Visual Analysis of Large Heterogeneous Social Networks by Semantic and Structural Abstraction

1. Visual Analysis of Large Heterogeneous Social Networks by Semantic and Structural Abstraction IEEE TRANSACTIONS ON VISUALIZATION AND COMPUTER GRAPHICS, VOL. 12, NO. 6, NOVEMBER/DECEMBER 2006 Authors: ZeqianShen, Kwan-Liu Ma and Tina Eliassi-Rad Presented By - Neha Agarwal

2. Introduction • Social network is a structure made up of individuals or organizations called as actors. • These actors are connected to each other through a relationship. • Analysis of social networks • Provides understanding of social processes and behaviors. • Active area of study beyond sociology. • Has become important in a variety of areas • Web • Organizational studies • Homeland security • Uncovers the invisible and hidden relationships between actors in a network. • Help identify critical actors or links.

3. In Visual analysis the social network is represented in the form of a graph where actors are shown as nodes and relationships as links. • Visual analytictools and techniques are used for social network analysis.

4. Challenges • Social network analysis for real-world applications presents several challenges: • The network can be very large containing upto millions of nodes • It makes drawing the entire network computationally infeasible. • Results in a mismatch between the density of information & the screen resolution available. • Large social networks commonly have complex relations among there actors • It makes visual analysis difficult. • A social network may have heterogeneous nodes and links Eg : A terrorism network nodes may be terrorists, organizations , attacks etc. • This creates additional complexity of visual analysis.

5. Visualization of Social Network • Visualization of the Entire Terrorism Network. There are 2,374 nodes and 8,767 links. • This is called the semantic or original graph.

6. Author’s Approach • Apply a set of information visualization techniques • Semantic abstraction • Structural abstraction • Importance filtering • Key Idea: Use a visual analytic tool OntoVis to guide the analysis with an auxiliary graph called ontology along with the techniques listed above.

7. Ontology Graph • Ontology is used to model the network i.e. the type of objects and/or concepts that exist, and their relations • Ontology Graph is a visual representation of an ontology. • Is much smaller (as in the number of actors and links) than the social network. • OntoVis uses it for selecting node types.

8. OntoVis • Is a social network visual analytics tool used for understanding large social networks. • Uses information in ontology to prune large heterogeneous network. • Makes large networks manageable and facilitate analytical reasoning. • Allows users to • examine different relationships, • infer new relationships • reveal hidden knowledge in the networks.

9. OntoVis – Semantic Abstraction • Provides a simple view of interest to help isolate key actors in social networks • Derived graphs are made by including only nodes whose types are selected in the ontology graph. This is called semantic abstraction. • Nodes that are not displayed in the derived graph are not deleted; rather they are converted into attributes of the displayed nodes. • User can browse the attribute list of nodes and convert a particular attribute back to a node.

10. Example of a very small terrorism network using semantic abstraction Classification Terrorist_Organization Country_Area A semantic abstraction that only contains terrorist organizations. Others become attributes. Ontology Graph

11. Leftists and Antiglobalization are converted and added. This views shows that two organizations are classified as Leftists and the other two are classified as Antiglobalization. Greece is converted to a node. This view shows that all four organizations are located in Greece.

12. OntoVis – Structural Abstraction • Structural Information (degree one nodes, duplicate paths) are used to condense the network. • OntoVis achieves this by removing • Nodes of degree one • duplicate paths i.e. the nodes which connect two nodes that are also directly connected • Reduces the visual complexity such that the user can readily focus on the key actors in the network and their relationships.

13. Example of structural Abstraction. .. Original network has many one-degree nodes and duplicate paths, which increase the visual complexity of the graph. Structural abstraction after removing one-degree nodes and duplicate paths. The key nodes in each cluster and the connections between them become clearer.

14. OntoVis – Importance Filtering • Helps users to decide which Types and Relationships should be selected and investigated. • Node size is used to indicate the importance of the node. Larger the size, more important it is. • In Original and Derived graph • Node Size is calculated from node degree.

15. In Ontology graph, • Node Type Size is calculated from disparity of connected types .Large value means that it preferentially connects to a small number of type. • The strong and weak links are determined by the frequency of links between two connected types. This is called node type disparity. • Large value indicates strong relation. • Small value indicates weak relation.

