SYSTEM AND METHOD FOR IDENTIFYING SIMILAR MEDIA OBJECTS

• US 20090049091A1
• Filed: 08/13/2007
• Published: 02/19/2009
• Est. Priority Date: 08/13/2007
• Status: Active Grant

• Current Assignee: R2 Solutions LLC (Acacia Research Corporation)
• Original Assignee: Yahoo! Inc. (Apollo Global Management, Inc.)
• Inventors: SLANEY, Malcolm, White, William
• Granted Patent: US 7,849,092 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/433 using audio data
• CPC Class Codes: G06F 16/434 using image data, e.g. images, photos, pictures taken by a user
• CPC Class Codes: G06F 16/48 Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
• CPC Class Codes: Y10S 707/913 Multimedia

Abstract

The systems and methods described create a mathematical representation of each of the media objects for which user ratings are known. The mathematical representations take into account the subjective rating value assigned by a user to the respective media object and the user that assigned the rating value. The media object with the mathematical representation closest to that of the seed media object is then selected as the most similar media object to the seed media object. In an embodiment, the mathematical representation is a vector representation in which each user is a different dimension and each user'"'"'s rating value is the magnitude of the vector in that dimension. Similarity between two songs is determined by identifying the closest vectors to that of the seed song. Closeness may be determined by subtracting or by calculating the dot product of each of the vectors with that of the seed media object.

Claims

1. A method for identifying similar media objects comprising:

accessing a datastore of user ratings of media objects including a seed media object, a first media object and a second media object, each user rating including a rating value associated with a user identifier and a media object;

identifying a seed set of user ratings associated with the seed media object, a first set of user ratings associated with the first media object and a second set of user ratings associated with the second media object;

for each user identifier appearing in user ratings in both the seed set and the first set, generating a first object user similarity value based on the rating values of the user ratings of the seed media object and the first media object;

calculating a first media object total similarity value based on the generated first object user similarity values;

for each user identifier appearing in user ratings in both the seed set and the second set, generating a second object user similarity value based on the rating values of the user ratings of the seed media object and the second media object;

calculating a second media object total similarity value based on the generated second object user similarity values;

comparing the first media object total similarity value with the second media object total similarity value; and

identifying the first media object as more similar to the seed media object than the second media object based on results of the comparing operation.

2. The method of claim 1 further comprising:

biasing the rating value cf each user rating by a predetermined factor before calculating the first object user similarity value and second object user similarity value.

3. The method of claim 1 further comprising:

maintaining a library of media objects including a seed media object, a first media object and a second media object;

receiving the user ratings from users accessing the library of media objects; and

storing the user ratings in the datastore.

4. The method of claim 1 further comprising:

receiving a request for one or more media objects similar to the seed media object; and

transmitting a response identifying the first media object as a media object that is similar to the seed media object.

5. The method of claim 1 wherein identifying the seed set, the first set and the second set further comprises:

identifying as the seed set all user ratings associated with the seed media object contained in the datastore,

identifying as the first set all user ratings associated with the first media object contained in the datastore; and

identifying as the second set all user ratings associated with the second media object contained in the datastore.

6. The method of claim 1 wherein identifying the seed set, the first set and the second set further comprises:

identifying a subset of user identifiers as super-users;

identifying as the seed set all user ratings associated with the seed media object and a super-user contained in the datastore,

identifying as the first set all user ratings associated with the first media object and a super-user contained in the datastore; and

identifying as the second set all user ratings associated with the second media object and a super-user contained in the data store.

7. The method of claim 1 wherein generating the first and second object user similarity values further comprises:

for each user identifier having a user rating for the seed set and the first set, multiplying the rating value of the seed media object with the rating value of the first media object; and

for each user identifier having a user rating for the seed set and the second set, multiplying the rating value of the seed media object with the rating value of the second media object.

8. The method of claim 7 wherein calculating the first and second object total similarity values further comprises:

adding each of the generated first object user similarity values to obtain the first object total similarity value; and

adding each of the generated second object user similarity values to obtain the second object total similarity value.

9. The method of claim 1 wherein generating the first and second object user similarity values further comprises:

for each user identifier having a user rating for the seed set and the first set, subtracting the rating value of the first media object with the rating value of the seed media object; and

for each user identifier having a user rating for the seed set and the second set, subtracting the rating value of the second media object with the rating value of the seed media object.

