comprising:

a computer system comprising computer hardware, the computer system programmed to implement;

an item attributes analysis component configured to;

identify a plurality of items represented in a data repository from which to select items to recommend to a target user, each item comprising one or more attributes;

calculate degrees of fit of the attributes of a first item with a second item, the degrees of fit reflecting the relevance of the attributes of the first item to the second item, wherein the item attributes analysis component calculates the degrees of fit by at least;

providing an initial matrix of values, each of the values reflecting whether the first and second items possess a selected attribute,

calculating a singular value decomposition of the initial matrix, the singular value decomposition comprising a left matrix, a singular value matrix, and a right transpose matrix,

reducing the dimension of one or more of the left, singular value, and right transpose matrices to create a reduced singular value decomposition, and

multiplying the left matrix, the singular value matrix, and the right transpose matrix to form a reduced-dimension matrix approximating the initial matrix, wherein selected values of the reduced-dimension matrix reflect the degrees of fit of the attributes of the first item with the second item; and

an association analysis component configured to;

calculate a degree of association between the first and second items based at least in part on the calculated degrees of fit, the degree of association between the first and second items indicating the relatedness of the first and second items; and

store the calculated degree of association in computer storage.

18. The apparatus of claim 17, wherein the item attributes analysis component is further configured to assign values to the attributes based at least in part on attribute type.

19. The system of claim 17, wherein at least some of the values of the initial matrix are weighted based at least partly on a type of the attributes.

20. A computer-implemented method of detecting associations between media works, the method comprising:

by a computer system comprising computer hardware;

identifying a plurality of media works represented in a data repository from which to select for recommending to a target user;

calculating degrees of fit between contributors and the media works, said calculating the degrees of fit comprising;

providing an initial matrix comprising first values, each of the first values reflecting whether one of the contributors contributed to one of the media works,

calculating a singular value decomposition of the initial matrix, the singular value decomposition comprising a left matrix, a singular value matrix, and a right transpose matrix,

reducing the dimension of one or more of the left, singular value, and right transpose matrices to create a reduced singular value decomposition, and

multiplying the matrices of the reduced singular value decomposition to form a reduced-dimension matrix comprising second values, each of the second values reflecting a degree of fit between one of the contributors and one of the media works;

calculating associations between selected ones of the media works based at least in part on the calculated degrees of fit; and

programmatically accessing one or more of the associations to generate recommendations for the target user.

21. The method of claim 20, wherein the works comprise at least one of the following:

movies, television shows, music, and books.

22. The method of claim 20, wherein the contributors comprise at least one of the following:

actors, authors, musicians, directors, producers, narrators, and writers.

23. The method of claim 20, wherein calculating a degree of fit further comprises assigning values to the contributors based at least in part on contributor type.

24. The method of claim 20, wherein calculating a degree of fit further comprises assigning values to the contributors based at least in part on customer reviews of the contributors.

25. The method of claim 20, wherein at least some of the values of the initial matrix are weighted based at least partly on a type of the attributes.

26. The method of claim 20, wherein at least some of the values of the initial matrix are given greater weight for contributors that are actors than contributors that are producers.

27. The method of claim 20, wherein at least some of the values of the initial matrix are given greater weight for contributors that are producers than contributors that are directors.

28. The method of claim 20, wherein at least some of the values of the initial matrix are given greater weight for contributors that are supporting actors than contributors that are lead actors.

29. The method of claim 20, wherein at least some of the values of the initial matrix are given greater weight for contributors that produce or direct fewer movies than other contributors.

30. The method of claim 20, wherein computer system comprises a plurality of physical computers.