Multi-facet classification scheme for cataloging of information artifacts

Patent metadata

Field	Value
Publication number	US-9710760-B2
Application number	US-82560910-A
Country	US
Kind code	B2
Filing date	Jun 29, 2010
Priority date	Jun 29, 2010
Publication date	Jul 18, 2017
Grant date	Jul 18, 2017

Abstract

A system and method for constructing a hierarchical multi-faceted classification structure includes organizing a plurality of visual categories into a multi-relational reference ontology that accounts for a plurality of different types of relationships. Media artifacts are categorized into the plurality of visual categories. The categories of artifacts are refined based on faceted ontology relationships or constraints from the multi-relational reference ontology. The multi-relational reference ontology and the one or more media artifacts with relationships are stored as the hierarchical multi-faceted classification structure in computer readable memory storage.

First claim

What is claimed is:

1. A method for constructing a hierarchical multi-faceted classification structure comprising:
classifying a plurality of visual categories into a multi-relational reference ontology that accounts for a plurality of different types of relationships, the multi-relational reference ontology including a plurality of faceted nodes of visual categories that can co-occur in an image and a plurality of regular nodes of visual categories that are mutually exclusive in the image;
determining a plurality of content-based tags for one or more media artifacts, wherein each content-based tag of the plurality of content-based tags is determined based on a confidence score using the plurality of visual categories;
for a set of conflicting content-based tags from the plurality of content-based tags, removing all but one conflicting content-based tag from visual categories that are mutually exclusive for each of the one or more media artifacts;
storing the multi-relational reference ontology, remaining content-based tags of the plurality of content-based tags, and the one or more media artifacts with relationships as a hierarchical multi-faceted classification structure in computer readable memory storage; and
retrieving at least one media artifact from the one or more media artifacts by at least one content-based tagging application using the multi-relational reference ontology and the remaining content-based tags.

2. The method as recited in claim 1 , wherein refining includes refining the categorizing based on at least one of faceted ontology relationships and constraints from the multi-relational reference ontology.

3. The method as recited in claim 2 , wherein the at least one of faceted ontology relationships and constraints includes at least one of mutual exclusivity, co-occurrence and relatedness.

4. The method as recited in claim 2 , wherein refining includes at least one of selecting, boosting and excluding category labels using relationship rules extracted from the multi-relational reference ontology.

5. The method as recited in claim 2 , wherein refining includes at least one of selecting, boosting and excluding category labels using statistical methods based on relationships in the multi-relational reference ontology.

6. The method as recited in claim 2 , wherein refining includes at least one of selecting, boosting and excluding category labels based on detecting confidence scores for categories.

7. The method as recited in claim 2 , wherein refining includes at least one of selecting, boosting and excluding category labels based on characteristics of categories in the multi-relational reference ontology.

8. The method as recited in claim 1 , wherein the plurality of different types of relationships includes at least one of parent-child relationships, sibling relationships, negation relationships and relationships among visual semantic concepts.

9. The method as recited in claim 1 , wherein the hierarchical multi-faceted classifications structure is constructed in accordance with plurality of rules comprising:
representing categories as a unique semantic idea having a category node;
linking each category node to one or more children nodes, if present, that reflect a semantic decomposition of a parent category such that children nodes are all category nodes or facet nodes;
partitioning children category nodes using a semantic category with respect to parent nodes wherein every children category folder has at least one sibling and children nodes of each parent are mutually exclusive and complete with respect to the semantics of the parent category; and
semantic perspective partitioning facet nodes where a facet node is complete with respect to the parent category and provides a perspective for further decomposing the parent category into parallel semantic categorical decompositions and peer facet nodes are redundant with respect to each other.

10. The method as recited in claim 1 , wherein categorizing includes decomposing a parent node of the multi-relational reference ontology as parallel decompositions of faceted nodes.

11. The method as recited in claim 1 , wherein the plurality of faceted nodes and the plurality of regular nodes are orthogonally partitioned with respect to each other to indicate respective membership in a corresponding one of the plurality of faceted nodes and the plurality of regular nodes, and wherein a first partitioning orientation indicates membership to the plurality of faceted nodes and a second partitioning orientation orthogonal to the first partitioning orientation indicates membership to the plurality of regular nodes.

12. The method as recited in claim 1 , wherein conflicting sibling nodes that appear as mutually exclusive in the multi-relational reference ontology are eliminated by selecting at most one of the conflicting sibling nodes for recommendation for the image.

13. The method as recited in claim 1 , wherein each of the plurality of faceted nodes functions as a respective root node for a respective subset of the plurality of regular nodes, the multi-relational reference ontology includes a plurality of horizontal planes, and co-occurrence between any of the plurality of faceted nodes is indicated by being on a same one of the plurality of horizontal planes.

14. The method as recited in claim 13 , wherein the multi-relational reference ontology includes a plurality of vertical planes, and co-occurrence between any of the plurality of regular nodes is indicated by being on a same one of the plurality of vertical planes.

15. A computer readable storage medium comprising a computer readable program for constructing a hierarchical multi-faceted classification structure, wherein the computer readable program when executed on a computer causes the computer to perform the steps of:
classifying a plurality of visual categories into a multi-relational reference ontology that accounts for a plurality of different types of relationships, the multi-relational reference ontology including a plurality of faceted nodes of visual categories that can co-occur in an image and a plurality of regular nodes of visual categories that are mutually exclusive in the image;
categorizing one or more media artifacts into the plurality of visual categories;
refining the categorizing to provide refined categories;
performing content-based tagging of the one or more media artifacts based upon a confidence score using the refined categories to determine a plurality of content-based tags for the plurality of media artifacts;
for a set of conflicting content-based tags from the plurality of content-based tags, removing all but one conflicting content-based tag from visual categories that are mutually exclusive for each of the one or more media artifacts;
storing the multi-relational reference ontology, remaining content-based tags of the plurality of content-based tags, and the one or more media artifacts with relationships as the hierarchical multi-faceted classification structure in computer readable memory storage; and
retrieving at least one media artifact from the one or more media artifacts by at least one content-based tagging application using the multi-relational reference ontology and the remaining content-based tags.

16. The computer readable storage medium as recited in claim 15 , wherein classifying a plurality of visual categories includes constructing the hierarchical multi-faceted classification structure in accordance with a plurality of rules including:
representing categories as a unique semantic idea having a category node;
linking each category no

Assignees

• Hill Matthew
• Kender John R
• Natsev Apostol
• Nguyen Quoc-Bao
• Smith John R
• Tesic Jelena
• Xie Lexing
• Yan Rong
• IBM

Inventors

• Hill Matthew
• Kender John R
• Natsev Apostol
• Nguyen Quoc-Bao
• Smith John R
• Tesic Jelena
• Xie Lexing
• Yan Rong

Classifications

• G06V10/765

  using rules for classification or partitioning the feature space · CPC title

• G06F18/24765

  Rule-based classification · CPC title

• G06N20/00 Primary

  Machine learning · CPC title

• G06F17/30265

  Physics · mapped topic

• G06N99/005 Primary

  Physics · mapped topic