The project explores a dataset from IBM and builds various recommender systems to recommend articles to new and existing users.
The main file is a Jupyter Notebook called Recommendations_with_IBM.ipynb. The notebook has four main sections:
- I. Exploratory Data Analysis
- II. Rank Based Recommendations
- III. User-User Based Collaborative Filtering
- V. Matrix Factorization
Various recommendation models was tested and the accuracy of recommendations may not be good as there is a large imbalance in the dataset. Content based recommendations were not included but may provide better recommendations, especially for new users. A/B testing can be used to test the various recommender systems to which ones gives a better click through rate, meaning more relevant recommendations for the users.
Libraries and virtual environment details are located in the Pipfile which can be used with pipenv.
Thanks to Sanjeev Yadav for his article on Medium when I got stuck. Also, Purva Huilgol, for her Medium article on Accuracy vs. F1-Score.
Project based on Udacity Data Science Nanodegree.