cpurta / adclickprediction

Notebooks for the Outbrain Click Prediction competition

AdClickPrediciton

This is my implementation of some techniques used to predict the probability that an ad will be clicked for the Outbrain Click Prediction Competition. Some of the methods includes a statistical method based on the mean of ads clicked / ads served. One uses the FTRL-Proximal algorithm as described in this paper. Also I have been researching methods that have won past ad-click prediciton competitions such as Field Aware Factorization Machines as described by the team that one the Criteo Ad Click Prediction competition. Their paper can be found here.

Notebooks

If you have Jupyter installed you can open up the notebooks with that and this will allow you to interact with the notebooks. You will have to download the data files for the notebooks to work properly. Those can be found here.

Best Accuracy

The best accuracy that I have achieved was 0.64175 with the FTRL-Proximal script.

Future work

I am still tinkering with the data set provided to use the libffm implementation to try and get better results.

LICENSE

MIT