Investigating Public Concerns About AI

A group project for the Natural Language Processing course, studying how artificial intelligence is discussed and framed in public discourse.

We assembled a dataset of more than 20,000 tweets and 500 news articles, then fine-tuned Twitter-RoBERTa and DistilBERT for sentiment classification, reaching roughly 70% F1-score on the detection of negative sentiment. To move from sentiment to substance, we applied topic modelling (NMF and LDA) to extract the structured themes that recur in media framing of AI.

Methods: Twitter-RoBERTa, DistilBERT, NMF, LDA, Hugging Face.