METHOD OF ORGANISING CONTENT ITEMS

• US 20100138427A1
• Filed: 04/24/2008
• Published: 06/03/2010
• Est. Priority Date: 05/01/2007
• Status: Active Grant

• Current Assignee: Koninklijke Philips Electronics N.V. (Koninklijke Philips N.V.)
• Original Assignee: Koninklijke Philips Electronics N.V. (Koninklijke Philips N.V.)
• Inventors: Van De Par, Steven Leonardus Josephus Dimphina Elisabeth, McKinney, Martin Franciscus, Skowronek, Janto
• Granted Patent: US 8,140,542 B2
• Time in Patent Office: Days
• US Class Current: 707/748
• CPC Class Codes: G06F 16/634 Query by example, e.g. query by humming
• CPC Class Codes: G06F 16/639 using playlists
• CPC Class Codes: G06F 16/68 Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
• CPC Class Codes: G06F 16/683 using metadata automatically derived from the content

Abstract

A method of organising content items includes obtaining information (19-21) associating each of a first and at least one further label with respective sets of first content items. At least one first content item is associated with the first label in response to user input. A second content item is obtained. For each of the first and further labels, a similarity measure is calculated between a feature vector characterising the second content item and each of the feature vectors characterising first content items associated with the label. The first label is associated with the second content item upon determining that the second content item is similar to the first content items associated with the first label, according to calculated values of the similarity measure.

Claims

1. Method of organising content items, including:

obtaining information (19-21) associating each of a first and at least one further label with respective sets of a number of first content items,

wherein at least one first content item is associated with the first label in response to user input;

obtaining a second content item;

for each of the first and further labels, calculating a similarity measure between a feature vector characterising the second content item and each of the feature vectors characterising first content items associated with the label; and

associating the first label with the second content item upon determining that the second content item is similar to the first content items associated with the first label, according to calculated values of the similarity measure.

2. Method according to claim 1, wherein

a user is allowed to define the first label by means of the user input.

3. Method according to claim 1, wherein

the first label is associated with the second content item by storing information linking the first label to the second content item.

4. Method according to claim 1, further including

allowing a user to associate at least one first content item with the first label by means of the user input.

5. Method according to claim 1, wherein the first label is associated with the second content item upon determining that the second content item is more similar to the first content items associated with the first label than to the first content items associated with the further labels.

6. Method according to claim 1, wherein at least one co-ordinate of each feature vector is obtainable by performing a pre-defined signal analysis on a signal comprised in the content item that is characterised by that feature vector, to determine a parameter value representative of a characteristic of the signal that is perceptible when rendered on a reproduction device.

7. Method according to claim 1, further including associating with the second content item data representative of at least one value based on the value of the similarity measure between the feature vector characterising the second content item and the feature vectors characterising content items associated with the first label.

8. Method according to claim 1, wherein the first label is associated with the second content item only upon determining that at least one value based on the value of the similarity measure between the feature vector characterising the second content item and the feature vectors characterising content items associated with the first label lies within a pre-determined range.

9. Method according to claim 1, wherein the similarity measure is calculated by:

calculating a weighted distance between feature vectors according to the weighted distance measure and

applying a function based on a probability distribution to the calculated weighted distance.

10. Method according to claim 9, wherein the function based on a probability distribution is parameterised across the sets of first content items associated with the first and further labels.

11. Method according to claim 9, wherein a similarity measure between a feature vector characterising the second content item and any feature vector characterising a content item associated with a label is normalised by a factor dependent on the number M_c of first content items comprised in the set associated with the label.

12. Method according to claim 11, wherein the similarity measure is normalised by a factor inversely proportional to the number M_c of first content items comprised in the set associated with the label, exponentiated with an exponent having a value in the range of zero to one.

13. Method according to claim 9, wherein the function based on a probability distribution includes a variable parameter specifying a width of the probability distribution, wherein the variable parameter is assigned a value depending on the total number of associations between a first content item and the first or a further label.

14. Method according to claim 13, wherein, for a particular one of the first and further labels, the variable parameter is assigned a value that is a function of values of a number of the smallest distances according to the weighted distance measure between pairs of feature vectors in the set of first content items associated with that label.

15. Method according to claim 13, wherein the variable parameter is a function of an average distance according to the weighted distance measure between arbitrary pairs of all feature vectors characterising the first content items associated with the first or a further label.

16. Method according to claim 1, including calculating the similarity measure using a weighted distance measure differentiating between coordinates of the feature vectors and based on a feature weighting matrix obtainable by

obtaining a training set of feature vectors, each characterising one of a plurality of content items,

each of the feature vectors being comprised in at least one of a plurality of sub-sets, and

populating the weighting matrix with values optimised such that pairs of feature vectors comprised in a common sub-set will have a large weighted distance relative to pairs of feature vectors comprised in disjoint sub-sets.

19. Computer programme including a set of instructions capable, when incorporated in a machine-readable medium, of causing a system (1;

39) having information processing capabilities to perform a method according to claim 1.

17. Device for organising content items, including electronic circuitry (2-11;

40-50), the electronic circuitry being operative to;

associate at least one first content item with a first label in response to user input,

obtain information associating each of at least one further label with respective sets of a number of first content items;

obtain a second content item;

for each of the first and further labels, calculate a similarity measure between a feature vector characterising the second content item and each of the feature vectors characterising first content items associated with the label; and

to associate the first label with the second content item upon determining that the second content item is similar to the first content items associated with the first label, according to calculated values of the similarity measure.

18. Device according to claim 17, configured to execute a method of organising content items, including:

obtaining information (19-21) associating each of a first and at least one further label with respective sets of a number of first content items,

wherein at least one first content item is associated with the first label in response to user input;

obtaining a second content item;

for each of the first and further labels, calculating a similarity measure between a feature vector characterising the second content item and each of the feature vectors characterising first content items associated with the label; and

associating the first label with the second content item upon determining that the second content item is similar to the first content items associated with the first label, according to calculated values of the similarity measure.