This model is a fine-tuned version of the ClimateBERT (distilroberta-base-climate-f) model (Webersinke et al., 2021) and created as part of a Bachelor Thesis at the University of St.Gallen (HSG).

The use-case of the FossilBERT model is the identification of climate-related tweets that try to "downplay the severity and certainty of climate-change related risks". The initial climate-classification of the tweets is based on another fine-tuned version of ClimateBERT on the "ClimaText" dataset provided by Varini et al. (2020).

The fine-tuning procedure involved a training dataset of 2933 hand-labeled tweets (1458 "downplaying" / 1475 "underscoring") of five polarizing participants of the climate change discourse as well as tweets from the American Petroleum Institute and Greenpeace.

Full Credits of the underlying ClimateBERT model belong to: 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://doi.org/10.48550/arXiv.2110.12010

ClimaText source: Francesco S. Varini and Jordan Boyd-Graber and Massimiliano Ciaramita and Markus Leippold (2020). ClimaText: A Dataset for Climate Change Topic Detection, In: Tackling Climate Change with Machine Learning workshop at NeurIPS 2020, Online, 11 December 2020 - 11 December 2020.