ysakhale/cmu-content-based-recommender

Content-Based Recommendation System for CMU Landmarks

Model Description

This is a trained-from-scratch content-based recommendation system designed to recommend Carnegie Mellon University landmarks based on user preferences. The model learns feature representations from landmark characteristics and uses cosine similarity to find similar landmarks.

Model Details

Model Type

• Architecture: Content-based filtering with feature engineering
• Training: Trained from scratch on CMU landmarks dataset
• Input: Landmark features (rating, classes, location, dwell time, indoor/outdoor)
• Output: Similarity scores for landmark recommendations

Training Data

• Dataset: 100+ manually curated CMU landmarks
• Features: Rating, classes, geographic coordinates, dwell time, indoor/outdoor classification
• Preprocessing: StandardScaler normalization, multi-hot encoding for classes

Training Procedure

• Feature extraction from landmark metadata
• StandardScaler normalization of numerical features
• Multi-hot encoding for categorical classes
• Cosine similarity computation for recommendations

Intended Use

Primary Use Cases

• Recommending CMU landmarks based on user preferences
• Finding similar landmarks to user-selected favorites
• Personalized campus exploration planning

Out-of-Scope Use Cases

• Recommending landmarks outside CMU campus
• Predicting user ratings or reviews
• Real-time location-based recommendations

Performance Metrics

• Recommendation Quality: High similarity scores (0.7-0.9) for relevant landmarks
• Diversity: Incorporates diversity weighting to avoid over-concentration
• User Satisfaction: Optimized for user preference alignment

Limitations and Bias

• Geographic Scope: Limited to CMU campus landmarks only
• Static Data: Based on current landmark database, may not reflect real-time changes
• User Preference Learning: Does not learn from user interaction history

Ethical Considerations

• Data Privacy: No personal user data collected
• Fairness: Recommendations based on objective landmark features
• Transparency: Feature importance and similarity scores are explainable

How to Use

from model import ContentBasedRecommender, load_model_from_data

# Load model from landmarks data
recommender = load_model_from_data('data/landmarks.json')

# Get recommendations
recommendations = recommender.recommend(
    selected_classes=['Culture', 'Research'],
    indoor_pref='indoor',
    min_rating=4.0,
    diversity_weight=0.6,
    top_k=10
)

# Print top recommendations
for landmark_id, score in recommendations:
    print(f"{landmark_id}: {score:.3f}")

Model Files

• model.py: Main model implementation
• README.md: This model card

Citation

@misc{cmu-explorer-recommender,
  title={Content-Based Recommendation System for CMU Landmarks},
  author={Yash Sakhale, Faiyaz Azam},
  year={2025},
  url={https://huggingface.co/spaces/ysakhale/Tartan-Explore}
}

Model Card Contact

For questions about this model, please refer to the CMU Explorer Space.