Product Reviews Analysis

Goal:

To understand the following about Product Reviews - what are most frequently used words, distribution of the length of the reviews, understanding fake reviews, spelling errors, predicting product rating using the reviews, etc.

Data Description:

• [kaggle dataset] (https://www.kaggle.com/datafiniti/grammar-and-online-product-reviews)

• 71000+ reviews for 1000 products

Features

• id
• reviewsText
• reviewTitle
• reviewRating
• purchasedProduct
• recommendProduct
• brand / manufacturer
• categories
• pack_size
• descriptions
• dateAdded

WORKFLOW

• EDA (data cleaning)
• NLP (tokenization, removing stop words, stemming)
• TOPIC MODELING (NMF with TF-IDF Vectorizer)
• RECOMMENDATION SYSTEM (content based filtering using the terms from reviews using cosine similarity matrix)
• CLASSIFICATION SYSTEM (Logostic Regression and Decision Tree Classifier)

Recommendation System example -

Input - 'skin cream'

Recommendations:

• 'Olay Total Effects Daily Moisturizer
• '7-In-1 Anti-Aging, 0.5oz'
• 'Olay Regenerist Deep Hydration Regenerating Cream'
• "L'oreal Paris Revitalift Triple Power Deep-Acting Moisturizer"
• 'CeraVe SA Renewing Cream'
• 'Shea Moisture African Black Soap Problem Skin Facial Mask'
• 'Tree Hut Shea Body Butters, Coconut Lime, 7 oz'
• 'Nivea Extended Moisture Body Lotion

I have done Dimensionality Reduction using PCA, and also tried K-Means clustering.