Comparative Study of Word Embedding Representations for Sentiment Classification

This project compares different word embedding techniques and machine learning models for sentiment classification using Amazon product reviews.

We evaluate:

Pretrained embeddings (word2vec-google-news-300)

Domain-specific Word2Vec embeddings trained on Amazon reviews

Across multiple models:

Perceptron

SVM

MLP (2 hidden layers: 50 → 10)

CNN (2 Conv1D layers: 50 → 10 channels)

Both binary (positive vs negative) and ternary (positive, negative, neutral) classification tasks are performed.