A two-state epidemic model in networks with links mimicking two kinds of relationships between connected nodes is introduced. Links of weights w1 and w0 occur with probabilities p and 1 − p, respectively. The fraction of infected nodes ρ(p) shows a non-monotonic behavior, with ρ drops with p for small p and increases for large p. For small to moderate w1/w0 ratios, ρ(p) exhibits a minimum that signifies an optimal suppression. For large w1/w0 ratios, the suppression leads to an absorbing phase consisting only of healthy nodes within a range pL ≤ p ≤ pR, and an active phase with mixed infected and healthy nodes for p < pL and p > pR. A mean field theory that ignores spatial correlation is shown to give qualitative agreement and capture all the key features. A physical picture that emphasizes the intricate interplay between infections via w0 links and within clusters formed by nodes carrying the w1 links is presented. The absorbing state at large w1/w0 ratios results when the clusters are big enough to disrupt the spread via w0 links and yet small enough to avoid an epidemic within the clusters. A theory that uses the possible local environments of a node as variables is formulated. The theory gives results in good agreement with simulation results, thereby showing the necessity of including longer spatial correlations.

Detecting communities in large-scale networks is a challenging task when each vertex may belong to multiple communities, as is often the case in social networks. The multiple memberships of vertices and thus the strong overlaps among communities render many detection algorithms invalid. We develop a Partial Community Merger Algorithm (PCMA) for detecting communities with significant overlaps as well as slightly overlapping and disjoint ones. It is a bottom-up approach based on properly reassembling partial information of communities revealed in ego networks of vertices to reconstruct complete communities. Noise control and merger order are the two key issues in implementing this idea. We propose a novel similarity measure between two merged communities that can suppress noise and an efficient algorithm that recursively merges the most similar pair of communities. The validity and accuracy of PCMA is tested against two benchmarks and compared to four existing algorithms. It is the most efficient one with linear complexity and it outperforms the compared algorithms when vertices have multiple memberships. PCMA is applied to two huge online social networks, Friendster and Sina Weibo. Millions of communities are detected and they are of higher qualities than the corresponding metadata groups. We find that the latter should not be regarded as the ground-truth of structural communities. The significant overlapping pattern found in the detected communities confirms the need of new algorithms, such as PCMA, to handle multiple memberships of vertices in social networks.

The conventional notion of community that favors a high ratio of internal edges to outbound edges becomes invalid when each vertex participates in multiple communities. Such a behavior is commonplace in social networks. The significant overlaps among communities make most existing community detection algorithms ineffective. The lack of effective and efficient tools resulted in very few empirical studies on large-scale detection and analyses of overlapping community structure in real social networks. We developed recently a scalable and accurate method called the Partial Community Merger Algorithm (PCMA) with linear complexity and demonstrated its effectiveness by analyzing two online social networks, Sina Weibo and Friendster, with 79.4 and 65.6 million vertices, respectively. Here, we report in-depth analyses of the 2.9 million communities detected by PCMA to uncover their complex overlapping structure. Each community usually overlaps with a significant number of other communities and has far more outbound edges than internal edges. Yet, the communities remain well separated from each other. Most vertices in a community are multi-membership vertices, and they can be at the core or the peripheral. Almost half of the entire network can be accounted for by an extremely dense network of communities, with the communities being the vertices and the overlaps being the edges. The empirical findings ask for rethinking the notion of community, especially the boundary of a community. Realizing that it is how the edges are organized that matters, the f -core is suggested as a suitable concept for overlapping community in social networks. The results shed new light on the understanding of overlapping community.