Exploratory Analysis of Multi-Edge Knowledge Graphs with Paired Clustered Grids

Auteurs

Durant E., Medoc N., Ghoniem M.

Référence

Proceedings - 2024 IEEE Visual Analytics Science and Technology VAST Challenge, Vast-Challenge 2024, pp. 15-16, 2024

Description

VAST Challenge 2024's Mini-Challenge 1 (MC1) aimed at discovering bias within a knowledge graph made of nodes and multi-edges extracted by two LLMs from articles related to fishing practices. Traditional force-based layouts lead to much visual clutter due to the overabundance and diversity of edges. Instead, we modeled the graph as a multilayer network [3] and arrange multi-edges in clustered adjacency grids where glyphs summarize edge aspects. Our interactive web application supports network exploration, pattern comparison, and outlier identification. Additionally, we used Papyrus, a corpus visualization tool to link topics to their sources, and discovered that analyst 'Harvey Janus' falsified edges related to SouthSeafood Express Corp, a company caught fishing illegally.

Lien

doi:10.1109/VASTChallenge64683.2024.00012