Community detection is a fundamental task that aims to partition a network into cohesive node groups. In real-world networks, partial prior information is available. However, existing community detection methods that utilize prior information ignore the sensitivity to different prior information and the dynamic evolution of node relevance to the expanding community, which lead to unstable results and an imbalance between annotation cost of prior information and performance. To address these issues, we propose ANS-CD, a community detection method via active node selection. Specifically, ANS-CD adopts a multi-perspective strategy to identify valuable nodes that exhibit high representativeness or uncertainty as core community seeds, ensuring efficient use of limited annotations. It then expands communities around them through an innovative conflict triggered labeling mechanism, which corrects the conflicts between communities and nodes during the expansion process. Therefore, compared with traditional methods that rely on prior information, ANS-CD can extract valuable prior information and dynamically capture high-valuable information with low annotation costs. Extensive experiments on real-world attributed networks demonstrate that ANS-CD outperforms comparison methods with fewer labeled nodes.