Description:

climateattention-10k classifies if a given sequence is related to climate topics. As a fine-tuned classifier based on climatebert/distilroberta-base-climate-f (Webersinke et al., 2021), it is using the following ClimaText dataset (Varini et al., 2020):

AL-10Ks.tsv : 3000 (58 positives, 2942 negatives)

The training set is highly unbalanced. You might want to check the upscaling version: 'kruthof/climateattention-10k-upscaled'

How to use:

from transformers import AutoTokenizer, pipeline,RobertaForSequenceClassification

tokenizer = AutoTokenizer.from_pretrained("climatebert/distilroberta-base-climate-f")
climateattention = RobertaForSequenceClassification.from_pretrained('kruthof/climateattention-10k',num_labels=2)

ClimateAttention = pipeline("text-classification", model=climateattention, tokenizer=tokenizer)

ClimateAttention('Emissions have increased during the last several months')

>> [{'label': 'Yes', 'score': 0.9993829727172852}]

Performance:

Performance tested on the balanced ClimaText 10K test set, featuring 300 samples (67 positives, 233 negatives) (Varini et al., 2020)

Accuracy	Precision	Recall	F1
0.9633	1	0.8358	0.9106

References:

Varini, F. S., Boyd-Graber, J., Ciaramita, M., & Leippold, M. (2020). ClimaText: A dataset for climate change topic detection. arXiv preprint arXiv:2012.00483.

Webersinke, N., Kraus, M., Bingler, J. A., & Leippold, M. (2021). Climatebert: A pretrained language model for climate-related text. arXiv preprint arXiv:2110.12010.

https://kruthof.github.io