A data analysis project using IBM Granite LLM to extract sentiment-based insights from user reviews across three Disneyland parks: California, Paris, and Hong Kong. This project combines exploratory data analysis (EDA) with advanced NLP through large language models (LLMs) to summarize reviews and uncover patterns in visitor experiences.
This project aims to extract insights from thousands of Disneyland reviews sourced from TripAdvisor. By combining data visualization with AI-based text summarization, we seek to:
- Understand visitor satisfaction and pain points
- Compare user sentiment across Disneyland California, Paris, and Hong Kong
- Showcase the value of LLMs (IBM Granite) in real-world review analysis
The dataset is sourced from Kaggle - Disneyland Reviews and includes:
Review_Text: Review contentRating: Score (1–5)Branch: Park location (California, Paris, Hong Kong)Reviewer_Location: Country of originYear_Visited: Visit year
-
Rating Distribution
Majority of reviews have a 5-star rating, indicating high satisfaction. -
Review Volume by Park
California has the highest number of reviews, followed by Paris and Hong Kong. -
Reviewer Location
Top reviewers come from the US and UK, reflecting travel and proximity trends. -
Keyword Themes
- Positive reviews:
ride,kid,fun,attraction - Negative reviews:
queue,staff,wait,price
- Positive reviews:
-
Time Trends
Review activity peaked around 2015 and gradually declined, possibly influenced by global events or park developments.
The IBM Granite LLM (Large Language Model) was used for:
-
Summarization
Generating concise summaries of reviews per park. -
Insight Extraction
Identifying top 3 recurring praises and complaints per location. -
Sentiment Classification
Create positive, neutral and negative sentiments from each review.
- Handles unstructured text effectively
- Understands context, sarcasm, and nuance
- Saves time vs manual or rule-based sentiment tagging
From the following reviews, identify:
1. The top 3 most frequent complaints
2. The top 3 most frequent praises
Praises:
- Exceptional customer service
- Magical atmosphere and iconic rides
- FastPass system improves experience
Complaints:
- Expensive food and merchandise
- Long wait times for rides
- Overcrowding during peak seasons
Praises:
- Well-organized and clean park
- Immersive Disney-themed atmosphere
- FastPass feature appreciated
Complaints:
- Long queues and ride closures
- High prices, even outside the park
- Some staff perceived as indifferent
Praises:
- Magical and family-friendly environment
- Easy access via public transport, great shows
- Shorter lines during weekdays
Complaints:
- Smaller size than other Disney parks
- Mixed quality of food relative to price
- Long queues during peak hours
Conclusion
- Overall sentiment is positive across all parks.
- FastPass, customer service, and theming are consistently appreciated.
- Common challenges include price, queue time, and seasonal overcrowding.
- Each park has distinct strengths and operational weaknesses.
Recommendations For Disneyland Management:
- Improve Queue Management
- Enhance FastPass or introduce app-based queueing
- Reassess Pricing Strategy
- Offer budget-friendly dining or bundling options
- Staff Training & Consistency
- Emphasize hospitality, particularly in Paris and Hong Kong
For Visitors:
- Visit Off-Season
- November and weekdays offer better comfort
- Use FastPass Strategically
- Prioritize popular attractions and book in advance
- Plan & Budget Accordingly
- Expect premium prices for food and experiences
- Python (Pandas, Seaborn, Matplotlib, WordCloud)
- IBM Granite LLM via Replicate API
- Google Colab Notebook
- GitHub