ZLZhenyu Li

Zhenyu Li

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Zhenyu Li has a total of 2 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Experimental analysis of position embedding in graph transformer networks

    by Man Hu, Dezhi Sun, Zhenyu Li

    The Transformer structures have achieved excellent performances in an increasing number of fields. In the field of natural language processing, the original Transformer structure was used to work on a full connectivity graph, which depicts all connections between words in a phrase. It has been subsequently applied in areas such as computer vision. Recently, many studies have attempted to apply Transformer structures to graph data and achieved good performance. However, it is still a challenge to embedding the structural information of the graph data. To understand the representation of graph data, we employed three distinct ways of embedding the positional and structural information of graph data based on the Transformer structure in this work. To learn the feature representation of the graph data, the encoded structural information and original feature information of the graph data were fed into the Transformer structure. For the learned feature representations, we used them for node classification and graph regression, and then compared and analyzed the three different encoding methods.

  • An evolutionary analysis approach based on core-edge structured communities for data mining

    by Zhenyu Li, Dezhi Sun, Man Hu

    With the rapid development of communication technology and social media, people are constantly interacting with each other, generating a large amount of behavioral data and forming a complex social network. With time, the interaction behaviors between people also change, which results in a constantly changing social network. Analysis of evolutionary social networks can effectively mining people's behavioral data, understanding people's behavior models in depth. Community detection is one of the important methods for social network analysis, and evolutionary analysis of dynamically changing communities, especially the core stable community structure in dynamic networks, can reveal the community structure at different periods and reveal the social network evolution patterns. In this paper, we define communities on real complex networks as core-edge structures with a tight core and sparse periphery, which can represent the importance of nodes within the community. And we propose a community detection algorithm based on the core-edge structure communities. Moreover, based on the characteristics of the core-edge community structure, we propose a community similarity calculation method based on the importance of members and an evolution analysis method based on the core-edge community structure.

  • Open Access

    Adaptive Aggregation-Transformation Decoupled Graph Convolutional Network for Semi-Supervised Learning

    by Dezhi Sun, Man Hu, Zhenyu Li

    Graph Convolutional Network (GCN) has achieved significant success in many graph representation learning tasks. GCN usually learns graph representations by performing Neighbor Aggregation (NA) and Feature Transformation (FT) operations. Deep Adaptive Graph Neural Network (DAGNN) improves NA operation in the aggregation-transformation decoupled GCN, which enables the model to obtain a large receiver field. However, the problem with GCN is that when the model is trained to a deeper level, the performance decreases. In particular, the influence of NA and FT on model degradation has not been fully considered. In this work, we propose a new decoupled GCN architecture to enhance the performance of deep GCN. First, we conduct an experimental analysis of the impact of NA and FT operations on the degradation of the deep GCN model. Subsequently, we propose an Adaptive Aggregation-Transform Decoupled Graph Convolutional Network (AATD-GCN) which divides the model into two depths $D_{NA}$ and $D_{FT}$ , and proposes improved approaches in NA and FT, respectively. The AATD-GCN reduces the influence of NA and FT operations on performance degradation of deep GCN, and obtains a large receiver field while extracting complex feature information. We investigate the requirements of model depth $D_{NA}$ and $D_{FT}$ for graphs of different structures and sizes. Finally, the effectiveness of the proposed architecture is verified through extensive experiments on real-world datasets, the experimental results show that AATD-GCN is superior in terms of accuracy and robustness.