|
Automatic Concept Maps :asA and :inA Dynamic Wiki
Navy SBIR FY2013.2
| Sol No.: |
Navy SBIR FY2013.2 |
| Topic No.: |
N132-128 |
| Topic Title: |
Automatic Concept Maps :asA and :inA Dynamic Wiki |
| Proposal No.: |
N132-128-1243 |
| Firm: |
Commonwealth Computer Research, Inc.
1422 Sachem Pl., Unit #1
Charlottesville, Virginia 22901 |
| Contact: |
Kevin Corby |
| Phone: |
(434) 284-9406 |
| Web Site: |
www.ccri.com |
| Abstract: |
Representing knowledge in a triple store is trivial, yet querying and visualizing the resulting knowledge is difficult and inefficient when the number of triples is large. Needing to understand the data models from each of the contributing processes and how these data models overlap or interact further complicates this problem. Visualization tools for knowledge stored in the Resource Description Framework (RDF) tend to simply enable visualization of the data via a graph. While this does show the available data in a relatively intuitive manner, it simply does not scale.
We will automatically identify intelligible, useful concepts that show how entities relate and expose undeclared relationships in the knowledge base. We will develop tools and techniques for concept generation to augment class/concept structures available from ontologies describing the knowledge store.
We address the main problem in two steps: (1) feature selection, (2) analytics and visualization. This proposal describes our proposed methodology for extracting features of entities described in an RDF knowledge base, and the application of these features to automatic concept map generation. We propose to develop a scalable manifold learning algorithm for concept extraction that will also enable a broader application of machine learning algorithms to RDF data at scale. |
| Benefits: |
In this effort, CCRi will develop prototype concept extraction algorithms and visualization from contextual queries. This will allow users to not only search across very large RDF stores but automatically filter the results down to meaningful and digestible sets of conceptually related data. Additionally, the prototype will develop flexible visualization tools to allow users to easily interact and learn from the data.
This system will be designed to transition to any major intelligence program of record such as any of the Distributed Common Grounds Systems or the Intelligence Community equivalents.
|
Return
|