Web Graph Centrality Used to Optimize Language Model Pretraining Data Selection
Researchers have introduced WebGraphMix, a framework that uses the structural topology of the web — specifically how central or peripheral a website is in the Common Crawl host-level graph — to guide the selection of pretraining data for language models. Unlike existing methods, it requires no model training, labeled data, or downstream supervision, making it computationally lightweight. The approach achieved 41.4% average accuracy across 23 tasks compared to 39.8% for uniform sampling, and reached 43.8% when combined with traditional content-quality classifiers.
WebGraphMix is a data selection framework proposed by researchers that leverages structural centrality scores derived from the Common Crawl host-level web graph to curate pretraining data for large language models. The core hypothesis is that central web hosts — those with many connections — expose models to broadly reusable abstractions, while peripheral hosts encode specialized, long-tail knowledge, and that both types of content are complementary. The framework computes these centrality scores efficiently at web scale without requiring any model training or labeled datasets, distinguishing it from classifier-based data selection methods. Models trained at 400M and 1B parameter scales using a 1:1 mixture of central and peripheral documents outperformed uniform sampling (41.4% vs. 39.8% average across 23 diverse tasks). Further gains were achieved by combining WebGraphMix's structural scores with document-level quality classifier scores, reaching 43.8%. The authors argue that web graph topology represents a meaningful and largely orthogonal axis for data curation relative to existing content-based approaches. The work was integrated into the DataComp-LM pipeline and submitted to arXiv in June 2026.
What's missing
The study does not report statistical significance or confidence intervals for the performance differences, making it difficult to assess whether the gains over uniform sampling are robust. It is also unclear how WebGraphMix generalizes beyond Common Crawl to other pretraining corpora, or how sensitive results are to the specific centrality metric chosen. The paper has not yet undergone peer review.
What different sources said
- arXiv cs.AICenter
Hubs or Fringes: Pretraining Data Selection via Web Graph Centrality
Related
Gut Bacteria Enzyme Found to Break Down Heat-Processed Food Compounds, Producing Novel Biogenic Amines
Researchers have discovered that an enzyme in common gut bacteria can degrade N-epsilon-carboxymethyllysine (CML), a compound formed during thermal food processing, producing previously unknown biogenic amines. The enzyme, ornithine decarboxylase SpeC from enterobacteria, acts on CML and related modified lysine derivatives through a low-level 'underground' catalytic activity. This finding suggests a previously unrecognized communication axis between thermally processed dietary compounds and gut microbial physiology, with potential implications for host health.
Full-Length Gene Sequencing Reveals Two Distinct Bacterial Communities in Black-Legged Ticks Expanding Into Canada
Researchers used Oxford Nanopore full-length 16S rRNA gene sequencing to characterize the microbiome of Ixodes scapularis black-legged ticks collected in Nova Scotia, Canada, distinguishing between tick-adapted bacteria and environmentally acquired bacteria. The study comes as I. scapularis — the primary vector of Lyme disease — is rapidly expanding northward into Canada due to climate change. The findings suggest that environmentally derived bacteria in tick microbiomes are not mere contamination, which has implications for how tick microbiome data is collected and interpreted across surveillance studies.
Study Identifies Metabolic Link Between Cell Envelope Stress and Biofilm Formation in Bacteria
Researchers have discovered that the metabolite acetyl-CoA directly inhibits enzymes that degrade the bacterial signaling molecule c-di-GMP, connecting cell envelope biosynthesis stress to biofilm formation in Pseudomonas aeruginosa. The study found that sub-inhibitory concentrations of antibiotics targeting early peptidoglycan biosynthesis — but not other antibiotic classes — elevate c-di-GMP levels by reducing phosphodiesterase activity, with acetyl-CoA competing for the enzyme active site. Because the relevant enzyme domain is broadly conserved across bacterial species, this checkpoint mechanism may be widespread and could have implications for understanding antibiotic-induced biofilm responses.