Recommendation system with multiple integrated recommenders

• US 7,949,659 B2
• Filed: 06/29/2007
• Issued: 05/24/2011
• Est. Priority Date: 06/29/2007
• Status: Active Grant

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Kim, Sung H., Lam, Shing Yan, Ionkov, George M., Chan, James D., Witt, Brett W., Chakrabarti, Kushal
• Primary Examiner(s): PATEL, NIKETA I
• Assistant Examiner(s): PATEL, NIKETA I
• Application Number:
  US11/772,010
• Publication Number:
  US 20090006373A1
• Time in Patent Office: 1,425 Days
• Field of Search: 707/748, 707/758, 707/732, 707/737, 707/769
• US Class Current: 707/732
• CPC Class Codes: G06F 16/335 Filtering based on additional data, e.g. user or group profiles filtering in web context G06F16/9535, G06F16/9536

Abstract

A recommendations system is provided in various embodiments for selecting items to recommend to a user. The system includes a recommendation engine with a plurality of recommenders, and each recommender identifies a different type of reason for recommending items. In one embodiment, each recommender retrieves item preference data and generates candidate recommendations responsive to a subset of that data. The recommenders also score the candidate recommendations. In certain embodiments, a normalization engine normalizes the scores of the candidate recommendations provided by each recommender. A candidate selector selects at least a portion of the candidate recommendations based on the normalized scores to provide as recommendations to the user. The candidate selector also outputs the recommendations with associated reasons for recommending the items.

Claims

1. A recommendations system for selecting items to recommend to a user, the system comprising:

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

a recommendation engine comprising a plurality of recommenders, each of the recommenders configured to implement a different recommendation algorithm such that each recommender is configured to generate recommendations targeted to a different detected interest of a user by at least;

retrieving item preference data reflective of actions performed by a user;

detecting the interest of the user from the item preference data;

generating candidate recommendations responsive to the detected interest of the user,

identifying one or more reasons for recommending the candidate recommendations, and

scoring the candidate recommendations to provide relative indications of the strength of the candidate recommendations,

wherein at least some of the recommenders are modular, such that one or more of the recommenders can be selectively removed from the recommendation engine in response to outputting recommendations of less usefulness than recommendations of other of the recommenders, and such that one or more new recommenders can be selectively added to the recommendation engine to target one or more additional user interests;

a normalization engine configured to normalize the scores of the candidate recommendations output by each recommender, wherein the normalization engine is further configured to;

apply weights to the recommenders to adjust the normalized scores to produce adjusted normalized scores for the candidate recommendations, and

adjust a selected one of the weights for a selected recommender in response to a demonstrated affinity by the user for items output from the selected recommender, to thereby emphasize the output of the selected recommender; and

a candidate selector component configured to;

select at least a portion of the candidate recommendations based on the adjusted normalized scores to provide as recommendations to the user, and

output the recommendations with associated textual reasons for recommending the items.

2. The system of claim 1, wherein a first one of the recommenders generates a first recommendation by determining whether the first recommendation can complete a series of items owned by the user, and wherein a second one of the recommenders generates a second recommendation by determining whether the second recommendation can be used as a gift for a friend of the user.

3. The system of claim 1, wherein the textual reasons comprise text indicating that a recommendation could be used as a gift for a friend'"'"'s upcoming birthday.

15. The system of claim 1, wherein the candidate selector component is further configured to output the recommendations and associated textual reasons for display to the user.

16. The system of claim 1, wherein the computer system comprises multiple physical computers.

4. A computer-implemented method of selecting items to recommend, the method comprising:

by a computer system comprising computer hardware;

retrieving item preference data reflective of actions performed by a user;

providing the item preference data to a plurality of recommenders, each recommender configured to implement a different recommendation algorithm such that each recommender is configured to generate recommendations targeted to a different detected interest of a user by at least;

generating candidate recommendations responsive to a subset of the item preference data, the subset of the item preference data reflecting the detected interest, and

identifying one or more reasons for recommending the candidate recommendations, the one or more reasons reflecting the detected interest;

wherein one or more of the recommenders are modular, such that the one or more recommenders are configured to not be used in response to outputting recommendations of less usefulness than recommendations of other of the recommenders, and such that one or more new recommenders are configured to be included to target one or more additional user interests;

applying weights to at least some of the recommenders and adjusting the weights over time in response to a demonstrated affinity by the user for items output from at least some of the recommenders, to thereby emphasize the output of at least some of the recommenders;

selecting at least a portion of the candidate recommendations to provide as recommendations to the user; and

outputting the recommendations and the one or more reasons for recommending the items, each of the one or more reasons comprising a textual explanation for recommending one or more of the items.

5. The method of claim 4, wherein each recommender is further configured to generate scores for the candidate recommendations.

6. The method of claim 5, further comprising normalizing the scores from each recommender.

7. The method of claim 4, wherein the one or more reasons indicate that a recommendation could complete a series of items owned by the user.

8. The method of claim 4, wherein the one or more reasons indicate that a recommendation could be used as a gift for a friend'"'"'s upcoming birthday.

9. The method of claim 4, wherein the one or more reasons indicate that a recommendation is similar to an item searched for by the user.

10. The method of claim 4, wherein the one or more reasons indicate that a recommendation is an item that might be purchased by an early adopter, and the user is an early adopter.

11. The method of claim 4, wherein the one or more reasons indicate that a recommendation is a top seller in a browse node of interest to the user.

12. The method of claim 4, wherein the item preference data comprises one or more of the following:

browse history data, series data, search data, purchase history data, friends data, browsing interests data, tags data, and top sellers data.

13. The method of claim 4, wherein each recommender is further configured to provide one or more reason types that map to reason text.

17. The method of claim 4, wherein said outputting the recommendations and one or more reasons for recommending the items comprises outputting the recommendations and the one or more reasons for display to the user.

14. A non-transitory computer-readable medium having instructions stored thereon which cause a computer system to perform a method of selecting items to recommend to a user, the method comprising:

retrieving item preference data reflective of actions performed by a user;

providing the item preference data to a plurality of recommendation algorithms, each recommendation algorithm corresponding to a different type of reason for recommending items, each recommendation algorithm configured to generate recommendations targeted to a different detected interest of the user by at least;

generating candidate recommendations responsive to a subset of the item preference data, the subset of the item preference data reflecting the detected interest, and

identifying one or more reasons for recommending the candidate recommendations, the one or more reasons reflecting the detected interest;

wherein one or more of the recommendation algorithms are modular, such that the one or more recommendation algorithms are configured to not be used in response to outputting recommendations of less usefulness than recommendations of other of the recommendation algorithms, and such that one or more new recommendation algorithms are configured to be included to target one or more additional user interests;

applying weights to the recommendation algorithms and adjusting the weights in response to a demonstrated affinity by the user for items output from at least some of the recommendation algorithms, to thereby emphasize the output of at least some of the recommendation algorithms;

selecting at least a portion of the candidate recommendations to provide as recommendations to the user; and

outputting the recommendations and the one or more reasons for recommending the items, each of the one or more reasons comprising a textual explanation for recommending one or more of the items.

18. The non-transitory computer-readable medium of claim 14, in combination with a computer system that executes said instructions to perform said method.