sustainable-robotics/robotics-arxiv-sustainability-classification
ArXiv Robotics Sustainability Dataset This dataset compiles sustainability annotations for robotics papers from arXiv metadata files in this repository. Dataset Structure Each record is based on an input paper object, with the following transformation rules: id - url to the paper on arXiv title - title of the paper authors - list of author names published - publication date in ISO format (YYYY-MM-DD) Added classification fields: sdg_analysis - original text from… See the full description on the dataset page: https://huggingface.co/datasets/sustainable-robotics/robotics-arxiv-sustainability-classification.
ArXiv Robotics Sustainability Dataset
This dataset compiles sustainability annotations for robotics papers from arXiv metadata files in this repository.
Dataset Structure
Each record is based on an input paper object, with the following transformation rules:
id- url to the paper on arXivtitle- title of the paperauthors- list of author namespublished- publication date in ISO format (YYYY-MM-DD)- Added classification fields:
sdg_analysis- original text from thesdg_analysisfield in the input paperaligned_with_sdgs: list of SDGs extracted from section1ofsdg_analysismotivated_by_sdgs: list of SDGs extracted from section2ofsdg_analysismentions_social_impact: boolean flag indicating if the paper mentions social impact in thesdg_analysismentions_ecological_impact: boolean flag indicating if the paper mentions ecological impact in thesdg_analysismentions_sustainability_impact: boolean flag indicating if the paper mentions sustainability impact in thesdg_analysismentions_un_sdgs: boolean flag indicating if the paper mentions UN SDGs in thesdg_analysis
About the dataset
This dataset is generated by extracting 49k+ papers from the arXiv database in the cs.RO category, staring from 2015 till early 2026.
Each paper is classified based on the presence of sustainability-related content in the sdg_analysis field, with respect to the UN Sustainable Development Goals (SDGs). The dataset includes boolean flags indicating whether the paper has social, ecological, sustainability, or UN SDG impacts, as well as lists of specific SDGs that the paper is aligned with or motivated by.
The data was generated by passing the full text of each paper through a custom prompt to a LLM (Deepseek v3) to extract the relevant information, which was then parsed and structured into the final dataset format.
A more comprehensive notes on the dataset and the analysis that motivated the dataset creation can be found in the article Robotics research should think (and do) more about sustainability!.
The dataset is not perfect, it is not created by an expert in sustainability of any kind. It is an approximation of the sustainability-related content in the papers, based on the LLM's understanding of the text. It is meant to be a starting point for further analysis and research, and not a definitive classification of the papers.
The code with which this dataset was generated with can be found on github: https://github.com/huggingface/arxiv-robotics-sustainability-classification
