[Demo Paper] SPREADVIZ: Decoding the Viral DNA of Social Networks through Analytics and Visualization

TL;DR

SPREADVIZ is a comprehensive web-based tool that allows researchers to visualize and analyze how information, diseases, or rumors propagate through social networks. By combining structural graph metrics (like k-core and K-truss) with epidemiological models (SIR, IC, LT), it provides a "sandbox" for identifying and testing the power of influential spreaders in real-world datasets.

Strategic Positioning: This work serves as a vital bridge between Graph Theory and Interactive Data Science, providing a unified GUI for complex algorithms that were previously restricted to command-line scripts.

The Problem: Why Spreading is Hard to Visualize

In the study of "Influence Maximization," we often deal with a "black box" problem. We know a node is central because its PageRank or Betweenness score is high, but we rarely see how that centrality translates into an actual cascade.

Prior tools were often limited to:

• Static Analysis: Showing centralities without the simulation.
• Single-Model Focus: Only supporting one type of epidemic model (e.g., only SIR).
• Scale Issues: Failing to provide intuitive ways to compare different "seed sets" of influential nodes side-by-side.

Methodology: The Three Pillars of SPREADVIZ

SPREADVIZ architecture is built on three interconnected modules that turn raw edge-lists into actionable insights.

1. Graph Exploration & Decomposition

Before simulating a virus or a marketing campaign, the tool performs a "structural X-ray." It calculates:

• Degree & PageRank: Local and global popularity.
• k-core & K-truss: Identifying the "dense heart" of the network. The authors highlight that nodes in the maximal k-core often possess superior spreading power even if their raw degree is lower.

2. Influential Node Detection

The system answers the fundamental question: Who should we start with?

• Single Spreader: Ranks nodes based on centrality.
• Multiple Spreaders: Implements the Greedy approach and its variants (like CELF++) to solve the NP-hard Influence Maximization problem, providing a near-optimal set of $k$ initial nodes.

3. Interactive Visualization

Using D3.js, SPREADVIZ provides a step-by-step playback of the infection. Users can watch "infected" nodes turn color and spread the "disease" to their neighbors based on the chosen probability $(\beta )$.

Simulation Models: SIR, LT, and IC

The tool supports three heavyweights of spreading theory:

1. SIR Model: Classic epidemiology (Susceptible $o$ Infected $o$ Recovered).
2. Linear Threshold (LT): A node flips only if enough of its neighbors are active (modeling collective behavior like social movements).
3. Independent Cascade (IC): Each active node has a one-shot chance to flip a neighbor (modeling information "bursts").

Experimental Insights & Results

The tool provides a comparative analytics suite. Users can plot Cumulative Influence (the total number of nodes infected over time) for different methods.

For example, a user can compare:

• A seed set chosen by Max Degree.
• A seed set chosen by K-truss.
• A seed set chosen Manually.

Ablation and Comparison: Through these visual plots, the authors demonstrate that the Greedy optimization typically outperforms heuristic methods (like high degree) in reaching the widest possible audience, though structural methods like K-truss provide a computationally "cheaper" and highly effective alternative for single-node targeting.

Note: The tool generates step-by-step infection curves to visualize the velocity and breadth of the spread.

Critical Analysis & Conclusion

Takeaway

SPREADVIZ is a significant utility for both Viral Marketing and Epidemiology. Its ability to swap between SIR and LT models while visualizing the graph's k-core structure allows researchers to see why certain nodes are influential—often because they connect disparate "communities" rather than just having many friends.

Limitations

• Scalability: While the tool handles medium-sized networks, the Greedy Influence Maximization algorithm is computationally expensive ($O(knm)$). For massive networks (millions of nodes), the server-side computation might lag.
• Temporal Dynamics: Most social networks change over time; SPREADVIZ currently focuses on static graph snapshots.

Future Outlook

As AI moves toward Graph Neural Networks (GNNs), tools like SPREADVIZ could evolve to visualize how GNNs learn "influence embeddings," potentially predicting spreading patterns without the need for 1,000+ Monte Carlo simulations.