Elsa C. Augustenborg Gary R. Danielson Andrew E. Beck Pacific Northwest National Laboratory PNNL-SA-75867
Overview Technical challenges Institutional challenges Architectural approach Examples: Promising tools and techniques Keys to effective implementation
Technical Challenges Overwhelming amount of information 1.2 Zettabytes (10 21 ) on the Internet in 2010 Types Structure Semi-Structured Unstructured Spatial/Temporal Multi-modal information Text Image Video Loosely related
Institutional Challenges More information but less time to analyze Less time to formulate analysis into a defensible analytical product Expertise continuously leaving institutions
Architectural Approach Lightweight Single function and minimal learning curve Work with desktop tools - exploits common GUI metaphors Tools can launch each other for added functionality In situ Exploit extensibility of application like Microsoft Office and Browsers (I.E., FireFox) Work within institutional security framework and support Track user actions Answer the question How did I get here? Storage and reuse of analytical techniques Preservation of institutional knowledge
Example:Lightweight Fuel Cycle App Capture of expert knowledge Map contains vetted expert knowledge about the nuclear fuel cycle As the map is zoomed the terms for that part of the fuel cycle are revealed The ontology can be filtered on relationship type and strength of applicability to selected portion of the fuel cycle Videos and other tutorial information can be attached to fuel cycle stages Ontology can be expanded by users Other maps can be created and linked to ontology
Example: Lightweight Search Assistant App Capture of expert knowledge Terms from the fuel cycle ontology, or other word lists, can be dragged to the Search App. The search app generates search phrases based on the selected terms and users logic Search phrases can be saved in a way that is not specific to a particular search engine Experts can create search phases for re-use Ontology can include translations
Example: Two Lightweight Taxonomy Driven Visualization Apps Rapidly understanding coverage Drag document collection/search results/rows from SQL query onto Fuel Cycle map (or any available map/ontology). Can use term clustering and strength of relevance to colour the map nodes. In situ use: Visualize web pages while browsing. In-depth exploration Only relevant sections displayed in pie. Drag slices of interest onto palette Slice explodes with more detail Drag terms back onto Search assistance to refine search Limit translation effort If Ontology has translations, non-english documents will be part of visualization.
Example: Statistical extraction of concepts app (non a priori) Rapidly Understand Themes No need to define a taxonomy or ontology. For a document collection concepts are identified based on statistical algorithms and thus are language non-specific Here terms are shown with a clickable temporal frequency visualization displaying term frequency over time based on document date
Example: Lightweight speed reader Concentrator app Understanding at a glance Drop a document and a set of terms of interest into the app Indicates clusters of terms or interest Use slider to select the amount of data displayed around each term or term cluster Algorithm looks for longest terms first and does not double count
Example: Frictionless Evidence Capture : Hunter-Gatherer Capture snippet with provenance Highlight information on a web page, Email, Word PDF, etc. Capture provenance and highlighted data into a repository Repository can be user-defined on the workstation or an enterprise-wide document management system Provenance can optionally be edited during capture via Outlook style notification window Lookup what is known Highlight information Request a search of one or more repositories
Example: Multi-modal data fusion Index using a single text metaphor Create a signature for an image or video that mimics signatures for documents Search engine treats the image like a document Images and documents - indexed together Images become self-annotating when indexed within the document containing them Special query interface allows user to combine exemplar images or portions along with search terms All of this combined/fused with: Time references (absolute or relative) Names and locations based on person, place and organization
Example: Use of game technology Analytical Gaming Framework A game hosting environment that enables player interactions from a game description and captures game play history Example Illicit Trafficking game Players in the roles of countries and companies Player Objectives: Obtain, or prevent, a weapons capability by obtaining goods and technologies. Used a simplified fuel cycle map for game definition and tracking acquisitions
Example: Lightweight Decision Maker App Process Modelling for Analysis Use Bayesian or Dempster-Schaffer based hierarchical model templates Decision-Maker Example User can attach evidence (images, documents, annotations) to nodes in the model and assign values indicating whether the evidence supports or refutes the proposition The model makes calculation that can help the analyst form and support the conclusion The model with attached evidence can be sent to others for review and comment Model with evidence can also be used to populate an Electronic briefing book.
Application Integration Portal based on accepted model Portal isn t lightweight but rather a place were lightweight apps can work in concert to perform larger tasks Could be distribution point for lightweight apps Clearing house for data and application May not contain data or applications but rather references them Can be used as the App store for light-weight applications The portal concept is in wide-spread use but basing portal structure on analytical structures adds significant value The structure of the portal mimics known and accepted models such as the physical model A simple structure example Top level with links to an App store with references and training for analytical application and a list of countries Clicking on a country would bringing up past present and future year folders each containing a similar structure for containing data and plans for that year
Keys to effective implementation User interface Leverage organic metaphors Leverage behavior of common applications Integrate with existing applications as much as possible Sustainable and supportable Leverage what exists and can be supported Organizational involvement End-users and support departments
Conclusion Lightweight apps help leverage the large quantities of available data to produce high quality, defensible analytical products Effective, frictionless, usable and supportable Information understanding based on well understood taxonomies Bringing information to the analyst and capturing information from the analyst Information fusion including multi-modal data Approaches to uniform organization, navigation and presentation Designed and developed with targeted client