透過您的圖書館登入
IP:3.12.108.236
  • 學位論文

基於圖卷積網路和自編碼器之分群演算法以其應用

Leanring Cluster Assignment with GCN and Autoencoder

指導教授 : 鄭卜壬

摘要


資料分群在資料探勘的領域中是一個非常重要的任務,而且對任何需要資料去輔助的應用程式也都是非常重要的。在不同的分群演算法中,屬性圖分群是想要從節點的屬性以及從圖的結構,去學習節點的表示方式以及分群的結果,而這個任務,是非常重要,但同時又不那麼容易完成的。先前的研究主要是專注在發展跟圖卷積網路相關的方法,而這些方法通常是在將圖的結構編碼的時候會有特別好的效果,但是不可避免的就是會減少節點屬性在分群結果上的影響力。而更重要的是,大部分GCN的方法會對圖的結構非常敏感,因為GCN本身的傳播就是透過圖的邊去進行傳播。而現實世界中的圖,又很常是充滿著噪音,可能會缺一個邊多一個邊。為了解決這個問題,我們提出一個框架去有效率地透過節點屬性以及圖結構去學習節點的表達方式,並且設計了一個運算子在訓練的過程去更新圖的結構,以此減少圖中的noise.

並列摘要


Data clustering is a fundamental task in data mining, and especially crucial to many data-driven applications. Among different kinds of clustering, attributed graph clustering, aiming to learn node representation and cluster assignment from node attribute and graph structure is an important yet challenging task. Previous studies focused on developing Graph Convolutional Network based approach, which are powerful at encoding graph structure, but inevitably reduce the influence of attribute on clustering. Furthermore, most GCN methods are sensitive to the quality of graph structures because of the propagation mechanism of GCN. To deal with these problems, we proposed a framework to effectively learn node representation with both structure and attribute feature, and design a combine operator to keep update the graph structure during training process

參考文獻


[1] E. Aljalbout, V. Golkov, Y. Siddiqui, M. Strobel, and D. Cremers. Clustering with deep learning: Taxonomy and new methods, 2018.
[2] D. Bank, N. Koenigstein, and R. Giryes. Autoencoders, 2020.
[3] D. Bo, X. Wang, C. Shi, M. Zhu, E. Lu, and P. Cui. Structural deep clustering network. In Proceedings of The Web Conference 2020. ACM, apr 2020.
[4] M. Capó, A. Pérez, and J. A. Lozano. An efficient k -means clustering algorithm for massive data, 2018.
[5] M.Caron,P.Bojanowski,A.Joulin,andM.Douze.Deepclusteringforunsupervised learning of visual features, 2018.

延伸閱讀