TDAM: A topic-dependent attention model for sentiment analysis

作者：

Highlights：

• A new recurrent unit derives text representations by an internal attention mechanism.

• Internal attention is applied to discover text patterns by means of topic embeddings.

• Patterns generated under multi-task supervision generates polarity-bearing topics.

• The proposed hierarchical model benefits sentiment classification and topic extraction.

摘要

•A new recurrent unit derives text representations by an internal attention mechanism.•Internal attention is applied to discover text patterns by means of topic embeddings.•Patterns generated under multi-task supervision generates polarity-bearing topics.•The proposed hierarchical model benefits sentiment classification and topic extraction.

论文关键词：Sentiment analysis,Neural attention,Topic modeling

论文评审过程：Received 5 June 2019, Revised 18 June 2019, Accepted 10 July 2019, Available online 18 July 2019, Version of Record 18 July 2019.

论文官网地址：https://doi.org/10.1016/j.ipm.2019.102084