In this work, we used implicit ratings and an auto-encoder with a modified cost function to make a GitHub Recommender System. First, we collect the data, construct the confidence and prediction matrices based on implicit rating schemes. Finally, we train an auto-encoder with a modified cost function and test the trained model using Recall metric.
report.pdf- gives detailed methodology and results of the project.
data_curation- contains all the scripts used for dataset creation.
Final Report- contains the .tex files of the
autoencoder.py- contains the autoencoder code.