Dynamic Network Visualization in 1 5 D Lei

Dynamic Network Visualization in 1. 5 D Lei Shi *, Chen Wang *, Zhen Wen † * IBM Research – China † IBM T. J. Watson Research Center

Mobile SMS Network – Spammer

Mobile SMS Network – Non-Spammer

Mobile SMS Network – Spammer/Non-Spammer

Outline p Problem p Related Works & Previous Solutions p Data Processing – Dynamic Ego Network – Event-based Dynamic Networks p Visualization – Metaphor – Graph layouts – Interactions p Case Study – Mobile SMS Networks – Infovis/VAST Conferences

Background & Research Problem p Dynamic networks are overwhelming in the reality, big value add-on with visualization – Demonstrate huge evolving social network over SNS/Twitter for community detection – Show the dynamically changing ad-hoc-routing sensor networks for diagnosis purpose – Visual evidence of growing telecom networks for role identification: employee retention p Problem with dynamic network visualization – How to encode the time dimension • 3 D? Video? Summarization? – How to deal with scalability • Finer time granularity => Larger network complexity => (visual clutter, bigger computation cost) – Usability for interactive analytics • Help automate pattern discovery

Related Works: Dynamic Movie Approach

Related Works: Small Multiple Display

Related Works: Dynamic Graph Drawing p Objective: preserve the user’s mental map [ELM 91][MEL 95] – Orthogonal ordering – Proximity relationships – Topology p Mental-map preserving dynamic graph drawing algorithms – Online dynamic graph drawing algorithms: compute the layout of one time frame only from its previous time frame and the graph change • Graph adjustment, e. g. force-scan algorithm [MEL 95] • Extension from KK model [BBP 07] • Incremental graph layout [North 95][DKM 06] – Offline dynamic graph drawing algorithms: take all the graphs in previous time frame into consideration • Optimize global stability [DGK 01][CKN 03] • Encode the graph change in multi-layer representation [BC 02] – Special graph/drawing types • Hierarchical graph [North 95][NW 02], clustered graph [HEW 98][FT 04] • Orthogonal graph [PT 98][GBP 04], radial graph [YFD 01]

1. 5 D Dynamic Network Visualization p Basic idea: only consider the dynamic ego network central to one node – Many network analytics applications are egocentric: person role analysis, company collaborations analysis – Rationality: demultiplex the data in network domain (1. 5 D Vis) v. s. time domain (movie approach) v. s. space domain (small multiple displays) p Benefits: – Fit both time and network info into a single static 2 D visualization (0. 5 D time, 1. 5 D network) – Reduced network size and layout computation complexity, less visual clutter – Better support dynamic network analytics, e. g. temporal network pattern discovery p Trade-offs: – Will lose the overall graph topology semantics and the topology evolving patterns – Compensate a little with interactions

Visual Metaphor Horizontal Glyph 2 -hop node central node sending/receiving trend 1 -hop node time-dependent edge time-independent edge Radial Glyph

Data Processing for 1. 5 D Visualization p 3 steps to generate the dynamic ego network data for 1. 5 D visualization – Slotting: – Extraction: reduce each slotted graph into the ego graph central to the selected node – Compression: aggregate the ego graphs into a single graph with timedependent and time-independent edges p Event-based dynamic networks – Insertion: the new event node is added to the graph, an edge is added between the event node and existing nodes if this event ever happens to it at a specific time

Graph Layout p Customized force-directed layout model for small/medium-sized networks: – Split the central trend node into several subnodes – Fix the sub-node locations at Y axis – Add stability constraints to non-central nodes to place them near their average time to the center – A balance of time-dependent and timeindependent edge forces p Circular graph layout for large networks – Partition – Sort – Assign

Graph Interactions p Timeline navigation zoom & pan zoom p Egocentric graph navigation drill-in to new central node view

Case Study — Mobile SMS Network p For each people, send only one message in one time For some people, send multiple messages in multiple times

Case Study — Mobile SMS Network p Unidirectional communication (no reply) Bidirectional communication (send & reply)

Case Study — Mobile SMS Network p No communications between receivers (friends) Connections between receivers (friends)

Case Study — Mobile SMS Network p Smooth transmissions (the automatic scanning with powerful machine) Irregular transmission pattern

Case Study — Conference Author Networks p Infovis author network: ego-edge mode, Prof. Stasko’s network

Case Study — Conference Author Networks p Infovis author network: network-edge mode Dr. Wong’s network Prof. Munzner’s network

Case Study — Conference Author Networks p VAST author network Overview Prof. Ribarsky’s network

Thai Korean Traditional Chinese Russian Gracias Thank You English Italian Obrigado Brazilian Portuguese Arabic Grazie Spanish Danke German Simplified Chinese Merci French Japanese Tamil 22 Hindi