Clustering multimedia search
View Patent ↗A method for clustering a set of web search results is disclosed. A first signature is compared based at least in part on an analysis of multimedia content associated with a first web search result with a second signature based at least in part on an analysis of multimedia content associated with a second web search result. The first web search result is clustered with the second web search result based at least in part on the comparison of the first signature with the second signature.
1. A method, comprising:
using a processor to precompute a first media content audio signature before a first web search, wherein the precomputation is based at least in part on an audio analysis of an audio portion of the video content, and wherein the first media content audio signature comprises a characteristic distinctive representation of the audio portion of the video content;
determining an entropy level of the first media content audio signature;
receiving a first web search result associated with the first web search; and
finding a video clip that sounds like the video content associated with the first web search result based at least in part on a comparison of the first media content audio signature using a distance metric and a determination that the entropy level of the first media content audio signature is high.
2. A method as recited in claim 1 , wherein finding comprises reducing a set of possible video clips found based at least in part on an analysis of textual metadata.
3. A method as recited in claim 1 , wherein video content includes any non-textual data or metadata.
4. A method as recited in claim 3 , wherein video content includes images, video and audio.
5. A method as recited in claim 4 , wherein the distance metric includes one or a weighted combination of: a Cartesian distance; a Manhattan distance; a Euclidean distance; and a byte difference.
6. A method as recited in claim 5 , wherein finding a video clip that sounds like the video content includes finding a video clip that sounds similar to the video content.
7. A method as recited in claim 6 , wherein finding includes highlighting web search results if the distance metric is below a similar-threshold but above an identical-threshold.
8. A method as recited in claim 7 , wherein each media content audio signature includes a hash value based at least in part on an audio feature.
9. A method as recited in claim 8 , wherein the audio feature is a percentage of laughter in an audio track.
10. A method as recited in claim 8 , further comprising labeling the video clip as a commentary of the video content in the event that the video clip has a more similar video signature comparison to the video content and a less similar audio signature comparison to the video content.
11. A method as recited in claim 8 , further comprising labeling the video clip as a remix of the video content in the event that the video clip has a less similar video signature comparison to the video content and a more similar audio signature comparison to the video content.
12. A method as recited in claim 8 , wherein the comparison is based at least in part on entropy.
13. A method as recited in claim 12 , further comprising finding video that looks like the video content associated with the first web search result.
14. A method as recited in claim 12 , further comprising finding audio that sounds like the video content associated with the first web search result.
15. A method as recited in claim 12 , further comprising sorting results for finding into bins.
16. A method as recited in claim 7 , further comprising receiving an indication that a “find similar” user interface button was activated.
17. A method as recited in claim 7 , further comprising adjusting the similar-threshold based at least in part on a “similarity slider” user interface setting.
18. A system, comprising:
a data store configured to store signatures of web search results; and
a processor coupled to the data store and configured to:
precompute a first media content audio signature before a first web search, wherein the precomputation is based at least in part on an audio analysis of an audio portion of the video content, and wherein the first media content audio signature comprises a characteristic distinctive representation of the audio portion of the video content;
determine an entropy level of the first media content audio signature;
receive a first web search result associated with the first web search; and
find a video clip that sounds like the video content associated with the first web search result based at least in part on a comparison of the first media content audio signature using a distance metric and a determination that the entropy level of the first media content audio signature is high.
19. A system as recited in claim 18 , wherein the processor is further configured to perform the first web search and the processor receives the first web search result from the processor itself.
20. A computer program product, the computer program product being embodied in a non-transitory computer readable storage medium and comprising computer instructions for:
precomputing a first media content audio signature before a first web search, wherein the precomputation is based at least in part on an audio analysis of an audio portion of the video content, and wherein the first media content audio signature comprises a characteristic distinctive representation of the audio portion of the video content;
determining an entropy level of the first media content audio signature;
receiving a first web search result associated with the first web search; and
finding a video clip that sounds like the video content associated with the first web search result based at least in part on a comparison of the first media content audio signature using a distance metric and a determination that the entropy level of the first media content audio signature is high.