16. OntoVis – Graph Layout and Animation • (a) Layout using the Linlog. The node-overlapping problem is obvious. • (b) Overlapping is reduced by refinement processing(pushes the smaller nodes away from the larger nodes). • (c) Unconnected nodes are placed in the peripheral area. • Also the movement of nodes from their current positions to the new positions is smoothly animated.

17. Case Study 1 – Movie Network • The network contains eight node types : person, movie, role, studio, distributor, genre, award and country. • There are 35,312 nodes and 108,212 links in this network. • The relationship between type person and type role is weak. In social network analysis weak linkages are often very important

18. Semantic Abstraction : Role – Actor relationship • Select Role type. As a result, all other types of nodes become the attributes of role nodes • choose five particular roles: hero, scientist, love interest, sidekick and wimp • go through each role to select from its attribute list “person,” which picks all the people who acted in that role. • the derived abstraction is constructed accordingly and shows exactly this role-actor relationship Role(in Red) and Person (in Blue)

19. Structural Abstraction : Role – Actor relationship • Woody Allen, Sandra Bullock, Archibald Leach Thomas Sean Connery, and Bill Paxton are the actors that played three different roles. • Woody Allen is related to actress Maria “Mia” Farrow. The relationship between two good actors can be interesting. Thus, we decided to do further investigation on their movies and relationships.

20. Select people nodes: Woody Allen and Maria “Mia” Farrow • Select movies from Woody Allen’s. Select Person from the movies • People who worked more often with Woody Allen are Maria “Mia” Farrow, Louise Lasser, and Diane Keaton. Movies (in orange) and Person (in blue) :related to Woody Allen.

21. Observation • The information on IMDB shows that • Maria “Mia” Farrow had children with Woody Allen • Louise Lasser was his ex-wife • He dated Diane Keaton. • Therefore, we conclude that Woody Allen often worked with his girlfriends on his movies.

22. Case Study 2 – Terrorism Network • There are2,374 nodes and 8,767 links in this network

23. Ontology Graph • The network contains nine node types : terrorist organization, classification, terrorist, legal case, country/area, attack, attack target, weapon, and tactic. • The number in the square brackets is the count frequency for the corresponding node type

24. Select node type Terrorist_organization in ontology graph. • Al-Qaeda is one of the most dominant organization.

25. Select Country_Area attribute from Terrorist organization node. • The organizations in the upper-left cluster are located in Gaza Strip, West Bank, and Israel. • Al-Qaeda has connections in US, Yemen, Austria, Iraq. • Al-Qaeda is an international organization with connections in many countries and areas Terrorist Organizations (in blue) and Country_Area (in red).

26. Al-Qaeda is the organization related to most legal cases and terrorists. • The case involving most terrorists is US versus WadihEiHage et al. 98-CR-1023. • There are only two legal cases against Hamas. Terrorists (in orange), Legal Cases (in light green) and Terrorist Organizations (in blue).

27. Visualization of Terrorist Attacks (in green) related to Al-Qaeda. Tactics (Pink), Weapons(purple) and targets(light blue) • Al-Qaeda’s major attacking tactic is bombing using explosive weapons. • Their attacking targets vary from business, airport, government to diplomatic objects

28. Future Work • Other social network analysis techniques developed in sociology research can be incorporated to extend the capabilities of OntoVis. • More interaction between the derived abstraction and the ontology graph can be useful to uncover the errors and help users correct and refine the ontology graph. • Connection sub graphs capture the strongest relationships between two nodes and can be readily incorporated into OntoVis.

29. Related Work • Graph drawing methods have been developed for network visualization and are applicable to the analysis of social networks. • NetDraw is a program to visualize large social networks. It can handle heterogeneous networks with two node types. • Pajek • widely used network visualization and analysis system. • however it does not use ontology information effectively, because it is not designed specifically for heterogeneous social network visual analysis.

30. Conclusion • OntoVis can analyze a large graph and lead users to some interesting and useful findings. • The ontology graph and the abstractions provide strong support in visual analysis. • Visual analytics tools such as OntoVis are expected to enable breakthroughs in data exploration where information overload is a barrier to insight.

31. Questions?