Dense cluster filtering
AssigneeA9.com, Inc.
InventorShabnam Ghadar
Life of this patent
Estimating the term…
Abstract
Systems and methods are described for determining clusters for large collections of content items. A fast cluster-identifying algorithm can be used to find high density areas where certain less interesting content items might be clustered in a feature space. An example algorithm is a mean shift algorithm. Once these high-density clusters are located, a system can remove them and proceed to analyze the remaining data. Removing these clusters of featureless content items can greatly reduce the collection size and also enhance the overall quality of the collection. Labels can then be applied to clusters and, when a content item is received, classification algorithms can be used to assign an appropriate label to the content item.
Claims20
A computer-implemented method for assigning a label to an image comprising:
- obtaining a collection of images;
- determining first feature vectors for individual images in the collection of images;
- analyzing the first feature vectors to generate second feature vectors for the individual images, wherein the second feature vectors have a lower dimensionality than the first feature vectors;
- executing a mode seeking algorithm using the second feature vectors to identify a group of images having a lower image quality compared to images of the collection of images, the group of images having a feature space density greater than a density threshold;
- generating a filtered collection of images by removing the group of images having the lower image quality from the collection of images;
- performing a cluster analysis algorithm on the filtered collection of images to generate one or more identified clusters;
- assigning a label to the images of a cluster of the one or more identified clusters;
- obtaining a query image from a client device;
- determining that the query image belongs to the cluster of the one or more identified clusters;
- and
- sending the label to the client device.
A computer-implemented method for determining clusters in a dataset, comprising:
- obtaining a collection of content items;
- determining respective item features for individual content items of the collection of content items;
- determining a density metric for a subset of content items of the collection of content items based at least in part upon the respective item features;
- determining that the density metric for the subset of content items is greater than a density threshold;
- generating a filtered collection of content items by removing the subset of content items having the density metric greater than the density threshold from the collection of content items;
- performing a clustering analysis on the filtered collection of content items to generate one or more identified clusters;
- and
- assigning a label to a cluster of the one or more identified clusters.
A system, comprising:
- at least one processor;
- and
- memory including instructions that, when executed by the at least one processor, cause the system to:
- obtain a collection of content items;
- determine respective item features for individual content items of the collection of content items;
- determine a density metric for a subset of content items of the collection of content items based at least in part upon the respective item features;
- determine that the density metric for the subset of content items is greater than a density threshold;
- generate a filtered collection of content items by removing the subset of content items having the density metric greater than the density threshold from the collection of content items;
- perform a clustering analysis on the filtered collection of content items to generate one or more identified clusters;
- and
- assign a label to a cluster of the one or more identified clusters.
Dependencies are inferred from the claim text.