Visualizing Social Networks

Unit-V
 Contents

 Introduction

 A Taxonomy of Visualizations

 Structural Visualization

 Semantic and Temporal Visualization

 Statistical Visualization

 The Convergence of Visualization,

Interaction & Analysis
 Structural and Semantic Filtering with Ontologies

 Centrality-based Visual Discovery
 Introduction

 Visualization is one of only two factors that are

responsible for the explosive development of all
fileds of modern science. (The other is

measurement.) 
 Visualization is an important tool to gain insight on
the structure and dynamics of complex social

networks. 
 Analytics and visualization allow users to gain
insight on the complexity of social dynamics.
 Online social networking sites

Academics

https://academic.microsoft.com/
https://aminer.org/

Music

http://liveplasma.com/
http://musicovery.com
https://last.fm

All in One

http://www.boostlabs.com/ibms-many-eyes-online-

data-visualization-tool/
References in Unit-II PPT.
 Taxonomy

 Structural

 Focuses precisely on structure, that can be thought of as the
topology of a graph that represents the actors and relationships in

a social network.
 Semantic

Nodes and Links can represent different aspects of the social



network.
 Temporal

 A particular case of a semantic diagram that uses time as the


main attribute.
 Statistical

 Useful for depicting the distribution of social network metrics.
 I. Structural Visualization

 The structure is the topology of a graph that represents the

 actors and relationships in a social network.
  Node-link diagrams
  Matrix-oriented Techniques
  Hybrid Techniques
 High level properties that a layout must satisfy:
 The positioning of nodes and links in a diagram should facilitate the
readability of the network.
 The positioning of nodes should help uncover any inherent
 clusterability of the network
 The position of nodes should result in a trustworthy representation
 of the social network.
 Low level properties :

 Proximity, color, shape, size, thickness, diameter etc.
 Node Link Diagrams

 In Sociograms, actors are represented as nodes, taking


geometric forms and relationships using lines between them.
 Different properties of the network can be encoded visually via

geometric properties, such as color, shape, size and thickness.
 Property-based Layouts

 Assigns the value of a node property as a location in a coordinate system.
 Force-directed and Energy-based Layouts
 Forces are designed to satisfy low-level properties that guarantee minimal overlap
of nodes and proximity of related nodes.

 Spectral Layouts
 Based on spectral algebra on key matrices that can be extracted from
the social structure. Eigenvectors of certain matrices can be used as
lower dimensional embedding of the graph.
 Radial Layouts

In radial layouts, nodes are placed in a circle and links are drawn as secant lines
through the circle. Target sociograms, use a centrality measure as radius.

Target Sociogram
 Force Directed Graph

• Attractive forces are defined between any pair of connected nodes,

while repulsive forces are defined between all pairs of nodes.
Layouts using networkx in Python
 Matrix Oriented Techniques

 Visualization metaphors designed to represent a social

network via an explicit display of its adjacency or
 incidence matrix.
 Each link is represented as a grid location with cartesian
 coordinates corresponding to the nodes in the network.
 Color and opacity are often used to represent important
 structural or semantic quantities.
 Challenge is to enable the users to visually identify
 local and global structures in the network.
 When the order of nodes is arbitrary, the matrix view
may not exhibit the clustering present in the data.
Matrix Representation of Social Network
 Hybrid Techniques

Rapid Graph Layout


 Nodes are first positioned in 2D space along a space-filing
curve. The position of a node along the curve is determined via
a clustering analysis to ensure the close proximity of related

nodes.
Path Visualization

 Uses hybrid matrix, where a basic layer depicts an adjacency

graph, while enabling path navigation on a different layer.
Node-Trix Visualization

 Represents small dense communities within a social network
as an adjacency matrix, and connects these within a node-link
diagram.
 II. Semantic Visualization

 Represent high level attributes and connections of actors

and links, either as specified explicitly in the data, or

implicitly inferred from cross-referencing external sources.
 Ontology-based Visualization: An ontology is a graph
whose nodes represent node types and links represent types

of representations.
 Pivot Graph: Starts with a selection of dimensions in a set
of multivariate, connected data and uses a rollup algorithm
to aggregate the data into discrete categories.
 Pivot Graph

Communication patterns between people in different departments

(rows) and locations (columns). The width of the arrows represents
the amount of communication going on (emails). This is aggregated
information, not just along the edges but also in the nodes.
Visual Analytics of Cell Phone Data using
MobiVis and OntoVis
 III. Temporal Visualization

 Social interaction is a time-dependent phenomenon, it is only


natural to represent the temporal dimension using visual means.
 Types of dynamic visualizations: Flipbooks, where nodes remain
static while the edges vary over time, and movies, where nodes

are allowed to move as the relationships change.
 Challenges when designing temporal visualizations is the
incorporation of the possibly disparate scales and variable

granularity of the temporal data.
 Mobivis addresses this problem via interactive filtering and
multi-scale time charts.

