Researching Algorithms Design & Analysis, Graph Theory, and Social Network Analysis
Working in the Algorithms & Complexity Lab
Under the supervision of Professor Hovhannes Harutyunyan
Date: Aug 2024 – Now
My key role consisted of:
Designed Spider, a graph community detection algorithm combining geodesic expansion, modularity-guided refinement, and greedy merge matching.
Benchmarked Spider against Louvain, Leiden, Infomap, CNM, and Label Propagation on nine real-world networks (34–8,035 vertices) and LFR benchmarks, achieving top F1 and ARI on Karate Club and Political Blogs and NMI 0.88 on Primary School, where flow- and propagation-based baselines collapse to near-trivial partitions.
Introduced Weighted Average Geodesic Distance Modularity (wGDM), a size-consistent, label-free partition quality measure.
Conducted the first systematic study of metric backbone sparsification with Leiden, achieving 14–71% edge reduction across seven networks while preserving — and on dense contact networks improving — detection quality (High School F1: 0.834 → 0.970).
Built a fully reproducible experimental pipeline with fixed random seeds, baseline implementations, and automated evaluation scripts.