System and method for item recommendations

• US 8,700,448 B2
• Filed: 12/05/2001
• Issued: 04/15/2014
• Est. Priority Date: 05/10/2001
• Status: Expired due to Fees

• Current Assignee: Kyndryl Incorporated
• Original Assignee: International Business Machines Corporation
• Inventors: Bertram, Ralf, Schrimpf, Gerhard, Stamm-Wilbrandt, Hermann
• Primary Examiner(s): Hewitt, II, Calvin L.
• Assistant Examiner(s): SHERR, CRISTINA O
• Application Number:
  US10/007,583
• Publication Number:
  US 20020178057A1
• Time in Patent Office: 4,514 Days
• Field of Search: 705 7- 20, 705/14, 705 26- 27, 707 1- 6, 707 9- 10, 707102-103, 707/999, 709/203
• US Class Current: 705/14.1
• CPC Class Codes: G06F 16/90324 using system suggestions
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/0222 During e-commerce, i.e. online transactions
• CPC Class Codes: G06Q 30/0631 Item recommendations

Abstract

Means and a computerized method for recommending items such as books and audio compact disks. For each item, a user profile includes ratings provided by users of the system. Unlike present recommendation systems, the user profiles do not include pre-computed similarity factors measuring similarity between users. Rather, when an advisee requests a recommendation, similarity measures are computed comparing the advisee to other users, and the similarity measures are associated with the other users. A subset of the users is selected, where the subset includes the users most similar to the advisee. A recommendation is made based on the ratings by the members of the selected subset.

Claims

1. A computerized method for generating a recommendation of an item to an advisee, comprising the steps of:

receiving a recommendation request comprising a selected item list from an advisee for a recommendation by a recommendation system;

in response to the recommendation request, computing, by a processor, a plurality of similarity factors based on;

at least one advisee profile from at least one rated item and determining which at least one user has already rated the item, wherein the advisee profile comprises a plurality of records, each record including a user identifier, an item identifier, and a rating value, such that each record is linked in a first and a second dimension; and

items from the selected item list that indicate similarity between the advisee and a plurality of users of the recommendation system who have previously provided ratings of items from the selected item list;

selecting, from the plurality of users of the recommendation system, neighboring users to the advisee, according to the similarity factors;

generating a recommendation of at least one item of the selected item list, according to the previously provided ratings of the at least one item by the neighboring users,

wherein the similarity factors of the advisee are not calculated with respect to all users for every individual recommendation request.

2. The method of claim 1, wherein all items which the step of computing depends are included in the selected item list.

3. The method of claim 2, wherein the recommendation of at least one item includes only items that are included in the selected item list.

4. The method of claim 1, wherein the step of selecting neighboring users excludes, any user whose similarity with the advisee is below a predetermined threshold.

5. The method of claim 1, wherein the step of computing and the step of selecting are executed substantially in parallel by inserting each newly computed similarity factor into a neighbor list in decreasing order of similarity and by limiting length of the neighbor list by excluding a user with lowest similarity to maintain the neighbor list within a predetermined length.

6. The method of claim 1, further including the step of caching identifiers of the neighboring users, associated similarity factors, and time stamps.

14. The method of claim 1, further comprising updating the selected neighboring users each time a new rating is entered or inferred during the recommendation request.

16. The method of claim 1, wherein the selected neighboring users are users who have a similarity factor which is better than a predetermined threshold value.

18. The method of claim 1, further comprising creating a user profile for each of the plurality of users, wherein the user profiles do not include any pre-computed similarity factors.

20. The method of claim 18, further comprising selecting each neighboring user when the neighboring user'"'"'s similarity factor is better than a predetermined threshold factor.

22. The method of claim 20, further comprising:

updating only the similarity factors between the neighboring users and the advisee for a new rating being entered for the at least one advisee profile; and

using an algorithm to determine a ranked matching list of the neighboring users.

24. The method of claim 22, further comprising:

for each recommendation request by the advisee, computing for use the similarity factors measuring the similarity between the advisee and the plurality of users; and

associating the similarity factors with corresponding users to determine, per recommendation request, the neighboring users; and

using the neighboring users determined per recommendation request to provide a basis for calculating the recommendation.

