GreenTruth Claim Detector

Binary classifier flagging whether a sentence is an environmental claim. Text stage of GreenTruth, which checks environmental claims against Earth-observation evidence.

Intended use

Find checkable environmental claims in corporate disclosures, before evidence-grounded verification. Detecting a claim is not judging it.

Training

Fine-tuned climatebert/distilroberta-base-climate-f on the environmental_claims dataset (Stammbach et al., ACL 2023): 3 epochs, lr 2e-5, batch 16, max length 256.

Evaluation

See the notebook run for test precision/recall/macro-F1/accuracy. The paper reports macro-F1 around 85% for comparable models.

Limitations & ethics

English, listed-company text; domain shift is likely. Detects a claim, not its truth. Inherits CC BY-NC-SA 4.0 (non-commercial). Do not use to label a company deceptive.

Dataset

https://huggingface.co/datasets/climatebert/environmental_claims

Downloads last month
83
Safetensors
Model size
82.3M params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Rahilgh/greentruth-claim-detector

Finetuned
(36)
this model

Dataset used to train Rahilgh/greentruth-claim-detector