Review-Based-Sentiment-Modelling
🏨 Booking.com Review-Based Sentiment Modelling
📘 Overview
This project explores 1,500 customer reviews from Booking.com for La Veranda Hotel in Larnaca, Cyprus, to extract insights into customer sentiment. Using Natural Language Processing (NLP) techniques like sentiment scoring, topic modeling (LDA), and geospatial analysis, we uncover themes that impact guest satisfaction and provide actionable business recommendations.
📌 Objectives
- Classify guest reviews into positive and negative sentiments.
- Identify key topics discussed in both positive and negative reviews.
- Examine geographic patterns in guest satisfaction.
- Explore if cultural differences influence negative feedback.
📂 Dataset
- Source: Public dataset from Kaggle
- Size: 1,500 guest reviews
- Location: La Veranda Hotel, Larnaca, Cyprus
⚙️ Data Preprocessing
- Combined positive and negative reviews into a unified column.
- Removed missing values.
- Cleaned text: punctuation, stopwords, lowercase.
- Standardized date formats.
- Geocoded guest country into coordinates.
📊 Visualizations
🖼️ Fig 1: Preprocessed Text and Word Cloud
This shows the initial word cloud after text cleaning.
🖼️ Fig 2: Embedding Coordinates using Geocoding
Guest countries were mapped to their coordinates for sentiment mapping.
📈 Results and Discussion
📌 3.1 Sentiment Analysis
Sentiment scores were generated using NLP tools to classify reviews.
Summary:
- Positive reviews dominate the dataset.
- Neutral sentiments suggest room for improvement.
- Negative reviews are fewer but focus on cleanliness, noise, and check-in delays.
🔥 Highlights:
- Common praise: “amazing service”, “perfect location”
- Common complaints: “dirty bathrooms”, “noisy rooms”
🖼️ Fig 3: Heat Map of Sentiment Analysis
Sentiment polarity distribution by review.
💡 Business Implications:
- Improve cleaning standards and noise control.
- Enhance staff training during peak seasons.
📌 3.2 Topic Modeling (LDA)
Used Latent Dirichlet Allocation (LDA) to extract recurring themes.
Positive Reviews:
- Topics: Staff service, room comfort, hotel location
- Keywords: “friendly staff”, “spacious room”, “convenience”
Negative Reviews:
- Topics: Cleanliness, delays, room temperature
- Keywords: “dirty bathroom”, “slow service”, “noisy”
🖼️ Fig 4A: Word Cloud - Negative Reviews
🖼️ Fig 4B: Word Cloud - Positive Reviews
💡 Business Implications:
- Emphasize strengths (service, comfort) in marketing.
- Address operational weak points (cleanliness, delays).
📌 3.3 Geographic & Cultural Insights
Sentiment trends analyzed by region using guest country geocoding.
Observations:
- High ratings: Guests from Europe & North America
- Low ratings: Guests from regions with cultural/language differences
🖼️ Fig 5: Geo Map of Guest Sentiment by Country
🖼️ Fig 6: Countries with Higher Negative Sentiment
💡 Business Implications:
- Customize services for guests from low-rated regions.
- Focus on cultural awareness in staff training.
✅ Conclusion
Key actionable takeaways:
- Address cleanliness and noise complaints promptly.
- Showcase service quality in campaigns.
- Understand and cater to cultural expectations.
This project demonstrates the power of NLP and machine learning in generating customer-centric strategies.
📎 Appendix
- LDA Models: Positive vs Negative Topic Extraction
- Keyword Selection: Based on selected reviews for granular insight
- Initial Word Cloud: Created from raw (unprocessed) reviews
- Data Sampler: 70% of dataset used to optimize Orange performance
💻 Code Snapshot
Snapshot of Orange’s visual pipeline for this project.
🧪 Live Orange Workflow
You can explore the full Orange data mining workflow here:
🔗 Download the Orange Workflow (.ows)
📌 Open it in Orange to view the complete visual pipeline.
🚀 Getting Started
To replicate the analysis:
- Clone this repository
- Install Orange
- Open
booking_sentiment_workflow.owsusing Orange - Explore the sentiment scoring, word clouds, and geolocation mappings interactively
📚 Reference
👤 Author
Ramanav Bezborah
🔗 GitHub Profile