Tools
Tools
Shehnaj, Mohammad and Razzaq, Abdul (2026) Input-Length Effects on Transformer-based Misinformation Detection: An Empirical Study. In: Proceedings of the 41st ACM/SIGAPP Symposium on Applied Computing. ACM, Thessaloniki, Greece, pp. 900-907. ISBN 979-840072294-3
Preview |
PDF
Download (1MB) | Preview |
Abstract
Transformer models dominate misinformation detection, yet the role of input length—from headline snippets to full articles—remains under-examined. Our objective is to quantify how length affects both effectiveness and efficiency under a single, transparent protocol, and to distil deployment-ready guidance.
We harmonise four public news datasets to binary labels, de-duplicate them, and stratify inputs into short, medium, long, and a mixed-length set. Each architecture is fine-tuned once on the balanced training split and evaluated per length bin on the test set; 512-token encoders use head+tail cropping, long-context models ingest extended context, and the efficiency protocol fixes micro-batching and padding. We report macro-F1/Accuracy/AUC together with latency, throughput, and memory, using paired tests on saved predictions.
Results show a clear, monotonic length effect: encoder baselines remain strong on short/medium inputs, while long-context models recover recall on long articles; runtime, rather than memory, is the primary cost of longer inputs. These findings yield a simple serving policy: route short/medium inputs to a 512-token encoder and genuinely long inputs to a long-context encoder. We release artefacts to reproduce all tables and figures: https://doi.org/10.5281/zenodo.17187547.
| Item Type: | Book Section |
|---|---|
| Additional Information: | This work is licensed under a Creative Commons Attribution 4.0 International License: https://creativecommons.org/licenses/by/4.0 |
| Uncontrolled Keywords: | efficiency; evaluation protocol; input length; long-context models; misinformation detection; transformers |
| Subjects: | Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science B Philosophy. Psychology. Religion > BJ Ethics > Conduct of life > Reliability > Information integrity H Social Sciences > HM Sociology > Information Science > Communication > Mass media |
| Divisions: | School of Computing > Staff Research and Publications |
| Depositing User: | Tamara Malone |
| Date Deposited: | 21 Jul 2026 11:55 |
| Last Modified: | 21 Jul 2026 11:55 |
| URI: | https://norma.ncirl.ie/id/eprint/9478 |
Actions (login required)
 |
View Item |