Identifying AI’s Environmental Risks: Using NLP to Analyzing Public Consultation Feedback on the AI Act

This repository contains code and resources for analyzing public consultation feedback on the AI Act, focusing on environmental and climate-related concerns identified by stakeholders. The analysis involves data scraping, cleaning, preprocessing, and textual analysis to extract insights from feedback submitted by various stakeholders.

The workflow follows key Natural Language Processing (NLP) techniques, data scraping, sentence extraction, keyword filtering, and textual analysis to identify and analyze recurring themes and patterns in the feedback data.

The research note can be consulted here.

---

Repository Structure

data/

• Contains raw and processed datasets.

notebooks/

01_scrapping.ipynb

• Scrapes public consultation feedback data from the European Commission's website.

02_data_cleaning.ipynb

• Cleans and standardizes the scraped text data.

03_preprocessing.ipynb

• Performs Exploratory Data Analysis, extracts environmental risk mentions, generates word clouds, trains a Word2Vec model, and visualizes word embeddings

04_analysis.ipynb

• Applies sentence embeddings, UMAP dimensionality reduction, and KMeans clustering to group environmental risk feedback into semantic clusters.
• Summarizes cluster content using a pre-trained summarization model (BART) to highlight key insights

output/

• Stores outputs generated from the analysis notebooks,

---

Technologies Used

• Data Collection and Scraping: selenium
• Data Cleaning and Preprocessing: pandas, numpy, re
• Natural Language Processing (NLP): nltk, Word2Vec, transformers, langdetect, langid
• Exploratory Data Analysis (EDA): CountVectorizer
• Machine Learning Models:
  • Clustering: KMeans
  • Dimensionality Reduction: umap-learn
  • Sentence Embeddings: Word2Vec, transformers
  • Summarization: torch, transformers
• Visualization: matplotlib, seaborn, wordcloud