Processes for assessing user affinities for particular item categories of a hierarchical browse structure

• US 7,979,445 B2
• Filed: 10/07/2010
• Issued: 07/12/2011
• Est. Priority Date: 10/13/2003
• Status: Expired due to Fees

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Mason, Zachary J.
• Primary Examiner(s): Vo; Tim T.
• Assistant Examiner(s): Gortayo; Dangelino N
• Application Number:
  US12/900,259
• Publication Number:
  US 20110022606A1
• Time in Patent Office: 278 Days
• Field of Search: 707/1, 707/3, 707/5, 707/10, 707/100, 707/734, 707/748, 707/751, 707/784, 707/949
• US Class Current: 707/748
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising

Abstract

A computer-implemented system provides a browse tree in which items represented within a database are arranged within a hierarchy of item categories. Each time a user selects an item, an amount of credit is cumulatively assigned to the ancestor nodes (categories) of the selected item. The amount of credit assigned to a particular category of the browse tree over time for a given user represents the user'"'"'s predicted affinity for that category. The user'"'"'s relative preferences for some or all of the categories are predicted by calculating differences between the user'"'"'s predicted affinities for such categories and the predicted affinities of a population of users for such categories. Scores reflective of these relative category preferences are used to provide personalized recommendations or other personalized content to the user.

Claims

1. A computer-implemented method, comprising:

recording item selection events in which users select particular items represented in an electronic data repository, said items arranged in a hierarchical browse structure that comprises multiple levels of item categories;

generating relative preference profiles for particular users, each relative preference profile representing preferences of a respective user for particular item categories of said hierarchical browse structure relative to corresponding preference levels of a population of users, wherein generating the relative preference profiles comprises analyzing the recorded item selection events to assess users'"'"' preferences for particular item categories, wherein the relative preference profile for a user comprises a plurality of relative preference scores, each relative preference score corresponding to a respective item category and reflecting a degree to which the user'"'"'s preference level for the item category differs from the user population'"'"'s preference level for the item category; and

selecting items to recommend to users based at least partly on the relative preference profiles of the users;

said method performed in its entirely by a computer system.

2. The method of claim 1, wherein selecting items to recommend to users comprises identifying similar users based on comparisons of the relative preference profiles of the users, and using the identified similar users to implement a collaborative recommendations algorithm.

3. The method of claim 1, wherein generating the relative preference profiles comprises:

determining a total amount of credit to be distributed for an item selection event in which the user selected an item that is categorized in the hierarchical browse structure;

identifying a plurality of ancestor categories of the selected item in the hierarchical browse structure;

dividing said total amount of credit by the number of ancestor categories of the selected item to determine an amount of credit per ancestor category to be distributed for the item selection event; and

assigning said amount of credit per ancestor category to each of the ancestor categories.

4. The method of claim 1, wherein the method comprises generating the relative preference scores by a process that comprises:

generating a first normalized score representing a preference level of the user for a selected item category, said normalized score based on item selections of the user;

generating a second normalized score representing a preference level of the user population for the selected item category, said second normalized score based on item selections of a plurality of users; and

comparing the first normalized score to the second normalized score to determine a relative preference score for the selected item category.

5. The method of claim 4, wherein comparing the first normalized score to the second normalized score comprises applying an entropy function to the first and second normalized scores.

6. The method of claim 1, wherein generating the relative preference profiles comprises:

determining an amount of credit to be distributed for an item selection event in which a user selected an item; and

distributing said amount of credit among a plurality of item categories under which the item falls in said hierarchical browse structure.

7. The method of claim 1, wherein the computer system comprises a plurality of servers.

8. An interactive system, comprising:

a server system coupled to a communications network, said server system configured to provide interactive user access to a hierarchical browse structure in which items represented in a data repository are arranged in a hierarchy of item categories, said hierarchy comprising multiple levels of item categories, said server system configured to record item selection events in which users select particular items represented in the data repository;

an analysis component configured to use the recorded item selection events to generate relative preference profiles for particular users, each relative preference profile representing preferences of a respective user for particular item categories of said hierarchy relative to corresponding category preference levels of a population of users, wherein the relative preference profile for a user comprises a plurality of relative preference scores, each relative preference score corresponding to a respective item category and reflecting a degree to which the user'"'"'s preference level for the item category differs from the user population'"'"'s preference level for the item category; and

a recommendation module configured to use the relative preference profiles to select items to recommend to particular users.

9. The interactive system of claim 8, wherein the interactive system is configured to identify similar users based on comparisons of the relative preference profiles of the users.

10. The interactive system of claim 8, wherein the interactive system is configured to implement a collaborative recommendations process in which users are matched to similar users based on comparisons of the relative preference profiles of the users.

11. The interactive system of claim 8, wherein the recommendation module is configured to rank a set of item recommendations for presentation to a user based at least partly on the relative preference profile of the user.

12. The interactive system of claim 8, wherein the analysis component is configured to generate the relative preference scores by a process that comprises:

generating a first normalized score representing a preference level of the user for a selected item category, said normalized score based on item selections of the user;

generating a second normalized score representing a preference level of the user population for the selected item category; and

comparing the first normalized score to the second normalized score to determine a relative preference score for the selected item category.

13. The interactive system of claim 12, wherein the analysis component is configured to compare the first normalized score to the second normalized score using a relative entropy function.

14. The interactive system of claim 8, wherein the analysis component is configured to assess the user'"'"'s preferences for particular item categories by a process that comprises:

determining an amount of credit to be distributed for an item selection event in which the user selected an item; and

distributing said amount of credit among a plurality of item categories under which the item falls in said hierarchical browse structure.

15. The interactive system of claim 8, wherein the items are products represented in an electronic catalog.

16. The interactive system of claim 8, wherein the analysis component is configured to assess the user'"'"'s preferences for particular item categories by a process that comprises:

determining a total amount of credit to be distributed for an item selection event in which the user selected an item that is categorized in the hierarchical browse structure;

identifying a plurality of ancestor item categories of the selected item in the hierarchical browse structure; and

dividing said total amount of credit by the number of ancestor item categories of the selected item to determine an amount of credit per ancestor item category to be distributed for the item selection event.

17. The interactive system of claim 16, wherein determining the total amount of credit comprises selecting an amount that depends upon a type of the item selection event, such that different total credit amounts are used for different types of item selection events.

18. Non-transitory computer storage having stored thereon executable code that directs a computer system to perform a process that comprises:

recording item selection events in which users select particular items represented in an electronic data repository, said items arranged in a hierarchical browse structure that comprises multiple levels of item categories;

generating relative preference profiles for particular users, each relative preference profile representing preferences of a respective user for particular item categories of said hierarchical browse structure relative to corresponding preference levels of a population of users, wherein generating the relative preference profiles comprises analyzing the recorded item selection events to assess users'"'"' preferences for particular item categories, wherein the relative preference profile for a user comprises a plurality of relative preference scores, each relative preference score corresponding to a respective item category and reflecting a degree to which the user'"'"'s preference level for the item category differs from the user population'"'"'s preference level for the item category; and

selecting items to recommend to users based at least partly on the relative preference profiles of the users.