Abstract: To address the limitation of existing community detection methods in overlooking inherent group information within social networks, a multi-view adaptive team aware community detection algorithm (Adapt-TaCD) is proposed. By defining user interaction information as an interaction view and team information as a team view, a multi-view network is constructed. The Jaccard similarity is employed to measure the consistency of nodes' first-order structural proximity across these views. Building upon this multi-view network, a modularity-based method is adopted to uncover community structures and derive the optimal community partitioning. To validate algorithm performance, the benchmark dataset CDDS-SCHOLAT with multi-relational characteristics from raw SCHOLAT social network data is constructed. Experiments for selecting the optimal system parameter γ were conducted on the NET-3K network of the CDDS-SCHOLAT dataset, confirming γ=1.5 as the optimal value for the system parameter. Using this optimal value, the Adapt-TaCD algorithm was compared with AP, Ncut, Louvain, NSA, and VGAER algorithms in community detection experiments on networks of varying scales within the CDDS-SCHOLAT dataset. The experimental results demonstrate that the Adapt-TaCD algorithm maintains stable performance across varying network scales with standard deviations of 0.052 0 for ACC, 0.009 9 for NMI, and 0.050 3 for ARI; compared to AP, Ncut, Louvain, NSA, and VGAER, Adapt-TaCD algorithm achieved average improvements of 0.36, 0.38, and 0.39 in ACC, NMI, and ARI, respectively; and the Adapt-TaCD algorithm exhibits superior efficiency, when processing a network with 10 000 nodes, it takes only 30.25 s. Experimental results confirm that Adapt-TaCD de-monstrates significant advantages in accuracy, stability, and computational efficiency, and is particularly suited for community detection tasks in large-scale multi-relational networks.