LSTM Sentiment Analysis
Overview
Fine-tuned an LSTM sentiment classifier on RateMyProfessor reviews, comparing model performance with and without data augmentation techniques including synonym replacement, random deletion, and back-translation.
Details
The task predicts sentiment scores from 1 to 5 (in 0.5-point intervals) from review text. Augmentation functions transform the training set to increase linguistic diversity and improve generalization in lower-resource settings.