Title: Heterogeneous graph convolutional neural network for short text classification

Authors: Bo Huang; Peipei Li; Zhijun Fang; Lv Lei; Chenming Wang

Addresses: School of Electrical and Electronic Engineering, Shanghai University of Engineering Science, 333 Longteng Road, Shanghai, 201620, China ' School of Electrical and Electronic Engineering, Shanghai University of Engineering Science, 333 Longteng Road, Shanghai, 201620, China ' School of Electrical and Electronic Engineering, Shanghai University of Engineering Science, 333 Longteng Road, Shanghai, 201620, China ' CSG Smart Science and Technology Co., Ltd., 777 Sizhuan Road, Shanghai, 201619, China ' School of Electrical and Electronic Engineering, Shanghai University of Engineering Science, 333 Longteng Road, Shanghai, 201620, China

Abstract: Most existing short text classification methods treat each phrase as an independent homogeneous distribution, thus losing the association information between sentences. To solve this problem, we propose a heterogeneous GCN for short text classification. The heterogeneous text graphs are constructed using word nodes to take advantage of the structural features of corpus graph and sentence graph. Integrating different topologies of text graphs captures more neighbourhood information, thus addressing the sparsity of short text. To better fuse topological features, this paper extracts text graph structure features with the help of GCN, constructs hypernode to capture global information of text, and deeply interacts and fuses with local information captured by CNN. Experiments demonstrate that the model outperforms the other models on the three benchmark datasets. The model was tested on a fault text classification dataset provided by an automotive company, thus enabling effective validation of the model in a specific industry domain.

Keywords: short text classification; heterogeneous graph; graph neural network; vector fusion.

DOI: 10.1504/IJISTA.2023.134984

International Journal of Intelligent Systems Technologies and Applications, 2023 Vol.21 No.4, pp.344 - 365

Received: 07 Apr 2023
Accepted: 30 Jun 2023
Published online: 23 Nov 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article