Probabilistic recommendation system

• US 8,301,623 B2
• Filed: 05/22/2007
• Issued: 10/30/2012
• Est. Priority Date: 05/22/2007
• Status: Expired due to Fees

• Current Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Original Assignee: Amazon Technologies, Inc. (Amazon.com, Inc.)
• Inventors: Chakrabarti, Kushal, Smith, Brent
• Primary Examiner(s): MENG, JAU SHYA
• Application Number:
  US11/752,251
• Publication Number:
  US 20080294617A1
• Time in Patent Office: 1,988 Days
• Field of Search: None
• US Class Current: 707/723
• 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 recommendations system uses probabilistic methods to select, from a candidate set of items, a set of items to recommend to a target user. The methods can effectively introduce noise into the recommendations process, causing the recommendations presented to the target user to vary in a controlled manner from one visit to the next. The methods may increase the likelihood that at least some of the items recommended over a sequence of visits will be useful to the target user. Some embodiments of the methods are stateless such that the system need not keep track of which items have been recommended to which users.

Claims

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

generating a ranked set of candidate items to recommend to a target user based at least partly on item selection actions performed by the target user, the ranked set of candidate items comprising candidate items and associated rankings of the candidate items;

subsequent to said generating, probabilistically varying the rankings of at least some of the candidate items in the ranked set of candidate items automatically by a computer to generate a probabilistically-varied ranked set of candidate items,

said probabilistically varying the rankings comprising using a pseudo-random number generator to vary the rankings by an amount that is based at least in part on a number of the item selection actions performed by the target user;

selecting, from said probabilistically-varied ranked set, a first subset of the candidate items to present to the target user; and

subsequent to said selecting, repeating at least said probabilistically varying the rankings and said selecting to thereby select a second subset of the candidate items to present to the target user, wherein the second subset of the candidate items comprises one or more items that are not members of the first subset of the candidate items.

2. The computer-implemented method of claim 1, wherein selecting, from said probabilistically-varied ranked set of candidate items, the first subset of the candidate items to present to the user comprises selecting a most-highly ranked subset of items.

3. The computer-implemented method of claim 1, wherein at least some of the candidate items are associated with one or more identified reasons for being recommended to the target user.

4. The computer-implemented method of claim 3, wherein the amount which the ranking of a candidate item is probabilistically varied further depends on the number of identified reasons for recommending the candidate item to the target user.

5. The computer-implemented method of claim 4, wherein probabilistically varying the rankings comprises causing candidate items having relatively low numbers of reasons to be probabilistically varied to a greater degree than candidate items having relatively high numbers of reasons.

6. The computer-implemented method of claim 1, wherein probabilistically varying the rankings comprises causing candidate items having relatively low category scores to be probabilistically varied to a greater degree than candidate items having relatively high category scores.

7. The computer-implemented method of claim 1, wherein probabilistically varying the rankings comprises generating values for at least a portion of the candidate items and adding probabilistic noise to the values.

8. The computer-implemented method of claim 7, wherein adding probabilistic noise to the values comprises generating a probability distribution corresponding to each value, wherein the characteristics of the probability distribution depend on the corresponding value.

9. The computer-implemented method of claim 8, further comprising sampling values from each probability distribution to generate a score for said at least a portion of the candidate items, wherein the score is responsive to the sampled value.

10. The computer-implemented method of claim 1, wherein probabilistically varying rankings of at least some of the candidate items in the ranked set further comprises probabilistically varying the rankings to achieve a target overlap value of same recommended items between the first and second subsets.

11. The computer-implemented method of claim 1, wherein said generating, said probabilistically varying, and said selecting are implemented by one or more machines.

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

by a computer system comprising computer hardware;

identifying items selected from an electronic catalog by a target user;

for each of the identified items, selecting associated items from a table of associated items to produce a set of candidate recommendations, said selecting comprising providing an initial score for each candidate recommendation;

for each of the candidate recommendations,

identifying a number of the item selections made by the target user that resulted in the candidate recommendation being selected from the table of associated items, and

calculating a new score for the candidate recommendation using a value selected from a probability function, the probability function configured to provide the value based at least partly on the identified number of item selections, wherein the probability function at least uses a pseudo-random number generator to calculate the new score;

outputting a first subset of the candidate recommendations for presentation to the target user, each of the candidate recommendations being ranked in the output according to the new score for that candidate recommendation; and

subsequent to said outputting, repeating at least said calculating the new score for each of the candidate recommendations and said outputting to thereby output a second subset of the candidate recommendations, wherein the second subset of the candidate recommendations comprises at least some different items from the first subset of the candidate recommendations.

