restaurant-recommender/
├── src/ # Pipeline scripts (01-09)
│ ├── config.py # Central configuration (Not included)
│ ├── 01_yelp_etl.py # Data extraction
│ ├── 02_spatial_etl.py # Spatial clustering
│ ├── 03_sentiment.py # Hybrid VADER + Transformer pipeline
│ ├── 04_sbert.py # Hierarchical SBERT Chunking implementation
│ ├── 05_similarity_pop.py # Semantic similarity matrix computation
│ ├── 06_feature_builder.py # Final feature vector assembly (36-dim)
│ ├── 07_model_training.py # Model training
│ ├── 08_ablation.py # Ablation study
│ └── 09_mmr.py # MMR evaluation
│
├── data/processed/ # Feature datasets
│ ├── features_all.parquet
│ └── restaurant_embeddings.parquet
│
├── models/ # Trained checkpoints
│ ├── ols_model.joblib
│ ├── xgb_model.joblib
│ ├── dnn_model.keras
│ └── scaler.joblib
│
├── figures/ # Visualizations
│
├── requirements.txt
├── run_models.py # Quick-start (Local)
└── run_Models.ipynb # Quick-start (Colab)
# Clone repository
git clone https://github.com/YingWana/Restaurant-Recommender.git
cd restaurant-recommender
# Create virtual environment
python -m venv venv
source venv/bin/activate # Windows: venv\Scripts\activate
# Install dependencies
pip install -r requirements.txt
# Run model training & evaluation
python run_models.py
from google.colab import drive
drive.mount('/content/drive')
%cd /content/drive/MyDrive/capstone
!pip install -q xgboost sentence-transformers vaderSentiment
!python run_models.pyNote: This project was developed and tested using Google Colab with the following configuration:
- Runtime type: Python 3
- Hardware accelerator: NVIDIA T4 GPU
- RAM: High-RAM runtime
- Runtime version: Latest available at execution time
| Category | Features | Count | Description |
|---|---|---|---|
| Semantic | X_sim | 1 | Cosine similarity (two-phase design) |
| Sentiment | X_sent_mean, X_sent_std, X_sent_skew | 3 | Three-moment distribution |
| Popularity | X_pop | 1 | log(review_count + 1) |
| Amenities | TakeOut, Delivery, OutdoorSeating, etc. | 8 | Binary attributes |
| Cuisine | American, Mexican, Italian, etc. | 22 | One-hot encoded |
| Spatial | neighborhood_type | 1 | K-Means cluster |
| Total | 36 |
| Metric | Value |
|---|---|
| Total Restaurants | 10,841 |
| Total Reviews | 2,390,705 |
| Geographic Coverage | PA, FL, TN, AZ, LA |
| Minimum Reviews | 50 per restaurant |
| Train / Val / Test Split | 70% / 15% / 15% |
| Test Set Size | n = 1,627 |
| Reviews > 512 tokens | 78,043 (2.6%) |
Source: Yelp Open Dataset (2024)
# --- Core Data Science ---
numpy==1.26.4
pandas==2.2.3
scipy==1.14.1
# --- Machine Learning ---
scikit-learn==1.5.2
xgboost==2.1.1
# --- Deep Learning ---
# For standard systems:
tensorflow>=2.16.0
# For Apple Silicon, replace with:
# tensorflow-macos==2.16.2
# tensorflow-metal==1.2.0
# --- NLP ---
nltk==3.9.1
vaderSentiment==3.3.2
sentence-transformers==3.0.1
transformers==4.45.2
torch>=2.0.0
# --- Visualization ---
matplotlib==3.9.2
seaborn==0.13.2
# --- Utilities ---
tqdm==4.66.5
joblib>=1.3.0
pyarrow>=14.0.0
Apple Silicon (M1/M2/M3) Installation
pip install tensorflow-macos tensorflow-metal| # | Citation |
|---|---|
| [1] | Hutto & Gilbert (2014). "VADER: A parsimonious rule-based model for sentiment analysis." ICWSM |
| [2] | Reimers & Gurevych (2019). "Sentence-BERT: Sentence embeddings using siamese BERT-networks." EMNLP |
| [3] | Chen & Guestrin (2016). "XGBoost: A scalable tree boosting system." KDD |
| [4] | Abadi et al. (2016). "TensorFlow: A system for large-scale machine learning." OSDI |
| [5] | Carbonell & Goldstein (1998). "MMR diversity-based reranking." SIGIR |
| [6] | Liu (2009). "Learning to rank for information retrieval." FnTIR |
| [7] | Yelp Open Dataset (2024). https://www.yelp.com/dataset |
| [8] | Ng (2021). "Data-Centric AI." DeepLearning.AI |
- Yelp for providing the Open Dataset
- Sentence-Transformers team for the SBERT implementation
- NVIDIA for CUDA/GPU acceleration support
License - see LICENSE for details. Copyright © 2025 Ying Wana. All Rights Reserved.