Pyhton_Sentiment_Analyis_Cust_Segmentation

Customer Behavior Analysis using Python (Sentiment Analysis + RFM Segmentation)

Motivation

Understanding customer behavior is vital for any e-commerce business aiming to improve user experience, increase retention, and boost sales. While customer reviews provide emotional feedback, transactional data reveals behavioral patterns. This project combines both:

Sentiment Analysis to assess customer emotions from reviews
Customer Segmentation (RFM + K-Means) to group customers based on purchasing behavior

By integrating these perspectives, we deliver actionable insights for improving product offerings, delivery experience, and personalized marketing.

Dataset

The dataset is a blend of transactional and textual feedback, consisting of the following key features:

Customer_ID
Order_ID
Order_Date, Delivery_Date
Customer_Review (Review Text)
Delivery_Status
Shipping_Partner
City, State
Quantity, Unit Price, Total

Source: Google Analytics API and customer review systems
Time Period: Dec 2023 – May 2025

Project verview

We approached this analysis in two major parts:

Sentiment Analysis with VADER
To automatically classify thousands of customer reviews into detailed emotional categories.
Customer Segmentation with RFM & K-Means
To profile customers based on their purchase recency, frequency, and monetary value.

Together, these methods offer a 360° view of customer behavior — what they feel and how they act.

Methodology

Sentiment Analysis (NLP-VADER)

Used SentimentIntensityAnalyzer from nltk.sentiment.vader
Extracted four sentiment scores
pos (positive), neu (neutral), neg (negative), compound (overall)
Applied custom thresholds on compound score to classify reviews into:
Positive, Mixed Positive, Neutral, Mixed Negative, Negative

Example

“Nice product but delivery was delayed” → Mixed Positive
“Good features, poor customer service” → Mixed Negative

Visualizations

Polarity distribution

Max Score 0.8047
Min Score -0.6369

We used the compound score from VADER to classify each review into five sentiment categories:

Positive: compound between 0.5164 and 0.8047
Mixed Positive: compound between 0.2281 and 0.5164
Neutral: compound between -0.0603 and 0.2281
Mixed Negative: compound between -0.3486 and -0.0603
Negative: compound less than -0.3486

Sentiment category bar chart

Bar Chart: Count of reviews in each sentiment category (Positive, Mixed Positive, Neutral, Mixed Negative, Negative).
Helps identify which sentiments are most common among customers

Customer Segmentaion (RFM + K-Means)</b>

Engineered RFM metrics:

Recency: Days since last purchase
Frequency: Count of orders
Monetary: Total spend
Scored customers using quantiles
Standardized values and applied K-Means Clustering
Used Elbow Method and Silhouette Score to determine optimal k
Labeled segments:
Top Customers
Average Customers
Lapsed Customers

Visualizations

RFM score

Cluster distribution

Standardized RFM values using StandardScaler
Applied K-Means Clustering
Chose optimal k using Elbow Method and Silhouette Score

Segment-wise count chart

Cluster distribution using Seaborn
Bar charts to interpret segments
Labeled clusters (“Top Customers”, “Average Customers”, “Lapsed Customers”)

Tools & Libraries

Sentiment Analysis	Customer Segmentation
Python	Python
NLTK (VADER)	Pandas, NumPy
Pandas	Scikit-learn (KMeans)
Matplotlib, Seaborn	Matplotlib, Seaborn
Jupyter Notebook	Jupyter Notebook

Key Outcomes

Classified 4,356 reviews into 5 detailed sentiment categories using custom VADER logic
Identified major pain points in delivery and service through textual analysis
Segmented customers into high-, medium-, and low-value groups
Informed campaign strategies to re-engage lapsed customers and retain top spenders

Conclusion

This combined project showcases how textual feedback and purchase behavior can be jointly analyzed to drive customer-centric decisions. The sentiment layer surfaces emotional trends, while RFM clustering uncovers structural segments in the customer base.