Natural Language Processing (NLP)

Project Overview

Project: Sentiment Analysis of Amazon Cell Phone Reviews

Dataset: 20191226-reviews.csv

Problem Statement:

Online reviews play a crucial role in the online shopping experience and significantly influence consumers' purchasing decisions. While humans can easily interpret the sentiment of these reviews, the challenge is to determine whether a computer can perform sentiment analysis with similar accuracy. This project aims to address this challenge using Amazon reviews of cell phones.

Dataset Overview

Size: 9000 user reviews

Attributes: Username, date, rating, and review text

Steps Taken

1. Data Preprocessing:

• Cleaning the Data: Removed any null values and irrelevant data points.

• Text Preprocessing: Tokenized the text, removed punctuation and stop words, and normalized the case to ensure uniformity.

2. Sentiment Analysis with NLTK:

• Natural Language Processing (NLP): Utilized the Natural Language Toolkit (NLTK), an open-source library, for text processing.

3. Sentiment Scoring:

• Stop Words Removal: Eliminated common words that do not contribute to sentiment (e.g., "and", "the").

• Word Sentiment Scoring: Each word in the review was scored based on its sentiment.

• Overall Sentiment Calculation: Combined individual word scores to determine the overall sentiment of the review.

• Sentiment Categories: Classified reviews into positive, negative, and neutral categories based on their scores.

• Compound Scoring: Calculated an overall compound score to determine the general sentiment of the review.

4. Advanced Sentiment Analysis with Hugging Face Transformers:

• Pre-trained Model Usage:

• Model: Loaded the pre-trained RoBERTa model and tokenizer from the Hugging Face transformers library.

• Implementation: Used 'TFRobertaForSequenceClassification', a TensorFlow implementation of RoBERTa for sequence classification tasks.

5. Sentiment Classification:

• Tokenization: Processed the review text using RoBERTa's tokenizer.

• Sequence Classification: Applied the RoBERTa model to classify the sentiment of the review text.

Results and Analysis:

NLTK Sentiment Analysis:

• Distribution of Sentiments: Visualized the distribution of positive, negative, and neutral reviews.

• Accuracy: Evaluated the accuracy of the sentiment analysis by comparing with human-labeled sentiments.

• Insights: Identified key themes and common sentiments in the reviews.

RoBERTa Sentiment Analysis:

• Enhanced Accuracy: Achieved higher accuracy and more nuanced understanding of sentiments compared to traditional methods.

• Comparative Analysis: Compared results from the NLTK approach with those from RoBERTa to highlight improvements.

Learning and Impact

• NLP Techniques: Gained a deep understanding of various NLP techniques and their applications in sentiment analysis.

• Model Implementation: Learned to implement and fine-tune pre-trained models for specific tasks.

• Comparative Analysis: Improved ability to evaluate and compare the performance of different machine learning models.

Tools used: Python, NLP, TensorFlow, HuggingFace, Jupyter

Models: VADER (Valence Aware Dictionary and sEntiment Reasoner), RoBERTa (Robustly optimized BERT approach)