Reproduce “A Graph-based Stratified Sampling Methodology for the Analysis of (Underground) Forums” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Giorgio Di Tizio, Gilberto Atondo Siu, Alice Hutchings, Fabio Massacci
A graph-based stratified sampling methodology is introduced to efficiently select training data for machine learning classifiers in underground forums, achieving up to 30% improved recall for cybercrime detection compared to random sampling. The methodology leverages network centrality metrics to generate more representative and effective training samples, enhancing the detection of criminal activities.
Key insights
- Uniform distribution sampling based on post degree centrality substantially improves recall performance for cybercrime detection.
- Proportional stratified samples based on various centrality metrics did not significantly outperform random sampling.
- Classifiers trained with different sampling strategies can disagree on up to 33% of classifications, highlighting the impact of sampling choices on model outcomes and detected behaviors.
- Incorporating network structure through graph-based sampling can lead to more efficient resource utilization for data annotation and better detection rates for critical cybercrime activities.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.