Improving Data Forwarding in Mobile Social Networks with Infrastructure Support: A Space-Crossing Community Approach

In this paper, we study two tightly coupled issues: space-crossing community detection and its influence on data forwarding in Mobile Social Networks (MSNs) by taking the hybrid underlying networks with infrastructure support into consideration. The hybrid underlying network is composed of large numbers of mobile users and a small portion of Access Points (APs). Because APs can facilitate the communication among long-distance nodes, the concept of physical proximity community can be extended to be one across the geographical space. In this work, we first investigate a space-crossing community detection method for MSNs. Based on the detection results, we design a novel data forwarding algorithm SAAS (Social Attraction and AP Spreading), and show how to exploit the space-crossing communities to improve the data forwarding efficiency. We evaluate our SAAS algorithm on real-life data from MIT Reality Mining and UIM. Results show that space-crossing community plays a positive role in data forwarding in MSNs in terms of deliver ratio and delay. Based on this new type of community, SAAS achieves a better performance than existing social community-based data forwarding algorithms in practice, including Bubble Rap and Nguyen’s Routing algorithms.

💡 Research Summary

The paper addresses two tightly coupled problems in Mobile Social Networks (MSNs) that incorporate a small number of fixed Access Points (APs) alongside a large population of mobile users: (1) detecting “space‑crossing” communities that span physical proximity and AP‑mediated connections, and (2) exploiting these communities to improve opportunistic data forwarding.
Network model. The authors model the hybrid underlying network as a dynamic graph G = {G₀, G₁,…}, where each snapshot Gₜ = (Vₜ, Eₜ) contains mobile users and APs. Edges are created not only from direct Bluetooth/Wi‑Fi contacts but also from user‑AP associations and AP‑AP links (assumed to form a simple circle). To cope with heterogeneous contact frequencies, a weighted aggregation mechanism is introduced: a growing window for sparse traces and a sliding window for dense traces, both normalising contact counts by the total number of contacts in the considered interval. An edge is added when the pairwise encounter ratio exceeds the median of all ratios at that time.
Community definitions. Three community types are defined:
Physical Proximity (PP) Community – groups of nodes with dense internal contacts, detected by the FOCS algorithm (a static overlapping‑community detector that avoids modularity resolution limits).
AP Community – all users that can reach the same AP (including the AP itself), forming a complete sub‑graph because the AP acts as a hub.
Space‑Crossing (SC) Community – the union of PP and AP communities that satisfy specific overlap criteria, thereby connecting geographically distant nodes through AP‑mediated links.
Space‑crossing detection. The detection proceeds in two phases. In the initialization phase, PP communities are first extracted with FOCS, AP communities are identified, and then PP and AP communities are merged according to two criteria:

• Criterion Sa: If the shared sub‑structure (overlapped nodes and intra‑edges) between a PP and an AP community exceeds a threshold α (determined experimentally), the two are merged.
• Criterion Sb: Adjacent APs (connected in the assumed AP circle) are merged into a single community to avoid over‑centralisation.
  The result of this phase is a set of initial SC communities.
  Dynamic tracking. Because the network evolves, the authors classify changes into four elementary actions (node addition/removal, edge addition/removal) and propose local update rules for each. For example, when a node joins, the algorithm checks whether it belongs to existing PP communities, whether it creates a new PP community with its neighbours, and then re‑evaluates overlaps with AP communities using an extended version of Criterion Sa (named Criterion Sc). Similar logic applies for deletions and edge updates, ensuring that SC communities are continuously refreshed without recomputing the whole graph.
  SAAS forwarding algorithm. The proposed routing protocol, Social Attraction and AP Spreading (SAAS), operates in two complementary modes:

1. Social Attraction (non‑AP areas). Each node u has a local activity value defined as the ratio of its encounter probability with other members of its SC community to the total encounter probability inside that community. This activity is combined with the Pearson correlation coefficient between u and a destination d to obtain a similarity score. Messages are forwarded to neighbours with higher similarity, following a “social attraction” principle.
2. AP Spreading (AP‑controlled areas). When a message reaches an AP‑covered region, the AP replicates the message to all users within its community, effectively spreading multiple copies across the AP circle. This dramatically increases the number of carriers and reduces delivery latency, especially for destinations that are far from the source.
   Evaluation. Experiments use two real‑world traces: MIT Reality Mining (Bluetooth/Wi‑Fi logs from 100+ participants) and UIM (University of Illinois Movement). The authors compare SAAS against two well‑known social‑community routing schemes: Bubble Rap and Nguyen’s algorithm. Metrics include delivery ratio and average delay. Results show that SAAS consistently outperforms the baselines, achieving up to 30 % higher delivery ratios and 35 % lower delays, with the most pronounced gains in AP‑dense zones. An ablation study removing the SC‑community step confirms that the performance boost is primarily due to the space‑crossing community abstraction.
   Contributions and limitations. The paper’s main contributions are: (i) a novel space‑crossing community detection framework that merges physical proximity and infrastructure‑based groups; (ii) a lightweight dynamic tracking mechanism for evolving networks; (iii) the SAAS routing protocol that leverages community structure and AP spreading. Limitations include the simplifying assumption that APs form a single circular backbone and the reliance on a manually tuned α threshold. Future work could explore more realistic AP topologies, adaptive threshold learning, and integration of machine‑learning‑based similarity metrics.

Overall, the study demonstrates that even a modest deployment of infrastructure can be systematically exploited to create powerful cross‑space communities, leading to substantial improvements in opportunistic data forwarding for mobile social networks.