wanghaisheng / lululemon_reviews_nlp_insights

Metis Project 4: Deriving actionable insights from lululemon product reviews using NLP and unsupervised learning. Also deploying a streamlit app the lululemon review curator.

Actionable Insights from Lululemon Reviews

Amanda Cheney
Metis Project 4
November 13, 2020

Motivation

Lululemon customers are known for writing detailed product reviews aimed at giving the company feedback as well as helping other potential customers inform their purchases. However, there are two major impediments to deriving actionable insights from these reviews, first the majority of the data is unstructured (since it's mostly just text) and second, numeric ratings are unable to convey specific information.

Objective

Natural language processing & unsupervised learning exploration of customer reviews of lululemon’s best-selling sports bras to derive actionable insights for product development and management team and develop a recommender system to provide a curated collection of reviews specifically tailored to customer product needs.

Data Sources

9,000+ unique reviews of all 13 of lululemon's best-selling sports bras scraped from lululemon.com using Selenium.

Methods

Spacy for text preprocessing and lemmatization. Feature extraction with TFIDF and CV. Topic Modeling with LSA, LDA and NMF.

Final model uses TFIDF and NMF and selects 6 core topics using scree, cumulative variation explained and t-SNE plots to validate optimal number of topics.

Key Findings

6 core topics across all product reviews.

1. expectations vs experience
2. size/fit
3. support
4. style/design
5. comfort
6. performance

Positive reviews have an equal distribution across all topics while negative reviews are overwhelmingly concerned with product performance, size/fit as well as how the customer’s experience with the product compares with their expectations.

Analysis has value added for product design team as well as customer experience - as demonstrated in recommender app - Lululemon Review Curator.

Lululemon Review Curator app provides prospective customers a curated collection of reviews specifically tailored to their unique product needs in order to save time sifting through hundreds of reviews written by customers with different needs.

Technologies Used

• Gensim
• SpaCy
• t-SNE
• Jupyter Notebook
• Python
• Pandas
• Scikit-learn
• Matplotlib
• Selenium
• Streamlit

Skills Demonstrated

• Natural Language Processing
• Unsupervised Learning
• Dimensionality Reduction
• Topic Modeling
• Visualization
• Web scraping