Transformer-enhanced Hawkes process with decoupling training for information cascade prediction

作者：

Highlights：

• A novel coupled topological–temporal information cascade diffusion framework.

• Extending the traditional Hawkes process with a topological horizon by a two-level attention architecture.

• A learnable position-wise user embedding method for unattributed information directed acyclic graph.

• A decoupling training scheme to cope with the long-tailed distribution in cascade sizes.

摘要

•A novel coupled topological–temporal information cascade diffusion framework.•Extending the traditional Hawkes process with a topological horizon by a two-level attention architecture.•A learnable position-wise user embedding method for unattributed information directed acyclic graph.•A decoupling training scheme to cope with the long-tailed distribution in cascade sizes.

论文关键词：Information diffusion,Information cascade prediction,Hawkes point process,Attention mechanism,Popularity prediction

论文评审过程：Received 15 March 2022, Revised 16 August 2022, Accepted 17 August 2022, Available online 23 August 2022, Version of Record 5 September 2022.

论文官网地址：https://doi.org/10.1016/j.knosys.2022.109740