Mutual Community Detection across Multiple Partially Aligned Social Networks

Community detection in online social networks has been a hot research topic in recent years. Meanwhile, to enjoy more social network services, users nowadays are usually involved in multiple online social networks simultaneously, some of which can share common information and structures. Networks that involve some common users are named as multiple “partially aligned networks”. In this paper, we want to detect communities of multiple partially aligned networks simultaneously, which is formally defined as the “Mutual Clustering” problem. The “Mutual Clustering” problem is very challenging as it has two important issues to address: (1) how to preserve the network characteristics in mutual community detection? and (2) how to utilize the information in other aligned networks to refine and disambiguate the community structures of the shared users? To solve these two challenges, a novel community detection method, MCD (Mutual Community Detector), is proposed in this paper. MCD can detect social community structures of users in multiple partially aligned networks at the same time with full considerations of (1) characteristics of each network, and (2) information of the shared users across aligned networks. Extensive experiments conducted on two real-world partially aligned heterogeneous social networks demonstrate that MCD can solve the “Mutual Clustering” problem very well.

💡 Research Summary

The paper addresses the problem of detecting communities simultaneously across several online social networks that share only a subset of users, termed “partially aligned networks.” Existing community detection methods either focus on a single network or assume fully aligned multi‑view data, and thus cannot preserve the unique structural characteristics of each network while leveraging the shared users to improve clustering quality. To fill this gap, the authors propose a novel framework called Mutual Community Detector (MCD).

First, the authors model each social platform (e.g., Foursquare and Twitter) as a heterogeneous information network consisting of users, posts, locations, timestamps, and words, together with directed links representing various social interactions. They define seven social meta‑paths (e.g., user‑follow‑user, user‑post‑word‑post‑user, etc.) that capture both direct and indirect relationships among users. Based on these meta‑paths they introduce a new similarity measure, HNMP‑Sim, which aggregates the counts of meta‑path instances weighted by learnable coefficients. This similarity matrix serves as the foundation for all subsequent clustering steps.

Second, to preserve each network’s intrinsic structure, the authors perform independent clustering on each network by minimizing a normalized cut objective defined on the HNMP‑Sim matrix. The result is a confidence matrix H for each network, indicating the probability of each user belonging to each cluster.

Third, they focus on the shared (anchor) users. For any pair of networks, they compute a normalized discrepancy between the anchor users’ cluster assignments. By introducing a regularization term that penalizes this discrepancy, they formulate a joint optimization problem that simultaneously (a) respects the independent clustering solutions and (b) aligns the anchor users across networks. The joint objective is solved by an alternating optimization scheme: fixing the independent clustering results while updating the anchor assignments to reduce discrepancy, then re‑optimizing the independent cuts, iterating until convergence.

To evaluate the method, the authors propose a new metric, Integrated Quality & Consensus (IQC), which combines traditional community quality measures (e.g., modularity, NMI) with a consensus score reflecting alignment across networks. Experiments on real‑world Foursquare and Twitter datasets demonstrate that MCD outperforms several baselines, including single‑network meta‑path clustering, multi‑view co‑regularized clustering, and existing co‑regularized multi‑domain graph clustering. Notably, MCD achieves higher NMI and modularity while also attaining superior IQC scores, indicating that it both preserves network‑specific structures and produces consistent community assignments for anchor users.

The paper concludes that MCD effectively solves the “Mutual Clustering” problem by integrating heterogeneous similarity, network‑specific preservation, and cross‑network regularization. Future work is suggested in three directions: extending the model to many‑to‑many anchor mappings (users with multiple accounts), adapting MCD to dynamic networks where relationships evolve over time, and exploring deep‑learning approaches for automatic meta‑path discovery and weight learning. Overall, the study provides a solid methodological contribution to multi‑network community detection and opens avenues for richer cross‑platform social analysis.