analiza aiqbench
This audit reveals that high accuracy scores on the widely used ISOT/Kaggle fake news corpus largely reflect dataset degeneracies like disjoint subject metadata rather than genuine veracity detection capabilities. The study demonstrates that even advanced models like DistilBERT fail to generalize under topic shifts, collapsing to near-chance performance on independent benchmarks compared to simpler linear baselines. For practitioners relying on these standard datasets, the findings underscore the critical need for rigorous distribution-shift testing and metadata-aware evaluation protocols before deploying classifiers in real-world scenarios.
What Does 99% Accuracy Measure? A Reproducible Audit of Shortcut Learning in a Widely Used Fake News Corpus
arXiv:2609.25006v1 Announce Type: cross Abstract: Text classifiers trained on the ISOT/Kaggle "Fake and Real News" corpus routinely report accuracy and F1 above 0.98, a level of performance that sits uneasily beside the difficulty of assessing veracity. Using a transparent TF-IDF and linear-classifier pipeline as a measurement instrument, we audit the corpus along three leakage channels and two distribution-shift protocols, releasing all code and derived numbers. First, the benchmark is partly degenerate: a classifier given only the subject metadata field, with the article text discarded, attains F1 = 1.000, since the two classes have disjoint subjects. Second, removing all three leakage channels, metadata, a newswire source tag present in 99.2% of real articles, and 6,251 duplicate documents contaminating 19.4% of a naive test split, lowers F1 by only 1.21 points (0.9935 to 0.9814); the residual signal is diffuse editorial style rather than a few giveaway tokens, since deleting the 1,000 highest-weight unigrams still leaves F1 = 0.926. Third, this style signal does not transfer: under a topic-disjoint protocol, average precision falls from 0.9995 to 0.9475 and deployed F1 from 0.9905 to 0.8067, with a prior-matched analysis confirming a genuine 5.2-point loss of discrimination, while temporal transfer is nearly lossless. A fine-tuned DistilBERT is stronger in-distribution (F1 = 0.9993) but degrades far more under topic shift, losing 12.9 average-precision points against the linear model's 5.2. Transferred to the independent LIAR benchmark, all three models fall to near-chance ranking (ROC-AUC 0.54-0.57), none beating a majority-class baseline. We conclude that within-corpus scores here quantify source and topic separability rather than veracity, that added capacity exploits the shortcut rather than avoiding it, and we recommend metadata-only, small-sample, and topic-disjoint baselines as inexpensive diagnostics for future work.
Title:What Does 99% Accuracy Measure? A Reproducible Audit of Shortcut Learning in a Widely Used Fake News Corpus
Abstract:Text classifiers trained on the ISOT/Kaggle "Fake and Real News" corpus routinely report accuracy and F1 above 0.98, a level of performance that sits uneasily beside the difficulty of assessing veracity. Using a transparent TF-IDF and linear-classifier pipeline as a measurement instrument, we audit the corpus along three leakage channels and two distribution-shift protocols, releasing all code and derived numbers. First, the benchmark is partly degenerate: a classifier given only the subject metadata field, with the article text discarded, attains F1 = 1.000, since the two classes have disjoint subjects. Second, removing all three leakage channels, metadata, a newswire source tag present in 99.2% of real articles, and 6,251 duplicate documents contaminating 19.4% of a naive test split, lowers F1 by only 1.21 points (0.9935 to 0.9814); the residual signal is diffuse editorial style rather than a few giveaway tokens, since deleting the 1,000 highest-weight unigrams still leaves F1 = 0.926. Third, this style signal does not transfer: under a topic-disjoint protocol, average precision falls from 0.9995 to 0.9475 and deployed F1 from 0.9905 to 0.8067, with a prior-matched analysis confirming a genuine 5.2-point loss of discrimination, while temporal transfer is nearly lossless. A fine-tuned DistilBERT is stronger in-distribution (F1 = 0.9993) but degrades far more under topic shift, losing 12.9 average-precision points against the linear model's 5.2. Transferred to the independent LIAR benchmark, all three models fall to near-chance ranking (ROC-AUC 0.54-0.57), none beating a majority-class baseline. We conclude that within-corpus scores here quantify source and topic separability rather than veracity, that added capacity exploits the shortcut rather than avoiding it, and we recommend metadata-only, small-sample, and topic-disjoint baselines as inexpensive diagnostics for future work.
17 pages, 11 figures, 11 tables. Code, experiment scripts, and machine-readable results: this https URL
Computation and Language (cs.CL); Machine Learning (cs.LG)
arXiv:2609.25006 [cs.CL]
(or arXiv:2609.25006v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.25006 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 405 cuvinte
Subiecte conexe
Shortcut Sensitivity in Agent-Safety Evaluations
arXiv · 17 oct. 2026
Lessons From the Leaderboard: What 5,000+ Kagglers Taught Us About Improving AI Reasoning
NVIDIA · 14 iul. 2026
In open RLVR, “improvement” depends on the instrument — a small GRPO testbed separating what training optimizes, measures, and teaches
LessWrong · 15 iun. 2026
Auditing Preference Biases and Fine-Tuning Language Models with Direct Preference Optimization on Anthropic HH-RLHF Using TRL and LoRA
MarkTechPost · 20 aug. 2026