Full metadata record
|dc.contributor||Department of Computing||en_US|
|dc.contributor.advisor||Huang, Xiao (COMP)||en_US|
|dc.publisher||Hong Kong Polytechnic University||en_US|
|dc.rights||All rights reserved||en_US|
|dc.title||Collaborative graph neural networks for unsupervised representation learning||en_US|
|dcterms.abstract||Graph neural networks (GNNs) have been intensively applied to analyze real-world networks, especially attributed networks. Existing studies of GNNs mainly focus on exploiting network structures, while the exploitation of node attributes is rather limited as they only serve as node representations at the initial layer. Other than that, the training objectives of most GNNs also do not include node attributes. They perform training mainly based on network structures in an unsupervised manner or the given node labels. However, node attributes are a major information source and need to be better exploited, as they are highly correlated with and complementary to the network. Thus, it is encouraging to deeply involve node attributes in the key components of GNNs, including the graph convolution operations and training objectives. However, this is a nontrivial task since an appropriate way of integrating is required to maintain the merits of GNNs. Additionally, node attributes such as user posts are distinct from topological structures and not in line with graph convolutions. To bridge the gap, in this paper, we focus on unsupervised learning and propose COllaborative graph Neural Networks (CONN). It refines GNNs from two aspects. First, a collaborative aggregation mechanism is designed. It updates the representation of each node by aggregating the representations of not only its neighbors, but also its attribute categories. Second, a collaborative training objective is developed. It assesses both node-to-node and node-to-attribute-category interactions based on cross correlations. Experiments on real-world networks show that CONN consistently outperforms state-of-the-art embedding algorithms. It even outperforms an end-to-end baseline on social networks.||en_US|
|dcterms.extent||vii, 42 pages : color illustrations||en_US|
|dcterms.isPartOf||PolyU Electronic Theses||en_US|
|dcterms.LCSH||Neural networks (Computer science)||en_US|
|dcterms.LCSH||Hong Kong Polytechnic University -- Dissertations||en_US|
Files in This Item:
|5868.pdf||For All Users (off-campus access for PolyU Staff & Students only)||1.29 MB||Adobe PDF||View/Open|
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item: