2025 · NLP
EV Sentiment Analysis Using Transformer Models
NLP Deep Learning Transformers
EV Sentiment Analysis Using Transformer Models
University of Maryland | Sep 2024 – May 2025
Python Hugging Face BERT/RoBERTa PyTorch PRAW API
Project Overview
Developed comprehensive sentiment analysis pipeline to understand public perception of electric vehicles using transformer-based NLP models on large-scale social media data.
Key Contributions
Data Collection: Collected and preprocessed 1.1M+ Reddit posts spanning 2019–2024 using PRAW API with temporal stratification across EV-related subreddits
Model Development: Fine-tuned BERT, RoBERTa, and DistilBERT on 5,000+ manually labeled posts for domain-specific EV sentiment classification, achieving 91.6% accuracy (F1: 0.89)
Trend Analysis: Conducted longitudinal sentiment analysis correlating public perception shifts with EV policy milestones (e.g., IRA, state incentives), identifying 23% sentiment improvement post-2022
Academic Presentation: Presented findings at AAPOR 2025 conference, demonstrating novel integration of social media analysis with traditional survey methodology
Technologies Used
- Languages: Python
- ML/NLP: PyTorch, Hugging Face Transformers
- Models: BERT, RoBERTa, DistilBERT
- Data Collection: PRAW Reddit API
Key Results
| Metric | Value |
|---|---|
| Posts Analyzed | 1.1M+ |
| Model Accuracy | 91.6% |
| F1 Score | 0.89 |
| Time Period | 2019-2024 |
| Labeled Training Set | 5,000+ posts |
Publication
This research was presented at the 80th Annual AAPOR Conference (May 2025) in Arlington, VA.