It is an open competition online, and we found it very interesting and challenging.
What it wants
We are asked to derive an efficient and effective algorithm to find out most potential advertisement retargeting customers.
How we built it
We built it through data analysis, feature extraction, model derivations, and robust experiments.
Challenges we ran into
We encountered huge user-item networks, and it is hard to address such a big dataset. In the beginning, we didn't know how to improve our model performance. Besides, there are lots of tricky points in dealing with the dataset preprocessing and model building.
Accomplishments that we're proud of
We started everything from scratches; however, we at the end derived a novel algorithm with great performance.
What we learned
We learned how to run through complete machine learning framework in merely thirty-six hours.
What's next for Conversion Probability Prediction in Retargeting
Invest more time in feature creations and model derivations to improve our mode performance.