Sentiment-Driven E-commerce Optimization: Amazon Review

Sentiment-Driven E-commerce Optimization: Amazon Review Analysis & Rating Prediction.

Project Overview ✅

This project serves as a machine learning proof of concept designed to transform raw Amazon product reviews into actionable business insights. By automating the prediction of review ratings and analyzing customer sentiment, the system enables brands to optimize product listings, proactively address customer pain points, and drive higher conversion rates.

Process ✅

I developed an end-to-end pipeline covering data acquisition, complex text processing, and model deployment. Automated Data Scraping, Integrated the Apify API to extract real-time customer feedback directly from Amazon product URLs. I configured the scraper to handle up to 100 reviews per run, capturing critical metadata including rating scores, review descriptions, and verified purchase status.

Data Refinement & Feature Engineering:

Cleaned a dataset of approximately 1,944 reviews by removing noise (punctuation/symbols) and stop words using NLTK. I implemented TF-IDF Vectorization to convert text into numerical features and applied SMOTE (Synthetic Minority Over-sampling Technique) to address class imbalance, ensuring the model could accurately predict rare negative reviews.

Model Benchmarking ✅

Developed and compared three distinct architectures to identify the most robust predictor:

Web App Deployment: Built a dedicated Streamlit dashboard that allows non-technical stakeholders to input raw review text and receive instant rating predictions with confidence scores.

Technical Stack✅

Languages & Tools: Python, Apify Client.

ML & NLP Libraries: Scikit-learn (SVC, Naive Bayes, MLP), NLTK (Tokenization, Stopwords), Imbalanced-learn (SMOTE).

Deployment: Streamlit, Joblib (Model Serialization).

Visualization: Plotly, WordCloud, Matplotlib.

Key Results ✅

Achieved a peak accuracy of 95.27% using the Neural Network model, with the SVC model following closely at 94.46%.

Developed sentiment-based feedback loops within the app: high ratings (4-5 stars) trigger positive marketing recommendations, while low ratings (1-2 stars) alert teams to address product issues like battery life or build quality.

Enabled real-time competitive analysis by providing a user-friendly interface for cross-functional marketing and product development teams to audit customer sentiment at scale.