Wellness-centre NLP Analytics Dashboard

Wellness-centre NLP Analytics

A comprehensive analysis of customer feedback to extract actionable insights through advanced Natural Language Processing techniques

27,586

Total Reviews Analysed

5,824

Negative Reviews

600+

Wellness Centres

10

Topic Clusters

Project Overview

This project analyses customer reviews from Google and Trustpilot for Wellness-centre Group to identify key drivers of customer satisfaction and dissatisfaction. By leveraging advanced Natural Language Processing (NLP) techniques, we extract actionable insights to enhance the customer experience across 600+ wellness centres.

Data Sources 12 months of reviews from Google and Trustpilot platforms
Analysis Methods BERTopic, Emotion Analysis, Large Language Models (Falcon-7B), Gensim LDA
Primary Goal Identify key areas for improvement to enhance customer experience

Topic Distribution in Negative Reviews

Emotion Analysis

Top Locations with Negative Reviews

Rank	Location	Google Reviews	Trustpilot Reviews	Total Reviews
1	London Stratford	59	22	81
2	London Enfield	25	23	48
3	London Swiss Cottage	22	15	37
4	London Hayes	17	16	33
5	Bradford Thornbury	19	14	33

Topic Clusters

Showers & Cleanliness 191

Day Pass & Access 122

Equipment & Weights 112

Staff & Rudeness 109

Music & Classes 106

Equipment Maintenance 97

Crowding 79

Methodology & Process

Data Import & Preprocessing

Imported and cleaned review data from Google and Trustpilot sources, preparing it for NLP analysis.

Data Sources: Google_12_months.xlsx and Trustpilot_12_months.xlsx
Cleaning Operations: Removed missing values, cleaned text, tokenised reviews
Data Volume: After cleaning - Google (11,865 reviews) & Trustpilot (15,721 reviews)
Anonymisation: Removed personal identifiers and Wellness-centre specific terms
Preprocessing: Converted to lowercase, removed stopwords and numbers

Initial Data Exploration

Conducted preliminary analysis to understand review distribution and common themes across both platforms.

Unique Locations: Google (456) & Trustpilot (376) with 7 common locations in top negative reviews
Common Locations: London Stratford, London Enfield, London Swiss Cottage among the most frequently reviewed
Word Frequency Analysis: Identified most common terms in both positive and negative reviews
Negative Reviews: Google (2,425) & Trustpilot (3,399) negative reviews identified

Topic Modelling with BERTopic

Applied BERTopic to identify key themes and patterns in negative customer reviews, focusing on common locations.

Model Configuration: Used KMeans clustering with 10 topics
Input: 1,000 negative reviews from locations common to both platforms
Visualisation: Created intertopic distance maps and bar charts of most salient terms
Results: Identified distinct clusters representing specific customer concerns

Location-Specific Analysis

Expanded analysis to top 30 locations to identify geographical patterns and location-specific issues.

Top Locations Analysis: Identified London Stratford, Leicester Walnut Street, and London Enfield as having highest negative reviews
Cross-Platform Comparison: Examined review patterns across Google and Trustpilot
Location-Specific Topics: Applied BERTopic on reviews from top 30 locations
Contrasting Results: Identified differences between overall findings and top-location findings

Emotion Analysis

Used BERT-based emotion classification to identify emotional patterns in reviews, with particular focus on anger.

Model: bhadresh-savani/bert-base-uncased-emotion from Hugging Face
Emotions Analysed: Anger, sadness, joy, fear, love, surprise
Results: Anger was the dominant emotion in negative reviews (Google: 807, Trustpilot: 1,225)
Analysis: Applied BERTopic on anger-dominant reviews to identify key triggers

Advanced Analysis with Falcon-7B

Leveraged large language model capabilities to extract topics and generate actionable insights from negative reviews.

Model: tiiuae/falcon-7b-instruct from Hugging Face
Process: Extracted top 3 topics from each negative review
Topic Clustering: Applied BERTopic to the extracted topics
Action Generation: Generated specific recommendations for each topic cluster

Validation with Gensim LDA

Applied traditional LDA topic modelling as a validation method to compare with BERTopic findings.

