Nano-ESG: Extracting Corporate Sustainability Information from News Data

This is the supplementary material for the paper Nano-ESG: Extracting Corporate Sustainability Information from News Data.

In order to run the notebooks, we recommend creating a new environment with python=3.10 . Then, you can install the packages in the requirements.txt file using

pip install -r requirements.txt

In the supplementary material, you will find

1. The data
   • evaluation_data: The annotated datasets as described in Section 5 of the paper
   • full_data: The full dataset as described in Section 4 of the paper
2. The Evaluation Guidelines (Note that the guidelines were originally in German and translated to English by us for this paper):
   • summary_evaluation.md: Guidelines for the first evaluation task
   • sentiment_aspect_evaluation.md: Guidelines for the second evaluation task
3. The prompts given to the LLMs:
   • first_filter_prompt.md: The prompt for the first LLM filter step (Section 3.4)
   • final_prompt.md: The final prompt to extract the ESG information from each news article (Section 3.6)
4. Notebooks to replicate results in the paper:
   • evaluate_task1.ipynb: This notebook replicates the results we present in Section 5.2
   • evaluate_task2.ipynb: This notebook replicates the results we present in Section 5.3 and Section 6.2
   • investigate_data.ipynb: This notebook replicates the results we present in Section 4 and Section 6.1
   • explore_topics.ipynb: This notebook replicates the results we present in Section 6.3. In addition, it is possible to explore other interesting topics.