| Abstract |
In the development of smart cities, public service feedback submitted by citizens through mobile applications or other channels is often presented in the form of short texts. These short texts hold significant economic and social value in natural language processing (NLP) fields such as public opinion monitoring, service satisfaction analysis, and issue classification within smart cities. However, the inherent characteristics of short text data, such as data sparsity and contextual semantic scarcity, present considerable challenges for accurate classification. To address these issues, this paper proposes a smart city public service feedback short text classification model that combines multivariate semantic features and Graph Convolutional Networks (MSF-GCN). MSF-GCN utilizes the Language Technology Platform (LTP) from Harbin Institute of Technology to extract various semantic features from the short texts and constructs a multivariate heterogeneous graph incorporating these features along with the original text. This graph is then used as input for GCN, which learns deeper features of the short text. Finally, a Softmax function is applied to obtain the probability distribution for each category, thus achieving automatic classification of smart city public service feedback short texts. Experimental results show that the proposed model improves the F1-score by 4% compared to traditional single models, effectively enhancing classification accuracy. © 2024 IEEE. |
