Anomaly Detection of government data based on multi-scale structure reconstruction

Aiming at the problems of low processing efficiency of multi-source heterogeneous data, difficulty of cross-level collaboration and high data security risk in government service platform, a government data anomaly detection that integrates multi-scale structure reconstruction and local attribute comparison is proposed. By borrowing the unsupervised graph anomaly detection technique and combining the multi-tier governance of government data and the characteristics of multi-source heterogeneous data, a hierarchical graph coarsening and multi-scale reconstruction mechanism is designed to realize the structural consistency modeling of cross-tier data; at the same time, subgraph comparison learning is introduced to enhance the ability to capture global anomalous patterns and realize the semantic alignment. We innovatively propose a collaborative optimization strategy of "local attribute comparison-multi-scale global structure reconstruction", which effectively solves the problem of detecting contextual and structural anomalies in government data. Experiments show that we apply the model to government service data and compare six baseline models to achieve the SOTA effect, with performance improvement of 0.1%-0.4% for different datasets, which proves the effectiveness of the method in this paper.