A span-sharing joint extraction framework for harvesting aspect sentiment triplets

作者：

Highlights：

• We design a span-sharing joint extraction model to extract aspect sentiment triplets.

• The error propagation between different ABSA subtasks is effectively avoided in our model.

• Complex relations between aspect terms and opinion terms are effectively dealed with.

• The proposed model achieves outstanding performances for ASTE tasks and AOPE tasks.

摘要

•We design a span-sharing joint extraction model to extract aspect sentiment triplets.•The error propagation between different ABSA subtasks is effectively avoided in our model.•Complex relations between aspect terms and opinion terms are effectively dealed with.•The proposed model achieves outstanding performances for ASTE tasks and AOPE tasks.

论文关键词：Sentiment analysis,Aspect sentiment triplet extraction,Span-sharing,Syntactic dependency

论文评审过程：Received 19 September 2021, Revised 13 January 2022, Accepted 3 February 2022, Available online 10 February 2022, Version of Record 23 February 2022.

论文官网地址：https://doi.org/10.1016/j.knosys.2022.108366