End-to-end event factuality prediction using directional labeled graph recurrent network

作者：

Highlights：

• A more practical End-to-End setting for Event Factuality Prediction is proposed.

• Event Anchor Detection and Factuality Induction are performed in jointly modeling.

• Directional and labeled syntactic information graph enhances graph neural network.

• State-of-the-art results on four benchmarks are achieved against strong baselines.

摘要

•A more practical End-to-End setting for Event Factuality Prediction is proposed.•Event Anchor Detection and Factuality Induction are performed in jointly modeling.•Directional and labeled syntactic information graph enhances graph neural network.•State-of-the-art results on four benchmarks are achieved against strong baselines.

论文关键词：Event factuality prediction,Event Anchor Detection,Joint modeling,Graph neural network,End-to-end,Syntactic information graph

论文评审过程：Received 13 May 2021, Revised 17 November 2021, Accepted 19 November 2021, Available online 8 December 2021, Version of Record 8 December 2021.

论文官网地址：https://doi.org/10.1016/j.ipm.2021.102836