Knowledge-oriented convolutional neural network for causal relation extraction from natural language texts

作者：

Highlights：

• Knowledge-oriented convolutional neural network (K-CNN) performs better than CNN.

• Word filters capture linguistic clues of causal relation and alleviate overfitting.

• Word filter selection and clustering improve the performance of K-CNN.

• Semantic features improve precision and recall for complex causal relations.

• Combination of knowledge and data improves the performance of deep learning model.

摘要

•Knowledge-oriented convolutional neural network (K-CNN) performs better than CNN.•Word filters capture linguistic clues of causal relation and alleviate overfitting.•Word filter selection and clustering improve the performance of K-CNN.•Semantic features improve precision and recall for complex causal relations.•Combination of knowledge and data improves the performance of deep learning model.

论文关键词：Natural language processing,Convolutional neural network,Relation extraction,Causal relationship,Lexical knowledge base

论文评审过程：Received 26 April 2018, Revised 2 July 2018, Accepted 7 August 2018, Available online 8 August 2018, Version of Record 24 August 2018.

论文官网地址：https://doi.org/10.1016/j.eswa.2018.08.009