10. The method of claim 9 wherein calculating the first and second object total similarity values further comprises:

squaring each of the generated first object user similarity values to obtain a set of squared first object user similarity values;

adding each of the squared first object user similarity values to obtain the first object total similarity value;

squaring each of the generated second object user similarity values to obtain a set of squared second object user similarity values; and

adding each of the squared second object user similarity values to obtain the second object total similarity value.

11. A system for identifying similar media objects comprising:

a communications module that receives a request for a media object similar to an identified seed media object and that transmits a response to the request;

a datastore that contains a plurality of user ratings of media objects, each user rating having a user identifier, a rating value, and a media object identifier, the user ratings of media objects including user ratings of the seed object and user ratings of a plurality of first objects; and

a comparison engine that analyzes the plurality of user ratings and identifies, based on the plurality of user ratings, a first media object as similar to the seed media object.

12. The system of claim 11 further comprising:

a media object library that contains the plurality of first objects.

13. The system of claim 11 wherein the comparison engine further compares each of the plurality of first media objects to the seed media object to generate a similarity value for each of the plurality of first media objects, wherein each similarity value for each respective first media objects is calculated by comparing each user rating of the seed media object with a corresponding user rating having the same user identifier for the respective first media object.

14. The system of claim 13 wherein the comparison engine further characterizes each first media object as a vector having a plurality of dimensions and magnitudes.

15. The system of claim 14 wherein each vector comprises a plurality of magnitudes each associated with a different dimension in multi-dimensional space, wherein each dimension corresponds to a different user identifier and each magnitude corresponds to the rating value assigned to the vector'"'"'s respective media object by a user corresponding to the user identifier.

16. The system of claim 15 wherein the comparison engine further comprises:

a dot product module that calculates, for each of the first media objects, a dot product of the vectors of the seed media object with the first media object.

17. The system of claim 16 wherein the comparison engine further comprises:

a similarity ranking module that calculates the similarity value for each first media object based on the dot product of the vector of the seed media object with the vector of the first media object and identifies one of the first media objects as most similar to the seed media object based on the relative similarity values of the first media objects.

18. The system of claim 12 further comprising:

a media module that transmits media objects to computing devices associated with user identifiers.

19. The system of claim 18 wherein the media module further receives and stores in the datastore user ratings from the computing devices.

20. A computer readable medium storing computer executable instructions which when executed cause a computer to perform a method comprising:

receiving a request for a media object that is similar to an identified seed media object;

creating a mathematical .representation of the seed media object based on subjective user ratings of the seed media object;

comparing the mathematical representation of the seed media object to a plurality of mathematical representations of different first media objects based on subjective user ratings of the first media objects; and

generating a response to the request, the response identifying at least one of the first media objects as being similar to the seed media object based on results of the comparing operation.

21. The computer readable medium of claim 20, wherein the method further comprises:

creating the plurality of mathematical representations of the different first media objects,

and wherein creating a mathematical representation comprises;

retrieving the subjective user ratings associated with the seed media object and the first media objects; and

representing each of the seed media object and first media objects as a vector in multi-dimensional space based on its associated subjective user rating.

23. The computer readable medium of claim 21, wherein the method further comprises:

defining the vector as a plurality of magnitudes each associated with a different dimension in the multi-dimensional space, wherein each dimension corresponds to a different user identifier and each magnitude corresponds to a subjective rating value assigned to the associated media object by a user corresponding to the user identifier.

24. The computer readable medium of claim 21, wherein the method further comprises:

calculating a dot product of the vector of the seed media object with each of the vectors of the first media objects.

25. The computer readable medium of claim 24, wherein the method further comprises:

ranking each of the first media objects as similar to the seed media object based on relative values of the dot product of the vector of the seed media object with each of the vectors of the first media objects.

28. The computer readable medium of claim 20, wherein the method further comprises:

comparing the mathematical representation of the most similar one of the first media objects to the plurality of mathematical representations of different second media objects based on subjective user ratings of the first media objects;

selecting a most similar one of the second media objects based on results of the comparing operation; and

transmitting the selected most similar one of the second media objects to the requestor.

22. (canceled)

26-27. -27. (canceled)