Multimodal sentiment and emotion recognition in hyperbolic space

作者：

Highlights：

• Common methods for sentiment and emotion recognition are in the Euclidean space.

• Some data exhibit hierarchical structures, well modelled in the hyperbolic space.

• Hyperbolic learning is proposed to improve the recognition task.

• Hyperbolic models outperform classical Euclidean methods on benchmark datasets.

摘要

•Common methods for sentiment and emotion recognition are in the Euclidean space.•Some data exhibit hierarchical structures, well modelled in the hyperbolic space.•Hyperbolic learning is proposed to improve the recognition task.•Hyperbolic models outperform classical Euclidean methods on benchmark datasets.

论文关键词：Mmultimodal machine learning,Hyperbolic learning,Emotion recognition,Sentiment analysis,Deep learning

论文评审过程：Received 4 March 2021, Revised 10 June 2021, Accepted 26 June 2021, Available online 5 July 2021, Version of Record 8 July 2021.

论文官网地址：https://doi.org/10.1016/j.eswa.2021.115507