Image-Based Text Classification using 2D Convolutional Neural Networks - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2019

Image-Based Text Classification using 2D Convolutional Neural Networks

Résumé

We propose a new approach to text classificationin which we consider the input text as an image and apply2D Convolutional Neural Networks to learn the local andglobal semantics of the sentences from the variations of thevisual patterns of words. Our approach demonstrates thatit is possible to get semantically meaningful features fromimages with text without using optical character recognitionand sequential processing pipelines, techniques that traditionalnatural language processing algorithms require. To validateour approach, we present results for two applications: textclassification and dialog modeling. Using a 2D ConvolutionalNeural Network, we were able to outperform the state-of-art accuracy results for a Chinese text classification task andachieved promising results for seven English text classificationtasks. Furthermore, our approach outperformed the memorynetworks without match types when using out of vocabularyentities from Task 4 of the bAbI dialog dataset.

Dates et versions

hal-03081713 , version 1 (18-12-2020)

Identifiants

Citer

Erinc Merdivan, Anastasios Vafeiadis, Dimitrios Kalatzis, Sten Hanke, Joahannes Kroph, et al.. Image-Based Text Classification using 2D Convolutional Neural Networks. 2019 IEEE SmartWorld, Ubiquitous Intelligence & Computing, Advanced & Trusted Computing, Scalable Computing & Communications, Cloud & Big Data Computing, Internet of People and Smart City Innovation (SmartWorld/SCALCOM/UIC/ATC/CBDCom/IOP/SCI), Aug 2019, Leicester, United Kingdom. pp.144-149, ⟨10.1109/SmartWorld-UIC-ATC-SCALCOM-IOP-SCI.2019.00066⟩. ⟨hal-03081713⟩
44 Consultations
0 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More