Model: Gensim LDA with 10 topics
Input: All negative reviews from Google and Trustpilot
Visualisation: Created interactive topic visualisations with pyLDAvis
Comparison: Confirmed consistency with BERTopic findings, but with less granularity

Key Findings & Insights

Primary Areas of Concern

Across all analytical methods, four consistent areas of customer dissatisfaction emerged:

Equipment Issues: Broken machines, insufficient quantity, poor maintenance
Cleanliness Problems: Dirty showers, changing rooms, and toilets
Staff Behaviour: Rudeness, unresponsiveness, poor customer service
Membership & Billing: Cancellation difficulties, unexpected charges

These issues appeared consistently across Google and Trustpilot reviews, particularly in high-traffic locations.

Location-Specific Patterns

Analysis revealed significant geographical variation in customer concerns:

London Stratford, London Enfield, and London Swiss Cottage appeared consistently among highest negative review locations
Urban centres showed more complaints about crowding and equipment availability
Some locations showed specific issues (e.g., parking problems) not prevalent elsewhere
Top 30 locations showed more balanced feedback with both positive and negative elements

Emotional Triggers

Emotion analysis revealed key factors driving customer anger:

Membership Fee Issues (342 reviews): Highest trigger of anger, relating to billing disputes, cancellation difficulties
Cleanliness Issues (324 reviews): Second highest anger trigger, particularly regarding showers and changing facilities
Equipment Problems (309 reviews): Third highest anger driver, focusing on broken or unavailable equipment
Staff Attitude (295 reviews): Fourth highest, with complaints about rudeness and unprofessionalism

These findings help prioritise areas needing urgent attention to improve customer satisfaction.

Model Comparison

The project employed multiple analytical approaches, each providing complementary insights:

BERTopic: Provided most nuanced understanding of specific customer concerns with contextual relevance
Emotion Analysis: Identified emotional patterns and priority issues driving negative sentiment
Falcon-7B: Generated specific, actionable recommendations addressing identified problems
Gensim LDA: Statistically validated findings from other approaches while providing broader topic distributions

The consistency across methods strengthens confidence in the identified issues and recommendations.

Strategic Recommendations

Facilities & Equipment

Implement Equipment Quality Control: Establish daily equipment inspection protocols with 24-hour repair policy
Upgrade Shower Facilities: Invest in improved water heating systems and increase shower stall cleaning frequency to hourly checks
Address Space Constraints: Redesign floor layouts to optimise equipment placement and improve traffic flow during peak hours
Enhance Cleaning Protocols: Implement visible cleaning schedules with staff signatures for accountability

Staff Training & Management

Customer Service Excellence Programme: Develop comprehensive training for all staff focusing on communication and conflict resolution
Staff Accountability System: Implement staff performance metrics tied to customer feedback with incentives for positive reviews
Technical Training: Ensure staff are properly trained on equipment maintenance and basic troubleshooting
Management Visibility: Schedule regular floor time for managers to directly engage with members

Membership & Billing

Transparent Pricing: Simplify membership structure with clear explanation of all potential charges
Streamlined Cancellation: Implement online cancellation options with immediate confirmation
Improved Access Systems: Upgrade PIN/pass systems to reduce access issues and frustration
Membership Retention Programme: Develop incentives and special offers for long-term members

Monitoring & Continuous Improvement

Real-time Feedback System: Implement QR code-based feedback throughout facilities for immediate issue reporting
Location-specific Focus: Prioritise improvements at high-problem locations (London Stratford, London Enfield)
Ongoing Analysis: Continue NLP monitoring of reviews to track improvement progress
Member Advisory Panel: Create member advisory groups to provide direct input on improvement initiatives

Implementation Priority

Based on analysis of emotional triggers and frequency of complaints, we recommend the following implementation sequence:

First Priority: Membership & Billing improvements (highest anger trigger)
Second Priority: Cleanliness & Shower facilities (second highest anger trigger)
Third Priority: Equipment maintenance and availability
Fourth Priority: Staff training and accountability

This approach targets the most emotionally charged issues first to rapidly improve customer satisfaction.