We introduce the idea of utilizing a recurrent neural network based representation learning approach to extract and model the complex and sequentially dependent player behavior in games. Our approach is based on the dynamical systems of Echo State Networks, which are very simple to evaluate yet powerful temporal representation learners. We empirically evaluate our approach by illustrating a case study for predicting player churn.
Predicting Player Churn with Echo State Networks
Type: Inproceedings
Author: R. Sifa
Journal: 2021 IEEE Conference on Games (CoG)
Year: 2021
Citation information
R. Sifa:
Predicting Player Churn with Echo State Networks.
2021 IEEE Conference on Games (CoG),
2021,
1-5,
IEEE,
Copenhagen, Denmark,
https://doi.org/10.1109/CoG52621.2021.9619059