Analysis of Hotel Reviews: Transforming Feedback into Actionable Insights

Project Overview

This project develops a robust, automated data analysis workflow to transform unstructured hotel reviews into a structured business intelligence dashboard. The primary goal is to provide hotel management with precise, data-driven recommendations to enhance guest satisfaction and operational efficiency, ultimately turning raw feedback into a strategic asset.

Key Features

This project is more than a simple script; it's an end-to-end analysis system built with professional practices in mind:

• Hybrid Analysis System: Combines high-precision Rule-Based Logic (for topic extraction) with the contextual power of Large Language Models (for sentiment and summarization).
• Robust Multi-Model Engine: Features a smart fallback strategy, automatically switching to a secondary model if the primary one fails, ensuring system stability.
• Encapsulated Workflow: A structured Python Class encapsulates the entire process, making the system modular and easy to understand.
• Strategic Visualization: The final output is a Priority Matrix Dashboard, a powerful tool for managers to instantly identify what's working, what's broken, and where the hidden opportunities lie.

A Note on Dataset Sampling

To ensure efficiency during development and to demonstrate the system's full end-to-end functionality within a standard computing environment (like Google Colab), this analysis was performed on a representative random sample of 200 reviews from the full dataset. The system is designed to be fully scalable to the entire dataset given sufficient computational resources.

Summary of Findings

The analysis of the 200 sample reviews provided the following strategic insights:

• Core Strengths: Location, Hotel Facilities, and Cleanliness are the hotel's most consistently praised assets.
• Critical Issues: Room & Amenities and Staff & Service are the most frequent drivers of guest complaints and require immediate attention.
• Hidden Risk: The Check-in/Check-out Process, despite low frequency, has the lowest sentiment score, marking it as a significant point of guest frustration.
• Opportunities: Value for Money and Noise Level are viewed favorably and could be developed into new strengths.

Tech Stack

• Language: Python
• Core Libraries: Pandas, Matplotlib, Seaborn, adjustText
• NLP & Modeling: Hugging Face Transformers (nlptown/bert-base-multilingual-uncased-sentiment, facebook/bart-large-cnn), PyTorch