AirBnb Topic Modeling (NLP/LDA)
MY ROLES
TIMELINE
2020
TOOLS USED
NLP, LDA, Python, Data Cleaning, Sentiment Analysis, Topic Modeling, Statistical Analysis, Pandas, NLTK, Scikit-learn, Visualization
DESCRIPTION
This project applied Natural Language Processing (NLP) and Latent Dirichlet Allocation (LDA) to analyze Airbnb review data, extracting meaningful topics and patterns. By combining topic modeling with sentiment analysis, the system could automatically categorize feedback into specific areas (cleanliness, location, etc.) while identifying the sentiment (positive/negative) associated with each. This provides hosts with actionable insights about their property's strengths and areas for improvement.
OUTCOMES
- Successfully extracted 15+ distinct topics from hundreds of thousands of Airbnb reviews
- Developed sentiment classifier with 85%+ accuracy for rating prediction
- Created visualization dashboard showing topic distribution and sentiment patterns
FEATURES
Automated topic extraction, sentiment classification by category, statistical correlation between topics and ratings, interactive visualization of results, data preprocessing pipeline
CHALLENGES
Handling unstructured text data with varied writing styles and languages. Determining the optimal number of topics that are both statistically valid and semantically meaningful. Addressing class imbalance issues in sentiment analysis as most reviews tend to be positive.
APPROACH
Implemented text preprocessing pipeline for cleaning review data. Used LDA for topic modeling with coherence score optimization to determine ideal topic count. Applied VADER sentiment analysis to classify comments, then linked sentiment to specific extracted topics. Visualized results using interactive dashboards to show hosts specific areas of strength and concern.