Comparisons Between Open-LLMs For Fake News Detection
Bibliographic record
Abstract
The spread of fake news on social media is a growing concern with significant negative societal impacts. To address this challenge, researchers have proposed various methods and algorithms for fake news detection. Recently, large language models (LLMs) have emerged as promising tools for both detecting and explaining fake news. This paper presents results from classification and explanation of true and fake news using three LLMs—Mistral, LLaMA3, and LLaMA4—applied through zero-shot prompting via Together.ai. The models analyzed 1,228 claims from the Fin-Fact dataset, evenly divided between real and fake news. Our findings are threefold. First, none of the three LLMs achieved strong classification performance, with LLaMA4 recording F1-score of 0.59 for fake news and 0.51 for true news. Second, in terms of explanation quality, LLaMA4 outperformed the other models, achieving the highest METEOR score (0.072), also obtained the lowest negative COMET score (–0.260) demonstrating improvement over its predecessor, LLaMA3. Third, all models consistently performed better on fake news than on true news, a counterintuitive outcome, since fake news is generally considered more challenging to interpret. This consistency across metrics, METEOR, BERTScore F1 and COMET, reinforces the validity of this observation. The relatively high BERTScore F1 (0.6+) indicated that all models produced semantically consistent explanations. However, both METEOR and COMET scores were very low relative to their respective scales, suggesting that the LLMs’ justifications differed in wording and judgment from those in Fin-Fact. Those metrics are designed for short and precise texts and are unsuitable for evaluating the long narrative justifications present in the Fin-Fact dataset.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.011 | 0.041 |
| Meta-epidemiology (narrow) | 0.003 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.006 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.004 | 0.008 |
| Open science | 0.002 | 0.003 |
| Research integrity | 0.004 | 0.003 |
| Insufficient payload (model declined to judge) | 0.004 | 0.003 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".