Churn modeling with probabilistic meta paths-based representation learning

作者：

Highlights：

• We devise probabilistic meta paths representation learning for churn prediction.

• We propose a random walks sampling method based on the corresponding Markov models.

• Experiments on real-life datasets show the benefit of probabilistic meta path RL.

• We obtain promising insights on meta path type and predictive outcome interplay.

• Our method alleviates RL computational requirements by random walk sampling.

摘要

•We devise probabilistic meta paths representation learning for churn prediction.•We propose a random walks sampling method based on the corresponding Markov models.•Experiments on real-life datasets show the benefit of probabilistic meta path RL.•We obtain promising insights on meta path type and predictive outcome interplay.•Our method alleviates RL computational requirements by random walk sampling.

论文关键词：Representation learning,Heterogeneous networks,Meta path-based representation learning,Enriched (social) networks,Churn prediction in telco

论文评审过程：Received 14 November 2018, Revised 10 April 2019, Accepted 3 June 2019, Available online 22 June 2019, Version of Record 13 January 2020.

论文官网地址：https://doi.org/10.1016/j.ipm.2019.06.001