26. The method of claim 1, further comprising determining the similarity factors between the advisee and the neighboring users only for the items of the selected item list.

28. The method of claim 26, further comprising:

storing or updating all ratings in a single large sparse matrix in a computer'"'"'s main memory; and

calculating the similarity factors and the neighboring users for a concrete request for recommendation for the advisee being processed.

31. The method of claim 1, wherein each record further comprises:

a next-user link to another record of a next user in a sequence according to the user identifier; and

a next-item link to another record of a next item in a sequence according to the item identifier.

7. A computerized method for generating a recommendation of an item to an advisee, comprising the steps of:

receiving a recommendation request comprising a selected item list from an advisee for the recommendation by a recommendation system;

in response to the recommendation request, computing, by a processor, a plurality of similarity factors based on at least one advisee profile from at least one rated item and determining which at least one user has already rated the item;

selecting a first set of users from a group of users of the recommendation system based on the selected item list;

selecting neighboring users from the first set of users based on similarities between the advisee and each member of the first set of users; and

generating a recommendation of at least one item from the selected item list based on ratings provided by each neighboring user,

wherein the computing comprises determining the similarity factors between the advisee and the neighboring users only for the items of the selected item list.

8. The computerized method of claim 7, wherein the similarities are determined from an advisee profile and user profiles, and the advisee and user profiles are based on advise and user behavior including at least one of buying pattern, item ratings, bookmarked websites, website usage pattern, and user action relative to a particular item.

9. The computerized method of claim 8, wherein an advisee profile or a user profile is updated for a new piece of information being added thereto.

10. The computerized method of claim 7, further comprising assigning a confidence factor to each advisee profile and each user profile, wherein the confidence factor is based on the combined effect of selected pieces of information recorded in a user or advisee profile.

11. The computerized method of claim 7, further including determining similarities between the advisee and each member of the first set of users after receiving a selected item list from the advisee.

12. The computerized method of claim 7, wherein a member of the first set of users is selected as a neighboring user if the similarity between the advisee and the member of the first set of users is better than a predetermined threshold.

13. The computerized method of claim 7, further comprising assigning a weight to each neighboring user where the weight is greater for a neighboring user have greater similarity to the advisee and the weight is lower for a neighboring user having a lower similarity to the advisee.

15. The method of claim 7, further comprising updating the selected neighboring users each time a new rating is entered or inferred during the recommendation request.

17. The method of claim 7, wherein the selected neighboring users are users who have a similarity factor which is better than a predetermined threshold value.

19. The method of claim 7, further comprising creating a user profile for each of the users, wherein the user profiles do not include any pre-computed similarity factors.

21. The method of claim 19, further comprising selecting each neighboring user when the neighboring user'"'"'s similarity factor is better than a predetermined threshold factor.

23. The method of claim 21, further comprising:

updating only the similarity factors between the neighboring users and the advisee for a new rating being entered for the at least one advisee profile; and

using an algorithm to determine a ranked matching list of the neighboring users.

25. The method of claim 23, further comprising:

for each recommendation request by the advisee, computing for use the similarity factors measuring the similarity between the advisee and the plurality of users; and

associating the similarity factors with corresponding users to determine, per recommendation request, the neighboring users; and

using the neighboring users determined per recommendation request to provide a basis for calculating the recommendation.

27. The method of claim 7, wherein the similarity factors of the advisee are not calculated with respect to all users for every individual recommendation request.

29. The method of claim 27, further comprising:

storing or updating all ratings in a single large sparse matrix in a computer'"'"'s main memory; and

calculating the similarity factors and the neighboring users for a concrete request for recommendation for the advisee being processed.

30. A computerized method for recommending an item to an advisee utilizing a user profile comprising, for each of a multitude of items, at least a rating value and excluding any pre-computed similarity factor measuring similarity between users, wherein the method constructs a recommendation to the advisee and comprises:

calculating, by a processor, for use within a recommendation request only, a multitude of similarity factors measuring the similarity between the advisee and the multitude of other users;

associating the similarity factors with the other users;

determining a subset, from the multitude of users, of neighboring users with the similarity factors nearest in value to the advisee; and

recommending at least one item from a selected item list in the recommendation request from the advisee based on the similarity factors of the neighboring users and based on rating values of the items by the neighboring users.