Method and apparatus for efficiently recommending items using automated collaborative filtering and feature-guided automated collaborative filtering

• US 6,092,049 A
• Filed: 03/14/1997
• Issued: 07/18/2000
• Est. Priority Date: 06/30/1995
• Status: Expired due to Term

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: McNulty, John Edward, Chislenko, Alexander, Tiu, David D., Lashkari, Yezdezard, Metral, Max E.
• Primary Examiner(s): Stamber, Eric W.
• Assistant Examiner(s): CAUDLE, PENNY L
• Application Number:
  US08/818,533
• Time in Patent Office: 1,222 Days
• Field of Search: 705/27, 705/26, 705/8, 705/10, 705/5, 707/5
• US Class Current: 705/7.29
• 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
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G06F 16/9536 Search customisation based on social or collaborative filtering
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/0201 Market modelling; Market analysis; Collecting market data
• CPC Class Codes: G06Q 30/0203 Market surveys; Market polls
• CPC Class Codes: G06Q 30/06 Buying, selling or leasing transactions
• CPC Class Codes: H04N 21/252 Processing of multiple end-users' preferences to derive collaborative data

Abstract

A method for recommending items to users using automated collaborative filtering stores profiles of users relating ratings to items in memory. Profiles of items may also be stored in memory, the item profiles associating users with the rating given to the item by that user or inferred for the user by the system The user profiles include additional information relating to the user or associated with the rating given to an item by the user. Item profiles are retrieved to determine which users have rated a particular item. Profiles of those users are accessed and the ratings are used to calculate similarity factors with respect to other users. The similarity factors, sometimes in connection with confidence factors, are used to select a set of neighboring users. The neighboring users are weighted based on their respective similarity factors, and a rating for an item contained in the domain is predicted. In one embodiment, items in the domain have features. In this embodiment, the values for features can be clustered, and the similarity factors incorporate assigned feature weights and feature value cluster weights. In some embodiments, item concepts are used to enhance recommendation accuracy.

Claims

1. In a machine which recommends at least one item to a first user based, in part, on similarity factors between the first user and at least a second user, a method for operating the machine to determine a similarity factor between the first user and the second user, the method comprising the steps of:

(a) retrieving from memory, using the machine an item profile of each of a plurality of items rated by the first user so as to define a plurality of retrieved item profiles, each of said retrieved item profiles includes a plurality of first values, each of the plurality of first values representing a rating given to an associated one of the items by each of a plurality of users;

(b) determining from the retrieved item profiles, using the machine, whether the second user has previously rated any of the items;

(c) retrieving from memory, using the machine, a corresponding user profile of the first user so as to define a retrieved first user profile;

(d) retrieving, from memory, a corresponding user profile of the second user provided the second user has previously rated any of the items, so as to define a retrieved second user profile, wherein each of the retrieved first and second user profiles includes a plurality of second values, each of the plurality of second values representing a rating given to one of a plurality of items by a corresponding one of the users; and

(e) calculating, using the machine, a similarity factor between the first user and the second user responsive to the retrieved profiles of the first and second users, wherein the first and second user profiles are distinct from the plurality of item profiles.

2. The method of claim 1 wherein the step of calculating a similarity factor includes sub-steps of:

subtracting, using the machine, for each item rated by both users, the rating given to the item by the second user from the rating given to the item by the first user;

squaring, using the machine, each rating difference; and

dividing, using the machine, the sum of the squared differences by the number of items rated by both users.

3. The method of claim 1

wherein, in step (a), each item profile of each item rated by the first user is retrieved from a first memory and wherein, in step (c), the user profile of the second user is retrieved from a second memory.

4. The method of claim 1 further comprising the step of retrieving, using the machine, a user profile of the first user from a second memory before step (a).

11. The method of claim 1 wherein the step of calculating a similarity factor is performed only if the second user has previously rated the items.

12. The method of claim 3 where in the first memory is located on a first server and the second memory is located on a second server.

5. In computer-implemented apparatus, the apparatus having a processor and a memory, the memory having computer executable instructions stored therein, a method for recommending an item to a particular one of a plurality of users, the item not yet rated by the user, the method comprising steps, performed through the executable instructions, of:

(a) storing a user profile, in the memory, for each of a plurality of users so as to define a plurality of user profiles, wherein each of the user profiles includes a plurality of first values, each of the plurality of first values representing a rating given to one of a plurality of items by a corresponding one of the users;

(b) storing an item profile, in the memory, for each of the plurality of items so as to define a plurality of item profiles, wherein each of the item profiles includes a plurality of second values, each of the plurality of second values representing a rating given to an associated one of the items by each of the users;

(c) retrieving, from the memory, the item profile of each item rated by the particular one user so as to define a plurality of retrieved item profiles;

(d) determining from the retrieved item profiles, a set of users that have previously rated the items;

(e) retrieving, from the memory, the user profile for each of the set of users so as to define a plurality of retrieved user profiles, the plurality of user profiles being distinct from the plurality of item profiles;

(f) calculating, in response to the retrieved user profiles, a similarity factor between the particular one user and each of the set of users;

(g) selecting, in response to the similarity factors, for the particular one user, a plurality of neighboring users from the set of users;

(h) assigning a weight to each of the neighboring users; and

(i) recommending at least one of the plurality of items to the particular one user based on the weights assigned to the neighboring users and ratings given to the item by the neighboring users.

6. The method of claim 5 wherein the user profile storing step comprises the step of storing additional information, in the user profiles, other than ratings for the items.

7. The method of claim 5 wherein the item profile storing step comprises the step of storing additional information, in the item profiles, other than ratings for the items.

8. The method of claim 5 further comprising the step of storing the item profiles in a first memory and storing the user profiles in a second memory.

9. The method of claim 5 wherein the item profile retrieving step comprises the steps of:

receiving a rating from one of the plurality of users for one of the plurality of items so as to define a received rating;

updating the user profile of the one of the plurality of users with the received rating;

updating the item profile of the rated item with the received rating;

calculating, for the particular one user, a plurality of similarity factors, each of the plurality of similarity factors representing similarity between the particular one user and another one of the users.

10. In computer-implemented apparatus, the apparatus having a processor and a memory, the memory having computer executable instructions stored therein, a method for recommending an item to a particular one of a plurality of users, the item not yet rated by the user, the method comprising steps, performed through the executable instructions, of:

(a) generating, for the particular one user, a concept mask representing areas of interest of the one user;

(b) storing, in the memory, a user profile in a memory for each of a plurality of users, wherein the user profile of the user includes a plurality of values, each of the plurality of values representing a rating given to one of a plurality of items by the user;

(c) calculating, for each of the plurality of users, a plurality of similarity factor vectors representing the similarity between each user and another one of the plurality of users for at least one concept specified within the concept mask;

(d) selecting, for each of the plurality of users, a plurality of neighboring users based on the similarity factor vectors;

(e) assigning a weight to each of the neighboring users so as to define a plurality of weights assigned to the neighboring users; and

(f) recommending at least one of the plurality of items to the particular one user based on the weights assigned to the neighboring users and the ratings given to the non-rated item by the neighboring users.

13. The method of claim 10 wherein the concept mask includes a number of concepts, wherein each of the concepts may be associated with at least one item rating, and wherein each item rating may be associated with at least one concept mask.

14. The method of claim 10 wherein the concept mask includes every concept with which rated items are associated.

15. The method of claim 10 wherein the concept mask includes concepts which have a highest value of all concepts.

16. The method of claim 10 wherein the concept mask includes concepts which have a value which exceeds a predetermined threshold.

17. The method of claim 10 wherein the concept mask includes a predetermined limited number of concepts.