Chronoblox: Chronophotographic Sequential Graph Visualization

TL;DR

Chronoblox addresses the challenge of visualizing dynamic graphs by placing all time-sliced snapshots into a shared inter-temporal embedding space to reveal micro-to-meso evolution. The method builds meta-graphs of node groups, computes inter-phase similarity with $M(b,b')=\frac{|m(b)\cap m(b')|}{|m(b)\cup m(b')|}$, learns a $64$-dimensional embedding via Node2Vec, and projects to 2D with PaCMAP for chronophotographic visualization. Key contributions include a unified layout that preserves inter-temporal relationships, an interactive interface with Chronophotographic View, Inter-Temporal Lineages, and Alluvial View, and validation on synthetic SBM/Louvain cases as well as a real Impact Investing retweet network. The work highlights modularity and practical usefulness for exploring dynamic community structure, while noting scalability limits (approx. 1000 node groups) and future work on zooming and intra-temporal aggregation.

Abstract

We introduce Chronoblox, a system for visualizing dynamic graphs. Chronoblox consists of a chronophotography of a sequence of graph snapshots based on a single embedding space common to all time periods. The goal of Chronoblox is to project all snapshots onto a common visualization space so as to represent both local and global dynamics at a glance. In this short paper, we review both the embedding and spatialization strategies. We then explain the way in which Chronoblox translates micro to meso structural evolution visually. We finally evaluate our approach using a synthetic network before illustrating it on a real world retweet network.

Figures (2)

• Figure 1: The Chronoblox interface. (A) Layer tab, (B) Alluvial view, (C) Timeline, and (D) Chronophotographic view. It shows the evolution of $\Xi$, the Impact Investing retweet network with node groups computed, here as stochastic blocks of Twitter/X users (https://lobbeque.github.io/chronoblox_examples/impact_investing_sbm.html). For illustrative purposes we circle country groups in 2014 and UK lineages over the whole sequence, see text in Sec. \ref{['application']}.
• Figure 2: Chronophotographies of the synthetic graph sequences. (A) Nodes are grouped via the SBM method (https://lobbeque.github.io/chronoblox_examples/toy_model_sbm.html), (B) Nodes are grouped via the Louvain algorithm (https://lobbeque.github.io/chronoblox_examples/toy_model_louvain.html).