With the rapid expansion of social networks, an influx of social bots has poured into these platforms. Their malicious activities, such as manipulating public opinion and invading users’ privacy, seriously disrupt the normal operation of social networks. Therefore, detecting social bots becomes one of the important tasks to ensure the security of social networks. Existing studies on the detection of social bots focus on utilizing graph neural networks, which can effectively capture community features, i.e., user attribute features and network topology. However, unlike genuine users in social networks, social bots typically exhibit behavior patterns of high-frequency activity at specific times. In this paper, we first study the differences on behavior patterns among different categories of users, and then propose a novel graph neural network model that integrates dual-dimensional features, including user behavior features and community features. The proposed model introduces user behavior features, which enhances its adaptability to dynamic scenarios. We evaluate the performance of the proposed model on real world datasets. The experimental results show that the proposed model clearly outperform the baseline methods. Especially, the dual-dimensional fusion approach can significantly improve detection accuracy and F1 scores.