13. The computer-implemented method of claim 12, wherein the at least one characteristic of the probability function comprises entropy.

14. The computer-implemented method of claim 13, wherein the entropy of the probability function decreases as the number of reasons increases.

15. The computer-implemented method of claim 12, wherein the probability function is a normal distribution.

16. The computer-implemented method of claim 15, wherein the probability function is further configured to provide the value based at least partly on a mean configured to depend on the number of item selections.

17. The computer-implemented method of claim 15, wherein the probability function is further configured to provide the value based at least partly on a standard deviation configured to depend on the number of item selections.

18. The computer-implemented method of claim 12, wherein the new scores depend upon strength values associated with specific item selections.

19. The computer-implemented method of claim 12, wherein calculating the new score for the candidate recommendation further comprises assigning negative feedback score to the candidate recommendation, wherein the negative feedback score is dependent on the number of said item selections.

20. The computer-implemented method of claim 19, wherein calculating the new score for the candidate recommendation further comprises determining one or more category scores responsive to the number of item selections in one or more categories selected from the group of a purchase history category, an unbought interests category, and a recent activity category.

21. The computer-implemented method of claim 20, wherein calculating the new score for the candidate recommendation further comprises combining the one or more category scores and the negative feedback score.

22. The computer-implemented method of claim 12, wherein said identifying, said assigning, and said using are implemented automatically by one or more computer processors.

23. A recommendation engine for selecting items to recommend, the recommendation engine comprising:

a computer system comprising one or more processors, said computer system programmed to implement;

a candidate generator configured to;

generate a ranked set of candidate items to recommend to a target user, the ranked set of candidate items comprising candidate items and associated rankings of the candidate items;

a probabilistic scorer configured to, subsequent to the candidate generator generating the ranked set of candidate items, probabilistically vary the rankings of at least some of the candidate items in the ranked set of candidate items to generate a probabilistically-varied ranked set of candidate items, the probabilistic scorer configured to probabilistically vary the rankings by at least using a pseudo-random number generator to vary the rankings by an amount based at least in part on one or more item selection actions performed by the target user; and

a candidate filter configured to select, from said probabilistically-varied ranked set, a first subset of the candidate items to present to the user;

wherein the probabilistic scorer is further configured to probabilistically vary the rankings of at least some of the candidate items a second time to generate a second probabilistically-varied ranked set of candidate items, and wherein the candidate filter is further configured to select a second subset of the second probabilistically-varied ranked set of candidate items to present to the user, wherein the second subset comprises at least some different items from the first subset of the candidate items.

24. The recommendation engine of claim 23, wherein the degree to which the probabilistic scorer probabilistically varies the ranking of a candidate item depends on a number of identified reasons for recommending the candidate item to the target user.

25. The recommendation engine of claim 24, wherein probabilistically varying the rankings comprises causing candidate items having relatively low numbers of reasons to be probabilistically varied to a greater degree than candidate items having relatively high numbers of reasons.

26. A method for varying recommendations presented to users, the method comprising:

by a computer system comprising computer hardware;

selecting items as candidates for recommendation to a target user;

ranking the items using a recommendation process that depends at least in part on item selection actions performed by the target user, the rankings configured to estimate an optimal order for presenting the items as recommendations for the target user;

varying the rankings of the items to produce varied rankings by at least using a probabilistic function that depends at least in part on the item selection actions performed by the target user so as to estimate a less optimal order for presenting the items to the target user, the probabilistic function being implemented at least in part by a pseudo-random number generator, such that some of the rankings are increased for some of the items and other of the rankings are decreased for other of the items;

outputting a first subset of the items as a most highly-ranked subset of the items according to the varied rankings as additional recommendations for presentation to the target user; and

subsequent to said outputting, repeating said selecting, ranking, varying, and outputting, based on the same item selection actions performed by the target user, to thereby output a second subset of the items to recommend to the target user, the second subset of items comprising at least some different items from the first subset of items.

27. The method of claim 26, wherein said varying the rankings comprises adjusting the rankings based at least in part on a number of item selections performed by the target user.

28. The method of claim 27, wherein said varying the rankings comprises increasing ranking variability when the number of item selections is relatively higher and decreasing the ranking variability when the number of item selections is relatively lower.