The targeted covert collection of Internet data, which can not only effectively hide users’ traces but also efficiently navigate to target data, plays an important role in three-dimensional social security. Due to big data generated by online social networks, large-scale network analysis usually confront with unbearable computational complexity, resulting in failing to achieve expected results within the specified response time. To retrieve and analyze the most important data within predetermined time, we study the problem of targeted covert collection with multiple granularity levels for Internet data in this paper. To find target data (or web users interchangeably), we use a tree indexes with the largest spanning tree to rank Internet data. Finally, we propose an effective collect algorithm for Internet data acquisition under different granularity levels. We validate our proposed scheme with the real data, which are gathered from the an online social platform. The experiment results verify that our algorithm can find the target data and collect data covertly with different granularity levels.