Public Sentiment Analysis of the Iran–United States Conflict on BBC News YouTube Comments Using DistilBERT

Abstract

The rapid proliferation of social media platforms, particularly YouTube, has transformed public commentary on geopolitical events into a rich and accessible data source for sentiment analysis. This study examines public sentiment expressed in YouTube comments on BBC News coverage of the Iran–United States conflict, employing DistilBERT—a distilled variant of BERT—as the primary classification model. A corpus of 12,847 comments was collected via the YouTube Data API v3 across a period spanning January to December 2024, covering major escalation events including drone strikes, diplomatic breakdowns, and proxy confrontations. The dataset was preprocessed through noise removal, tokenisation, and label assignment using a lexicon-assisted annotation pipeline. DistilBERT was fine-tuned on the labelled corpus and evaluated against established baselines including Naive Bayes, Support Vector Machine (SVM), and standard BERT. Experimental results demonstrate that DistilBERT achieves an F1-score of 91.73%, surpassing SVM (84.12%), Naive Bayes (76.45%), and approaching standard BERT (92.18%) while requiring 40% fewer parameters and exhibiting substantially lower inference latency. Sentiment distribution revealed that negative sentiment dominated (58.3%), followed by neutral (27.4%) and positive (14.3%), with pronounced negativity spikes correlated with military escalation events. The findings contribute to the growing body of research on computationally efficient transformer-based sentiment analysis in politically sensitive domains and offer implications for media analytics, public diplomacy monitoring, and conflict communication research.