System and method for cascading image clustering using distribution over auto-generated labels
Embodiments of the present invention provide a system that can be used to classify a feedback image in a user review into a semantically meaningful class. During operation, the system analyzes the captions of feedback images in a set of user reviews and determines a set of training labels from the captions. The system then trains an image classifier with the set of training labels and the feedback images. Subsequently, the system generates a signature for a respective feedback image in a new set of user reviews using the image classifier. The signature indicates a likelihood of the image matching a respective label in the set of training labels. Based on the signature, the system can allocate the image to an image cluster.
1. A computer-implemented method for facilitating cascading image clustering, the method comprising:
selecting an initial image from a set of reviews and further selecting a set of neighboring images to form an initial seed cluster;
repeating the selecting a number of times with a different image and a different set of neighboring images from the set of reviews, wherein the selecting and the repeating forms a plurality of seed clusters;
determining whether an average distance between two clusters from the plurality of seed clusters is below a threshold value; and
in response to the average distance between the two seed clusters being below the threshold value, merging the two seed clusters.
2. The computer-implemented method of claim 1 further comprising:
allocating a semantic label to the two merged seed clusters.
3. The computer-implemented method of claim 1 , wherein the initial image is randomly selected.
4. The computer-implemented method of claim 1 further comprising:
repeating the determining and the merging until all mergeable seed clusters are merged.
5. The computer-implemented method of claim 4 further comprising:
allocating a semantic label to each seed cluster after all mergeable seed clusters have been merged.
6. The computer-implemented method of claim 5 further comprising:
receiving a new image;
classifying the new image in one semantic label of the semantic labels; and
adding the new image to the one semantic label of the semantic labels.
7. The computer-implemented method of claim 1 , wherein the number of times is based on empirical data.
8. The computer-implemented method of claim 1 , wherein the number of times is received from an administrator.
9. The computer-implemented method of claim 1 , wherein the different image is selected based on average distance from images within the initial seed cluster.
10. The computer-implemented method of claim 9 , wherein the different image is selected based on a largest average distance from images within the initial seed cluster.
11. The computer-implemented method of claim 1 further comprising:
generating a binary tree based on merged seed clusters.
12. The computer-implemented method of claim 11 , wherein the binary tree comprises a root, wherein the root is labeled with a semantically meaningful category.
13. The computer-implemented method of claim 11 , wherein traversal down the binary tree represents finer levels of granularity.
14. A computer-implemented method for facilitating cascading image clustering, the method comprising:
selecting an initial image from a set of reviews and further selecting a set of neighboring images to form an initial seed cluster;
repeating the selecting a number of times with a different image and a different set of neighboring images from the set of reviews, wherein the selecting and the repeating forms a plurality of seed clusters; and
merging a first seed cluster with a second seed cluster from the plurality of seed clusters to form a first merged seed cluster responsive to determining that similarity between the first seed cluster and the second seed cluster is within a threshold.
15. The computer-implemented method of claim 14 further comprising:
allocating a semantic label to the first merged seed cluster.
16. The computer-implemented method of claim 14 , wherein the initial image is randomly selected.
17. The computer-implemented method of claim 14 further comprising:
repeating the merging until all mergeable seed clusters are merged.
18. The computer-implemented method of claim 17 further comprising:
allocating a semantic label to each seed cluster after all mergeable seed clusters have been merged.
19. The computer-implemented method of claim 18 further comprising:
receiving a new image;
classifying the new image in one semantic label of the semantic labels; and
adding the new image to the one semantic label of the semantic labels.
20. The computer-implemented method of claim 14 , wherein the number of times is based on empirical data or is received from an administrator.
21. The computer-implemented method of claim 14 , wherein the different image is selected based on average distance from images within the initial seed cluster.