Hierarchical Multi-label Classifier Based on Transformer Encoder for Grassroots Social Network Governance

Hierarchical multi-label classification (HMC) involves assigning objects to multiple paths in a hierarchical label structure. This makes HMC a unique classification task. HMC approach can be leveraged to improve the grassroots social network governance. This paper discusses the problems associated with existing methods for solving HMC problems and proposes a hierarchical multi-label classifier based on Transformer encoder (HMC-TE) to address these issues. HMC-TE is a hybrid classifier with both local and global information. It is mainly implemented by Transformer-Encoder. Existing hybrid methods fail to incorporate the effect of word position information on the text, which makes it difficult to capture the semantics and positional information of words in the sequence. To address these limitations, the model integrates positional variables to capture sequential data and improve the hierarchical multi-label classification problem. The proposed model, HMC-TE, was compared to state-of-the-art baseline algorithms in experiments conducted on 8 real-world datasets. HMC-TE has been shown to significantly enhance the performance of hierarchical multi-label classification tasks.