The notion of trends in social networks has emerged as an important problem attracting the attention of researchers as well as the industry. Although, recent work has studied trends from various perspectives such as its temporal and geospatial properties, the structural properties of the network that creates such trends are ignored in trend detection. In this work, we propose two novel structural trend deﬁnitions called correlated and uncorrelated trends that leverage friendship information to detect interesting topics that would not be detected using traditional trend deﬁnitions. We experimentally and analytically show that correlated trends are significantly different from traditional trends whereas the difference for uncorrelated trends, although corresponding to a useful variation, is less pronounced. We show that both correlated and uncorrelated trends identify interesting activities in social networks. We also show that the new trend deﬁnitions can be used to detect or ﬁlter suspicious activity in the network. Detection of structural trends is inherently harder than traditional trend detection. Therefore we propose a sampling technique that provides computational gain while remaining within an acceptable error bound. Experiments performed on a large-scale social network data of 41.7 million nodes and 417 million posts show that even with a small sampling rate of 0.005, the average precision lies above 0.93 for correlated trends while keeping a perfect average precision of 1 for uncorrelated trends.