Skip to content

README.md

Latest commit

 

History

History
104 lines (77 loc) · 2.53 KB

README.md

File metadata and controls

104 lines (77 loc) · 2.53 KB
Raw

Course Recommend System

An interactive Streamlit-based course recommendation app.
Users can select completed courses, train different recommenders, and compare model behavior inside the UI.

Live Demo

Try the deployed app here:
https://course-recommend-system-v2.streamlit.app/

Features

  • Modern Streamlit interface with interactive course selection (streamlit-aggrid)
  • Multiple recommendation models:
    • Hybrid (Recommended)
    • Course Similarity
    • KNN Collaborative
    • Clustering with PCA
    • Neural Network (Keras)
  • Sidebar controls for model and hyperparameters
  • In-app model descriptions
  • Built-in offline comparison metrics:
    • Hit@K
    • Coverage

Project Structure

Course Recommend System/
├─ recommender_app.py      # Streamlit UI
├─ backend.py              # Model training/prediction logic
├─ ratings.csv             # User-item ratings
├─ sim.csv                 # Course similarity matrix
├─ course_processed.csv    # Processed course metadata
├─ courses_bows.csv        # Bag-of-words/content features
└─ requirements.txt

Installation

1) Create a virtual environment

python -m venv .venv

Windows (PowerShell):

.venv\Scripts\Activate.ps1

macOS/Linux:

source .venv/bin/activate

2) Install dependencies

pip install -r requirements.txt

Run the App

streamlit run recommender_app.py

Usage Flow

  1. Select a model from the sidebar.
  2. Adjust model hyperparameters.
  3. Select completed courses in the main table.
  4. Click Train Selected Model.
  5. Click Recommend New Courses.
  6. Use Model Evaluation to compare models with Hit@K and Coverage.

Model Notes

  • Hybrid: combines content similarity and popularity.
  • Course Similarity: pure item-item similarity approach.
  • KNN Collaborative: collaborative filtering from user-item interactions.
  • Clustering with PCA: dimensionality reduction (PCA) + KMeans clustering.
  • Neural Network: Keras embedding-based rating predictor.

Note: SCORE scales are model-specific.
Compare scores within the same model, not across different models.

Requirements

  • Python 3.10+
  • TensorFlow (CPU mode works)
  • scikit-learn
  • Streamlit

Future Improvements

  • Add missing models (User Profile, NMF, embedding-based regression/classification)
  • Add more metrics (Recall@K, MRR@K, MAP@K)
  • Add hyperparameter tuning and experiment tracking

License

If there is no license file in the repository, contact the project owner before reuse.