METHOD AND A SYSTEM FOR AUTOMATIC EVALUATION OF DIGITAL FILES

• US 20080228744A1
• Filed: 03/12/2007
• Published: 09/18/2008
• Est. Priority Date: 03/12/2007
• Status: Active Grant

• Current Assignee: Hitlab ULC
• Original Assignee: Hitlab ULC
• Inventors: DESBIENS, Jocelyn
• Granted Patent: US 7,873,634 B2
• Time in Patent Office: Days
• US Class Current: 707/5
• CPC Class Codes: G06F 16/683 using metadata automatically derived from the content
• CPC Class Codes: G06F 16/686 using information manually generated, e.g. tags, keywords, comments, title or artist information, time, location or usage information, user ratings

Abstract

There is provided a method for automatic evaluation of target files, comprising the steps of building a database of reference files; for each target file, forming a training set comprising files from the database of reference files and building a test set from features of the target file; dynamically generating a learning model from the training set; and applying the learning model to the test set, whereby a value corresponding to the target file is predicted.

Claims

1. A method for automatic evaluation of target files, comprising the steps of:

building a database of reference files;

for each target file, forming a training set comprising files from the database of reference files and building a test set from features of the target file;

dynamically generating a learning model from the training set; and

applying the learning model to the test set;

whereby a value corresponding to the target file is predicted.

2. The method of claim 1, further comprising storing the predicted value in a result database.

3. The method of claim 1, wherein said step of building a database of reference files comprises collecting files identified as references according to a predefined scheme, under a digital format;

obtaining feature vectors of each of the collected files; and

storing the feature vectors in a database of reference files.

4. The method of claim 3, wherein said step of building a database of reference files further comprises storing a rate, defined according to the predefined scheme, of each of the reference files in a score database.

5. The method of claim 3, wherein said step of obtaining feature vectors of each of the collected files comprises extracting, from the collected files, a number of features to yield reference feature vectors.

6. The method of claim 3, wherein said step of storing the feature vectors in a database of reference files comprises storing the feature vectors along with relevant information.

7. The method of claim 1, wherein said step of forming a training set comprises finding closest reference files in the database of reference files, versus which the target file is to be assessed.

8. The method of claim 1, wherein said step of forming a training set comprises extracting a feature vector of the target file;

and finding n closest feature vectors of the target file'"'"'s feature vector in the database of reference files.

9. The method of claim 1, wherein said finding n closest feature vectors comprises using one of:

i) Euclidian distance, ii) cosine distance and iii) Jensen-Shannon distribution similarity.

10. The method of claim 1, wherein said step of forming a training set comprising files from the database of reference files and building a test set from features of the target file further comprises reducing the dimensionality of the training set and reducing the dimensionality of the test set.

11. The method of claim 10, wherein said steps of reducing the dimensionality are done by using one of:

i) Principal Component Analysis (PCA) and ii) Singular Value Decomposition (SVD).

12. The method of claim 10, wherein said steps of reducing the dimensionality are done by a non-linear regression technique.

13. The method of claim 10, wherein said steps of reducing the dimensionality are done by one of:

Neural Networks, Support Vector Machines, Generalized Additive Model, Classification and Regression Tree, Multivariate Adaptative Regression Splines, Hierarchical Mixture of Experts and Supervised Principal Component Analysis.

14. The method of claim 7, wherein said step of dynamically generating a learning model comprises using closest reference files in the database of reference files.

15. The method of claim 8, wherein said step of dynamically generating a learning model comprises using the n closest feature vectors of the target file'"'"'s feature vector in the database of reference files.

16. The method of claim 7, wherein said step of dynamically generating a learning model comprises applying a prediction model to the closest reference files in the database of reference files.

17. The method of claim 8, wherein said step of dynamically generating a learning model comprises applying a prediction model to the n closest feature vectors of the target file'"'"'s feature vector in the database of reference files.

18. The method of claim 7, wherein said step of dynamically generating a learning model comprises applying a prediction model comprises applying a Support Vector Model.

19. The method of claim 8, wherein said step of dynamically generating a learning model comprises applying a Support Vector Model to the n closest feature vectors of the target file'"'"'s feature vector in the database of reference files.

20. The method of claim 1, wherein said step of dynamically generating a learning model comprises using MCubix™

by Diagnos.

21. The method of claim 1, further comprising discarding the learning model after prediction for the target file.

22. The method of claim 1, wherein said step of building a training set comprises rebuilding the training set as new files appear in the database of reference files.

23. The method of claim 7, wherein said step of forming a training set comprises finding new closest reference files in the database of reference files as new reference files appear in the database of reference files

24. The method of claim 8, wherein said step of forming a training set comprises updating the closest neighbours as new reference files appear in the database of reference files.

25. The method of claim 1, wherein said step of generating a learning model comprises automatically generating a learning model based on a dynamic neighborhood represented by the training set.

26. The method of claim 1, wherein the target files are song files, the reference files are hits, and the target files are assessed according to the hits.

27. A method for automatic evaluation of songs, comprising the step of:

building a database of hit songs;

for each song to be evaluated, forming a training set comprising songs from the database of hit songs and building a test set from features of the song to be evaluated;

dynamically generating a learning model from the training set; and

applying the learning model to the test set;

whereby a score corresponding to the song to be evaluated is predicted.

28. The method of claim 27, wherein said step of building a database of hit songs comprises collecting songs identified as hits according to a predefined scheme of rating.

29. The method of claim 27, wherein said step of building a database of hit songs comprises extracting, from the hint songs, a number of features representing the hint songs.

30. The method of claim 29, wherein said step of building a database of hit songs comprises extracting, from the hint songs, a number of features representing the hint songs, including at least one of timbre, rhythm, frequency and melody of the hit songs.

31. The method of claim 27, wherein said step of forming a training set and a test set comprise extracting a number of features, from the song to be evaluated, representing the song to be evaluated.

32. The method of claim 27, wherein said step of dynamically generating a learning model from the training set comprises using MCubix™

of Diagnos.