Investigating Techniques for Aspect-Based Sentiment Analysis in Employee Feedback
Built an aspect‑level sentiment system for employee feedback using BERT/RoBERTa/T5 and GPT variants, with domain adaptation and augmentation. Evaluated using F1 and human judgment.
PythonTransformersscikit‑learnPandasNotebook Experiments
- Extracted aspect–sentiment pairs (e.g., work‑life balance: positive; salary: negative)
- Improved interpretability with aspect taxonomy + example rationales
- Comparing different large language model (LLM) architectures to better understand their progress in fine-grained aspect-based sentiment analysis.