 PostHistory, helps visualize the dynamic social network that
arises from email exchanges. The visualization uses pixel-oriented
technique, which shows both relevance and quantity of emails over
time with color and size.
PostHistory Interface
 IV. Statistical Visualization

 Analysts often explore summary visualizations to understand

 the distributions of variables of interest.
 These variables correspond to network statistics such as
 degree, centrality and the clustering coefficient.
 Histograms

 Scatter Plots can be used to present centrality metrics such
as closeness, betweenness and degree, to understand the
 distribution in a network.
 Scatter plots are also useful in representing edge correlations.

 A point in the 2D space represents an edge, whose x
coordinate is the value of a given property of one of the
nodes connected to the edge and the y coordinate the value
of the same property of the node at the other end of the edge.
 The Convergence of Visualization,
Interaction and Analytics

 Tight integration of social network statistics and

visualization is necessary for effective exploration of social
networks.

 Cluster analysis of a social network drives the matrix
ordering for an effective adjacency matrix visual

representation and also node-link diagrams.
 Hierarchical clustering and summarization will be used

to visualize large social networks.
 Vizster and Social Action enhance the visualization with
an explicit representation of communities, enabling the
analyst to discover groups and interconnections
Visualization using Vizster
Social network analysis tool that integrates visualization
and statistics to improve the analytical process.
 Visualization and Interaction

 (1) They feature ego-centric views  Visual elements are arranged in relation
to a given context. Visualizations are produced for individual queries rather
than the whole.
 (2) They are playful  Primarily designed to help discover people and make sense
of structures of community.
 Tools such as Vizster and Touchgraph rely on force-directed layouts to
represent the local structure of the network and enable interaction to
expand or contract the network on demand.

 Dragging nodes around the visualization spaces lets users identify


community structures depending on the layout change over time.
 (3) They are dynamic.  PostHistory and Social Fragments are examples

where the visualization and interaction take time as the main dimension
 Structural and semantic filtering provides refined visual representations of

simplified and filtered data.
 The analysis of centrality and importance allows to design and test novel
representations and interactive techniques for exploration of social networks.
 Structural and Semantic Filtering with
Ontologies

 The ontology associates a specific type to nodes and

links that can be used to obtain a higher level of

abstraction.
 The authors compute a structural metric, based on
ontology, that specifies the disparity of links of a

given node.
 This disparity, measured in terms of the connectivity
variance of nodes of a given type, indicates whether a
particular node has weak or strong links to a certain
type or other nodes of the same type.
 Ontovis Output to represent Actor-Role

Heterogeneous data set that links actors, directors, movies, genres and
roles. Red nodes are Roles and Blue nodes are Actors.
 Centrality-based Visual Discovery and
Exploration

 A node centrality is a measure of the importance of a

 node, from a structural point of view.
 Metrics such as degree, betweenness, closeness and
Eigenvector centralities provide a ranking of the nodes in
terms of the number of connections, shortest paths or
 length of random walks.
 Target sociogram uses centrality metrics, so that
central nodes are positioned at the interior of concentric
 nodes and peripheral nodes in the outer regions.
 Edges can be ranked in terms of

  Importance of nodes connected to them
  Number of shortest paths that go through them
 Probability of appearing in a number of random walks through the
network.
 Hidden Link Discovery

 A centrality derivative defines the sensitivity of a node’s

 importance to the importance of all other nodes.
 By measuring the influence of a node onto all others, we can
 extract information that may be important to the user.
 A hidden link occurs when two nodes influence mutually to a
 great extent, but are not connected directly.
 This type of relationship often occurs between cluster centers
 or representatives.
 Largest magnitude derivatives often occur between cluster
centers while small magnitude derivatives occur within
 clusters.
 A leaf node in a cluster has a little influence on the cluster
center regarding its centrality.
Hidden Link Discovery
 References

 https://pdfs.semanticscholar.org/e3dc/01a7a02f5b5c8e959af6c80ad2


066d8df6f3.pdf
 https://cambridge-intelligence.com/keylines/social-networks/

 http://mark-kay.net/2014/08/15/network-graph-of-twitter-followers/

 https://www.cs.umd.edu/hcil/sna-

workshop/documents/SNAWorkshop_Nat.pdf
 https://www.cmu.edu/joss/content/articles/volume1/freeman.pdf

 https://www.researchgate.net/publication/229013647_Visual_Analyti

cs_of_Cell_Phone_Data_using_MobiVis_and_OntoVis
 http://hint.fm/papers/pivotgraph.pdf

 https://eagereyes.org/techniques/graphs-hairball


 http://alumni.media.mit.edu/~fviegas/posthistory/index.html

 http://www.cs.umd.edu/hcil/socialaction/