AI Paper Atlas

Research map / Graph learning / Graph self-supervised learning

Structurally Human, Semantically Biased: Detecting LLM-Generated References with Embeddings and GNNs

Melika Mobini, Vincent Holst, Floriano Tori, Andres Algaba, Vincent Ginis

ICLR 2026 · Graph learning › Graph self-supervised learning · approach: LLMs and graph foundation models · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use title embeddings instead of graph-property featuresRandom Forest · Ground truth vs GPTaccuracy: 0.6079 -> 0.8346Improved
use title embeddings instead of graph-property featuresGCN · Ground truth vs GPTtest accuracy: 57.73 -> 93.10Improved
use title embeddings instead of graph-property featuresGIN · Ground truth vs GPTtest accuracy: 51.71 -> 93.28Improved
use title embeddings instead of graph-property featuresGAT · Ground truth vs GPTtest accuracy: 57.40 -> 93.78Improved
use title embeddings instead of graph-property featuresGraphSage · Ground truth vs GPTtest accuracy: 55.59 -> 93.14Improved
use title embeddings instead of graph-property featuresRandom Forest · Ground truth vs Claudeaccuracy: 0.5389 -> 0.7688Improved
replace OpenAI embeddings with SPECTER (ablation)Random Forest · Ground truth vs GPTaccuracy: 0.8346 -> 0.7502Hurt
replace OpenAI embeddings with SPECTER (ablation)Random Forest · Ground truth vs Claudeaccuracy: 0.7688 -> 0.6878Hurt
replace semantic embeddings with i.i.d. random vectors of same dimension (ablation)Random Forest · Ground truth vs GPTaccuracy: 0.8346 -> 0.5047Hurt
use temporally constrained random baseline (ablation)Random Forest on graph properties · Ground truth vs GPTaccuracy: 0.6079 -> 0.6119No clear effect
swap generator at test time (train GPT-4o, test Claude)Random Forest on OpenAI embeddings · Ground truth vs generatedaccuracy: 0.8346 -> 0.7205Hurt
swap generator at test time (train GPT-4o, test Claude)GCN · Ground truth vs generatedaccuracy: 93.10 -> 0.8001Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy