A two-step hybrid unsupervised model with attention mechanism for aspect extraction

作者：

Highlights：

• An unsupervised method for aspect extraction reduces cost of manual annotations.

• A hybrid approach for the sentences which doesn't follow the language constraints.

• Embedding based similarity method to extract domain-specific aspect terms.

• Attention-based aspect extraction method for long sentences.

• The experimental evaluation on the SemEval-16 dataset validates our approach.

摘要

•An unsupervised method for aspect extraction reduces cost of manual annotations.•A hybrid approach for the sentences which doesn't follow the language constraints.•Embedding based similarity method to extract domain-specific aspect terms.•Attention-based aspect extraction method for long sentences.•The experimental evaluation on the SemEval-16 dataset validates our approach.

论文关键词：Aspect extraction,Aspect-level sentiment analysis,Attention model,Deep learning,LSTM,Unsupervised learning

论文评审过程：Received 12 October 2019, Revised 26 May 2020, Accepted 16 June 2020, Available online 26 June 2020, Version of Record 10 July 2020.

论文官网地址：https://doi.org/10.1016/j.eswa.2020.113673