Collaborative filtering-based recommendations

• US 8,468,155 B2
• Filed: 06/19/2007
• Issued: 06/18/2013
• Est. Priority Date: 06/22/2006
• Status: Active Grant

• Current Assignee: Infosys Limited
• Original Assignee: Infosys Limited
• Inventors: Sureka, Ashish
• Primary Examiner(s): Hasan, Syed
• Assistant Examiner(s): MUELLER, KURT A
• Application Number:
  US11/765,377
• Publication Number:
  US 20080010258A1
• Time in Patent Office: 2,191 Days
• Field of Search: 707/732
• US Class Current: 707/732
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation

Abstract

Various techniques can be used to implement a collaborative filtering-based recommendation engine. For example, different similarity measures can be used for different users. Different similarity measures can be used for a particular user across time. A superior similarity measure can be found for a user. User-defined similarity measures can be supported.

Claims

1. A computer-implemented method comprising:

choosing item recommendations for a plurality of users comprising at least a first user and a second user; and

outputting the item recommendations;

wherein choosing item recommendations for a plurality of users comprises;

for the first user, identifying, out of a plurality of possible user similarity measures, a first user similarity measure as most accurately identifying users who provided rating information similar to rating information provided by the first user;

for the first user, using the first user similarity measure to identify, from the plurality of users, a first set of similar users who provided rating information similar to the rating information provided by the first user;

for the first user, using the rating information provided by the first set of similar users to select a first item recommendation;

for the second user, identifying, out of the plurality of possible user similarity measures, a second user similarity measure as most accurately identifying users who provided rating information similar to rating information provided by the second user, wherein the second user similarity measure is different than the first user similarity measure;

for the second user, using the second user similarity measure to identify, from the plurality of users, a second set of similar users who provided rating information similar to the rating information provided by the second user; and

for the second user, using the rating information provided by the second set of similar users to select a second item recommendation.

2. The method of claim 1 further comprising:

selecting a similarity measure for a particular user via a stored mapping between users and user similarity measures.

3. The method of claim 1 further comprising:

choosing item recommendations for a same user via different user similarity measures when applying collaborative filtering for the same user at different times while choosing item recommendations for an other user via a same user similarity measure.

4. The method of claim 1, wherein identifying a first user similarity measure comprises:

testing a plurality of similarity measures;

selecting a best performing similarity measure; and

storing an association associating the best performing similarity measure with the first user.

5. The method of claim 1 further comprising:

receiving from a user an indication regarding a preferred similarity measure; and

employing the preferred similarity measure when choosing recommendations for the user.

6. The method of claim 1 further comprising:

receiving from a user a user-defined similarity measure; and

employing the user-defined similarity measure when choosing item recommendations for the user.

7. The method of claim 6 wherein the user-defined similarity measure comprises a plurality of user-defined weight values provided.

8. The method of claim 1 wherein the choosing is performed in a single recommender system.

9. The method of claim 1 wherein the choosing is performed in a single recommendation engine.

11. The computer-implemented method of claim 1 wherein:

the first user similarity measure and the second user similarity measure are chosen from the group consisting of;

Pearson Correlation Coefficient, Jacobian distance, Euclidean distance, Manhattan distance, and a custom similarity measure.

12. The computer-implemented method of claim 1, further comprising:

for the first user, using the first user similarity measure to identify, from the first set of similar users, a most similar user; and

for the first user, weighting rating information provided by the most similar user more heavily than rating information provided by other users in the first set of similar users when selecting a first item recommendation.

13. The computer-implemented method of claim 1 wherein identifying a first user similarity measure comprises:

dividing the rating information provided by the first user into a training data set and a testing data set;

using the plurality of possible user similarity measures to identify respective test sets of similar users based on the training data set;

using the test sets of similar users to obtain, for the possible user similarity measures, respective sets of predicted rating information; and

comparing the sets of predicted rating information to the rating information of the testing data set.

10. One or more computer-readable non-transitory storage media having encoded thereon computer-executable instructions for performing a computer-implemented method comprising:

choosing item recommendations for different users via different user similarity metrics when applying collaborative filtering for the different users, wherein a user similarity metric is designated for one user independently of other users; and

outputting the item recommendations;

wherein choosing item recommendations for different users comprises (a)-(e);

(a) for a first user, identifying, out of a plurality of possible user similarity metrics, a first user similarity metric measuring similarity among users, wherein the identifying is based at least on performance of the first user similarity metric with respect to the first user;

(b) for the first user, using the first user similarity metric to identify a most similar user;

(c) choosing a first recommendation for the first user via collaborative filtering using the first user similarity metric measuring similarity among users, wherein a weight assigned to an influence of the most similar user is greater than a weight assigned to an influence of any other user when choosing a first recommendation for the first user via collaborative filtering;

(d) for a second user, identifying, out of the possible user similarity metrics, a second user similarity metric measuring similarity among users, wherein the identifying is based at least on performance of the second user similarity metric with respect to the second user, wherein the second user similarity metric is different from the first; and

(e) choosing a second recommendation for the second user via the second user similarity metric measuring similarly among users;

wherein the first user similarity metric is implemented as a first similarity function that provides a value indicating distance between users; and

the second user similarity metric is implemented as a second similarity function different from the first similarity function.

14. A computer-implemented method comprising:

choosing item recommendations for different users via different user similarity metrics when applying collaborative filtering for the different users, wherein a user similarity metric is designated for one user independently of other users; and

outputting the item recommendations;

wherein choosing item recommendations for different users comprises (a)-(e);

(a) for a first user, identifying, out of a plurality of possible user similarity metrics, a first user similarity metric measuring similarity among users, wherein the identifying is based at least on performance of the first user similarity metric with respect to the first user;

(b) for the first user, using the first user similarity metric to identify a most similar user;

(c) choosing a first recommendation for the first user via collaborative filtering using the first user similarity metric measuring similarity among users, wherein a weight assigned to an influence of the most similar user is greater than a weight assigned to an influence of any other user when choosing a first recommendation for the first user via collaborative filtering;

(d) for a second user, identifying, out of the possible user similarity metrics, a second user similarity metric measuring similarity among users, wherein the identifying is based at least on performance of the second user similarity metric with respect to the second user, wherein the second user similarity metric is different from the first; and

(e) choosing a second recommendation for the second user via the second user similarity metric measuring similarly among users;

wherein the first user similarity metric is implemented as a first similarity function that provides a value indicating distance between users; and

the second user similarity metric is implemented as a second similarity function different from the first similarity function.

15. One or more computer-readable non-transitory storage media having encoded thereon computer-executable instructions for performing a computer-implemented method comprising:

choosing item recommendations for a plurality of users comprising at least a first user and a second user; and

outputting the item recommendations;

wherein choosing item recommendations for a plurality of users comprises;

for the first user, identifying, out of a plurality of possible user similarity measures, a first user similarity measure as most accurately identifying users who provided rating information similar to rating information provided by the first user;

for the first user, using the first user similarity measure to identify, from the plurality of users, a first set of similar users who provided rating information similar to the rating information provided by the first user;

for the first user, using the rating information provided by the first set of similar users to select a first item recommendation;

for the second user, identifying, out of the plurality of possible user similarity measures, a second user similarity measure as most accurately identifying users who provided rating information similar to rating information provided by the second user, wherein the second user similarity measure is different than the first user similarity measure;

for the second user, using the second user similarity measure to identify, from the plurality of users, a second set of similar users who provided rating information similar to the rating information provided by the second user; and

for the second user, using the rating information provided by the second set of similar users to select a second item recommendation.