Authors:
Yue Wang
1
;
2
;
Yujie Hu
1
;
2
;
Wenjing Chang
1
;
2
and
Jianjun Yu
1
Affiliations:
1
Computer Network Information Center, CAS, CAS Informatization Plaza No.2 Dong Sheng Nan Lu, Haidian District, Beijing 100083, China
;
2
University of Chinese Academy of Sciences, No.19A Yuquan Road, Shijingshan District, Beijing 100049, China
Keyword(s):
Risk Identification, Graph Neural Networks, Bipartite Graphs, Multi-Semantic Feature Fusion.
Abstract:
In the complex landscape of project management, ensuring the authenticity of participant involvement is paramount to achieving fairness, enforceability, and desired outcomes. Addressing the challenges posed by the heterogeneous nature of graphs, the underutilization of rich attribute information, and the scarcity of anomaly labels, we propose a Project Participation Authenticity Risk Identification Graph Neural Network (PARI-GNN), a novel architecture leveraging graph-based anomaly detection techniques to assess authenticity risks in project participation. PARI-GNN include a novel framework for risk identification using heterogeneous graphs. This method transforms heterogeneous graphs into bipartite graphs and combines multi-feature semantic fusion techniques with bipartite graph structures, providing a robust solution for identifying inauthentic participation. We evaluate our proposed model using real-world data. The experimental outcomes affirm the superior performance of PARI-GNN
in accurately discerning authenticity risks, demonstrating the efficacy and competitive advantage of the proposed framework over a variety of state-of-the-art methodologies.
(More)