KINDLE REVIEW ANALYSIS MODEL

This project builds a sentiment analysis model to classify Amazon Kindle reviews as positive or negative based on the text content. It uses NLP preprocessing, Bag-of-Words (BoW) and TF-IDF vectorization, and Gaussian Naive Bayes models for classification.

PROJECT OVERVIEW-------------------------------------------------------------

Goal: Predict sentiment of Kindle reviews from text data.

Dataset: 12,000 Kindle reviews with ratings from 1–5.

Classes:

Positive → rating ≥ 3

Negative → rating < 3

The notebook demonstrates:

Data cleaning and preprocessing

Feature extraction using BoW and TF-IDF

Model training with Gaussian Naive Bayes

Evaluation with accuracy, precision, recall, F1-score, and confusion matrix

DATASET------------------------------------------------------------------------

The dataset is sourced from Krishnaik’s GitHub repository and contains:

Column Description reviewText Text of the Kindle review rating Star rating (1–5) reviewerID Reviewer identifier reviewTime Date of review summary Short review summary helpful Helpfulness votes

For GitHub, only the relevant columns reviewText and rating are used.

DATA PREPROCESSING----------------------------------------------------------------

Lowercasing all text

Removing punctuation, numbers, and special characters

Removing stopwords

Removing HTML tags and URLs

Lemmatization using NLTK's WordNetLemmatizer

Converting ratings into binary sentiment:

Positive (1): rating ≥ 3

Negative (0): rating < 3

FEATURE EXTRACTION--------------------------------------------------------------------

Bag-of-Words (BoW): Represents text as token counts

TF-IDF Vectorization: Represents text with term importance

Both methods are applied to the training and test sets for model training.

MODEL TRAINING-------------------------------------------------------------------------

Algorithm: Gaussian Naive Bayes

Inputs: BoW and TF-IDF features

Outputs: Binary sentiment classification (0 = negative, 1 = positive)

Training is done on 80% of the dataset, and testing on 20%.

EVALUATION-----------------------------------------------------------------------------

The model performance is evaluated using:

Accuracy

Precision

Recall

F1-score

Confusion Matrix