| 摘 要: 为学习更丰富的语义表示以提升聚类效果,文章提出一种多层内部语义表示增强的深度文本聚类(Deep Document Clustering via Multi-layer Enhanced Internal Semantic Representation, DCISR)模型。首先,设计了一种语义融合策略,将其不同层次的外部结构语义表示逐层融入内部语义表示中。其次,充分利用编码层和解码层对语义补充的作用进行内部语义表示的补充增强。最后,设计了一种三重自监督机制,以监督模型参数更新。实验结果表明,该模型在4个真实文本数据集上的聚类性能均高于对比模型,验证了模型的有效性,可为未来开展相关工作提供参考。 |
| 关键词: 文本聚类;深度聚类;自编码器;语义表示;图卷积网络 |
|
中图分类号: TP391
文献标识码: A
|
| 基金项目: 贵州省教育厅职业教育科研项目(GZZJ-Q2022028);贵州轻工职业技术学院院级课题(23QY23) |
|
| Deep Document Clustering Model with Multi-layer Enhanced |
|
REN Lina1,2, YAO Maoxuan1
|
(1.Department of In formation Engineering, Guizhou Light Industry Technical College, Guiyang 550025, China;
2.College of Computer Science and Technology, Guizhou University, Guiyang 550025, China)
renlina111@163.com; yaomaoxuan@126.com
|
| Abstract: In order to learn richer semantic representations to enhance clustering performance, this paper proposes a Deep Document Clustering via Multi-layer Enhanced Internal Semantic Representation (DCISR) model. Firstly, a semantic fusion strategy is designed to gradually integrate external structural semantic representations from different levels into internal semantic representations. Secondly, the complementary enhancement of internal semantic representations is achieved by fully utilizing the encoding and decoding layers to supplement semantics. Additionally, a triple self-supervised mechanism is designed to supervise model parameter updates. Experimental results demonstrate that the proposed model outperforms comparative models in clustering performance on four real document datasets, which verifies the effectiveness of the model, providing reference for future related work. |
| Keywords: document clustering; deep clustering; autoencoder; semantic representation; graph convolutional network |
