[Project in collaboration with Telenor-NTNU AILab]
The use of recurrent neural network (RNN) have been proven to be effective for modelling sessions in a recommender engine for predicting the next product selected from a user.
Going further, the use of RNN give also the possibility of feeding and incorporating contextual information in the RNN and predicting WHEN the next item should be suggested (proactive action) other then WHAT should be recommended.
A less explored area, is the use of (Deep) Reinforcement Learning framework in a recommender angine, given the discrete action space (prediction of an item in a discrete space) and the definition of a reward function.
The student should explore the state of the art in the described areas and implement one or more related papers. It should also be able to understand problems and show the ability in tackling challenges
A minimal background in Machine Learning is requested (at least one of two courses)
[ 1] Recurrent Recommender Networks 
[ 2] Deep Reinforcement Learning in Large Discrete Action Spaces 
Keywords: Recsys, RNN, Deep Learning