Amazon Rating Prediction Project

A product manager wants to evaluate a NEW product in an EXISTING category/brand. Using historical performance data of similar products (same brand tier, category), predict how this new product will perform.

Example Use Case:

• Input: price=$25, category="Electronics/Accessories", brand="GenericBrand", description="Durable silicone case with kickstand"
• Output: "⚠️ 2.8/5 star rating expected - HIGH RISK product"

📊 Dataset

Using Amazon Reviews 2023 dataset from McAuley Lab:

• Source: https://amazon-reviews-2023.github.io/
• Category: Electronics (43.9M reviews, 18.3M users, 1.6M items)
• Sample Size: 30,000 merged records for this project

📁 Data Files Required

Download these files from the website:

1. Electronics.jsonl - Review data (ratings, text, user info)
2. meta_Electronics.jsonl - Product metadata (title, price, features, etc.)

Run file data_extraction.ipynb file to save the new data in csv format with everything you need.

📋 Output Dataset Features

From Product Metadata:

• main_category - Product category
• product_title - Product name
• average_rating - Overall product rating
• rating_number - Number of ratings
• price - Product price in USD
• description - Product description (list format)
• parent_asin - Unique product ID
• details - Product details (contains brand, size, etc.)

From Reviews:

• rating - Individual review rating (Target variable)
• review_title - Review title
• text - Review content
• helpful_vote - Review helpfulness votes

🛠️ Project Pipeline

1. data_cleaning.ipynb Handle missing values and outliers (e.g., drop products with missing price, filter unrealistic values). Normalize text fields (lowercasing, removing special characters, etc.). Save cleaned and merged dataset into a CSV file for downstream use. Output: cleaned_data.csv

2. feature_engineering.ipynb This notebook transforms the cleaned dataset into machine learning–ready features.