LeetCode Semantic Problem Recommender

A learning-to-rank recommendation system that recommends the next best LeetCode problem using semantic embeddings and a supervised ranking model. The system uses Sentence-BERT for contextual understanding and LightGBM for ordering problem relevance.

Live Demo: https://leetcode-problem-recommender.streamlit.app/

---

Key Features

Semantic Retrieval (Sentence-BERT)

Generates dense embeddings using all-MiniLM-L6-v2 to capture conceptual similarity between problems, enabling retrieval beyond keyword matching.

Feature Engineering

Each candidate problem is ranked using the following signals:

• Cosine Similarity: Measures semantic closeness.
• Jaccard Similarity: Measures overlap in problem tags.
• Difficulty Distance: Penalizes sudden jumps in difficulty.
• Popularity Score: Based on likes, acceptance rate, and submission statistics.

Learning-to-Rank with LightGBM

A LightGBM ranking model combines all features into a single relevance score, improving recommendation accuracy compared to similarity-only approaches.

---

Tech Stack

• Frontend: Streamlit
• Model Serving: LightGBM, Scikit-Learn
• Embeddings: Sentence-Transformers (HuggingFace)
• Preprocessing: Pandas, NumPy, NLTK
• Similarity Metrics: NumPy cosine similarity

---

Project Structure

├── app/
│   ├── recommender.py    # Core inference logic & feature extraction
│   └── __init__.py
├── data/
│   ├── preprocessed_data.csv  # Cleaned LeetCode dataset
│   └── sbert_embeddings.pkl   # Pre-computed vector embeddings (Cached)
├── models/
│   └── lightgbm_model.pkl     # Trained Ranking Model
├── Data_Pipeline/             # Jupyter Notebooks for EDA & Training
│   ├── leetcode_scraper.ipynb
│   ├── data_preprocessing.ipynb
│   └── lightGBM.ipynb
├── app.py                # Streamlit Application Entry Point
├── requirements.txt      # Python Dependencies
└── README.md             # Documentation