This repo demonstrates two approaches to sentiment analysis, executed with R and Python.

File: NLP_sentiment_analysis.R

This analysis uses child speech from the Child Language Data Exchange System (CHILDES), processing 3.7 million words from real-world conversations to provide insights into how children's speech conveys positive and negative emotions.

Overview:

• Use the childesr package to access the CHILDES TalkBank database.
• Download and cache the CHILDES English-North America corpus (3.7M+ tokens).
• Apply stemming and remove stop words for dimensionality reduction.
• Perform lexicon-based sentiment analysis with the Bing sentiment lexicon.
• Identify the top 20 most frequent positive and negative words in children's speech.
• Use ggplot2 to create a publication-ready visualization.

File: NLP_sentiment_analysis_RoBERTa.ipynb

This analysis uses deep learning for text classification using RoBERTa (Robustly Optimized BERT Approach), a transformer model with 125 million parameters. The model is fine-tuned on the NLTK movie_reviews corpus to classify reviews as positive or negative. The current, quick-and-dirty method delivers 90%+ accuracy, precision, recall, and F1-score after 3 training epochs.

Overview:

• Load the NLTK movie_reviews corpus and the pre-trained RoBERTa model.
• Tokenize text and convert tokenized text to PyTorch tensors for batching.
• Configure training arguments such as number of epochs.
• Evaluate performance with accuracy, precision, recall, and F1-score metrics.
• Use plotnine (ggplot2 for Python) to visualize accuracy by training epoch.