1 / 14 Integrated Visual Analysis of Global Terrorism Remco Chang Charlotte Visualization Center UNC Charlotte.

1 / 14

Integrated Visual Analysis of Global Terrorism

Remco ChangCharlotte Visualization Center

UNC Charlotte

2 / 14

Integrated Terrorism AnalysisMultimedia

Visual GTD

Real Time

Known Events

3 / 14

Video Analysis Goals

• to describe trends in news content over time

• to discover breaking news and hot topics over time

• to trace conceptual development of news

• to retrieve news of interests effectively

• to collect evidences and test hypotheses for intelligent analysis

• to compare group (such as different channels) differences in content

• to associate news content with social events

4 / 14

Multimedia Analysis

5 / 14

Video Analysis Example

CNN Fox News MSNBC• News contains view points and opinions• Find local, regional, national, and international reports of the

same event to get a complete picture

6 / 14

NVAC Collaborations

• PNNL – A. Sanfilippo (Content Analysis and Information Extraction of closed caption)

• PNNL – W. Pike (Emotional state extraction from closed caption)

• Penn State – A. MacEachren (Geographical analysis)• Georgia Tech – J. Stasko (Jigsaw, entity relationships)

• Visual Analytics is the point of integration!!

7 / 14

Integrating Terrorism Data Analysisand News Analysis

Terrorism Databases

Terrorism Visual

Analysis

News Story Databases

News Visual

Analysis

Jigsaw

TerrorismVA

BroadcastVA

Stab/TIBORReasoningEnvironment

Framing,Affective Analysis

NVAC

Next: full, Web-based multimedia content

8 / 14

Visual GTD Flow Chart

Entity Relationships(Geo-temporal Vis)

Dimensional Relationships(ParallelSets)

Entity Analysis(Search By Example)

9 / 14

Five Flexible Entry Components

10 / 14

11 / 14

Parallel Sets View

• Parallel Sets– Displays

relationships among categorical dimensions

– Shows intersections and distributions of categories

12 / 14

Parallel Sets View

• Dynamic filtering on continuous dimensions can show more information

• Here we see the large proportion of facility attacks and bombings in Latin America during the early 1980s

13 / 14

Analysis using Longest Common Sequence (LCS)• Two strings of data (each representing a series of events)

– GATCCAGT– GTACACTGAG

• Basic algorithm returns length of longest common subsequence: 6

• Can return trace of subsequence if desired:– GTCCAG

• GATCCAGT• GTACACTGAG

• Additional variations can take into account event gap penalties, time gap penalties, and exploration of shorter, or alternate, common subsequences

14 / 14

Grouping using MDS in 2D

• Each o represents a terrorist group

• Groups form cluster according to naturally occurring trend sizes

• Sharp divide between large clusters in right hemisphere

• Left hemisphere contains many smaller clusters

MDS Analysis by TargetType