Conference Papers Year : 2022

Using Multi-level Attention Based on Concept Embedding Enrichen Short Text to Classification

Abstract

Aiming at the defects of short text, which lack context information and weak ability to describe topic, this paper proposes an attention network based solution for enriching topic information of short text, which can leverage both text information and concept embedding to represent short text. Specifically, short text encoder is used to enhance the representation of short texts in the semantic space. The concept encoder obtains the distribution representation of the concept through the attention network composed of C-ST attention and C-CS attention. Finally, Concatenating outputs from the two encoders creates a longer target representation of short text. Experimental results on two benchmark datasets show that our model achieves inspiring performance and outperforms baseline methods significantly.
Fichier principal
Vignette du fichier
529309_1_En_13_Chapter.pdf (487.9 Ko) Télécharger le fichier
Origin Files produced by the author(s)

Dates and versions

hal-04178738 , version 1 (08-08-2023)

Licence

Identifiers

Cite

Ben You, Xiaohong Li, Qixuan Peng, Ruihong Li. Using Multi-level Attention Based on Concept Embedding Enrichen Short Text to Classification. 12th International Conference on Intelligent Information Processing (IIP), May 2022, Qingdao, China. pp.148-155, ⟨10.1007/978-3-031-03948-5_13⟩. ⟨hal-04178738⟩
37 View
3 Download

Altmetric

Share

More