A machine learning algorithm for recommending the top N results for a multi-class target using XGBoost. The model is implemented in both R (with html preview) and Python. There is also a Python implementation of the same type of model using Tensorflow.
This multi-class recommender system reads in a .csv file that has, on each row, a series of features and a prediction target. The sample data have independent 6 feature columns and two feature columns that contain the same feature as the target column. These last two columns represent user history and are prior views of the target item.
The recommender uses XGBoost to calculate the multi-class probabilities using a multi:softprob objective.
A recommendation is considered valid if the target class is among the top N predictions.