Multi-Class Sentiment Classifier

01 · The problem

Raw product reviews needed automated sentiment labels for analytics.

02 · Approach

1. 1

   Preprocessed 1,000 reviews: tokenization, stop-word removal, POS-tagged lemmatization via NLTK.

2. 2

   Engineered TF-IDF unigram/bigram features.

3. 3

   Compared Naive Bayes, SVM, and Logistic Regression.

03 · Outcome

All three classifiers reached full test accuracy on this synthetic dataset built from distinct linguistic templates.