NetHi

An Interactive Web Application for Visualization of Network Clustering and Hierarchical Data

Motivation

• Graphs are powerful tools for representation of hierarchies and relations in data
• Graph clustering and representation techniques are attractive fields for decades, even more so, the large networks and hierarchies challenge remains
• Graphs help us better understand data of almost any kind, since all real-world domain data contains some sort of connections and orders within

Problems

Networks:

• Networks often have millions of nodes
• Hard to visualize
• User can be interested only in “important” parts of the network

Hierarchies:

• Usually represented using a dendogram
• All leaf nodes are shown
• User can have better overview when only one part of hierarchy can be shown

Solution description

NetHi is a web application implemented using Dash (https://dash.plotly.com) There are three parts of the application:

1. Setup (uploading dataset, selecting algorithm, selecting attributes and clustering)
2. Visualization (selecting layout, navigating through the data)
3. Report (clustering report, graph report)

Methods

NETWORKS

• Louvain community detection
• Spring layout coordinates computation and translation
• Graph construction (community nodes, virtual community nodes connected to all community nodes)

HIERARCHIES

• Agglomerative clustering
• Creating hierarchical structure
• Graph construction (adding sub-cluster nodes and connecting them to the leaf nodes)

VISUALIZATION

• Dash Cytoscape
• Coloring with matplotlib.cm5
• Expanding and collapsing
• Node information presentation on hover

Architecture

Architecture of the solution resembles MVC architecture in a following way:

Usage

1. Upload a dataset (either network or hierarchical data)
2. Select the type of clustering
3. Select attributes you want to use
4. Click on "CLUSTER" button
5. Click on "CREATE VISUALIZATION" (6. Click on "EXPAND GRAPH" if you want to see the whole graph)