TY - GEN
T1 - Transformer-Based Contrastive Multi-view Clustering via Ensembles
AU - Zhao, Mingyu
AU - Yang, Weidong
AU - Nie, Feiping
N1 - Publisher Copyright:
© 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.
PY - 2023
Y1 - 2023
N2 - Multi-view spectral clustering has achieved considerable performance in practice because of its ability to explore nonlinear structure information. However, most existing methods belong to shallow models and are sensitive to the original similarity graphs. In this work, we proposed a novel model of Transformer-based contrastive multi-view clustering via ensembles (TCMCE) to solve the above issues. Our model integrates the self-attention mechanism, ensemble clustering, graph reconstruction, and contrastive learning into a unified framework. From the viewpoint of orthogonal and nonnegative graph reconstruction, TCMCE aims to learn a common spectral embedding as the indicator matrix. Then the graph contrastive learning is performed on the reconstructed graph based on the fusion graph via ensembles. Extensive experiments on six real-world datasets have verified the effectiveness of our model on multi-view clustering tasks compared with the state-of-the-art models.
AB - Multi-view spectral clustering has achieved considerable performance in practice because of its ability to explore nonlinear structure information. However, most existing methods belong to shallow models and are sensitive to the original similarity graphs. In this work, we proposed a novel model of Transformer-based contrastive multi-view clustering via ensembles (TCMCE) to solve the above issues. Our model integrates the self-attention mechanism, ensemble clustering, graph reconstruction, and contrastive learning into a unified framework. From the viewpoint of orthogonal and nonnegative graph reconstruction, TCMCE aims to learn a common spectral embedding as the indicator matrix. Then the graph contrastive learning is performed on the reconstructed graph based on the fusion graph via ensembles. Extensive experiments on six real-world datasets have verified the effectiveness of our model on multi-view clustering tasks compared with the state-of-the-art models.
KW - Contrastive graph learning
KW - Ensemble clustering
KW - Graph reconstruction
KW - Multi-view clustering
KW - Transformer
UR - https://www.scopus.com/pages/publications/85174446090
U2 - 10.1007/978-3-031-43412-9_40
DO - 10.1007/978-3-031-43412-9_40
M3 - 会议稿件
AN - SCOPUS:85174446090
SN - 9783031434112
T3 - Lecture Notes in Computer Science
SP - 678
EP - 694
BT - Machine Learning and Knowledge Discovery in Databases
A2 - Koutra, Danai
A2 - Plant, Claudia
A2 - Gomez Rodriguez, Manuel
A2 - Baralis, Elena
A2 - Bonchi, Francesco
PB - Springer Science and Business Media Deutschland GmbH
T2 - 23rd Joint European Conference on Machine Learning and Knowledge Discovery in Databases, ECML PKDD 2023
Y2 - 18 September 2023 through 22 September 2023
ER -