[Paper Review] Predicting Small Group Accretion in Social Networks: A topology based incremental approach

Ankit Sharma, Xiaodong Feng+3 more|arXiv (Cornell University)|Jul 12, 2015

Complex Network Analysis Techniques25 references4 citations

TL;DR

This paper proposes a topology-based incremental approach to predict small group accretion in social networks, focusing on two key mechanisms: incremental accretion (IA) and subgroup accretion (SA). It introduces three models—Katz-based, network alignment-based, and label propagation-based—using hypergraph structures and random walks on group-actor networks, achieving high recall (up to 0.605) in per-group prediction, demonstrating strong performance for higher-order link prediction in evolving social networks.

ABSTRACT

Small Group evolution has been of central importance in social sciences and also in the industry for understanding dynamics of team formation. While most of research works studying groups deal at a macro level with evolution of arbitrary size communities, in this paper we restrict ourselves to studying evolution of small group (size $\leq20$) which is governed by contrasting sociological phenomenon. Given a previous history of group collaboration between a set of actors, we address the problem of predicting likely future group collaborations. Unfortunately, predicting groups requires choosing from $n \choose r$ possibilities (where $r$ is group size and $n$ is total number of actors), which becomes computationally intractable as group size increases. However, our statistical analysis of a real world dataset has shown that two processes: an external actor joining an existing group (incremental accretion (IA)) or collaborating with a subset of actors of an exiting group (subgroup accretion (SA)), are largely responsible for future group formation. This helps to drastically reduce the $n\choose r$ possibilities. We therefore, model the attachment of a group for different actors outside this group. In this paper, we have built three topology based prediction models to study these phenomena. The performance of these models is evaluated using extensive experiments over DBLP dataset. Our prediction results shows that the proposed models are significantly useful for future group predictions both for IA and SA.

Motivation & Objective

• To address the challenge of predicting future small group collaborations (size ≤ 20) in social networks, which are often overlooked in favor of large community evolution.
• To model group evolution not as macro-level community shifts, but as incremental processes driven by actor additions or subgroup integrations.
• To reduce the intractable combinatorial explosion of $n \choose r$ possibilities by focusing on two dominant mechanisms: incremental accretion (IA) and subgroup accretion (SA).
• To develop topology-based prediction models grounded in social network theory and dyadic link prediction principles, adapted for higher-order group structures.
• To evaluate model performance on the DBLP dataset using both per-group and global metrics, emphasizing recall due to the high sparsity of positive cases.

Proposed method

• Extends the Katz centrality method to hypergraphs by modeling cycles that alternate between group networks and outer actor networks, capturing inter-group communication patterns.
• Proposes a network alignment-based supervised method that uses cycle-based similarity scores between groups and external actors to predict accretion likelihood.
• Introduces a label propagation-based method using bi-random walks on a composite network of groups and actors, where walks alternate between group cliques and external nodes.
• Employs a sequential bi-random walk algorithm (BiRW_seq) that alternates between group and outer network steps, with decay parameter $\alpha$ and maximum path lengths $l_g = l_o = 4$.
• Uses the group network adjacency matrix $\mathbf{N_g}$, inter-network matrix $\mathbf{I}$, and outer network matrix $\mathbf{N_o}$ to compute influence scores via iterative matrix multiplication.
• Applies these models to predict future group formations by ranking external actors based on their predicted attachment likelihood to existing groups.

Experimental results

Research questions

• RQ1 What are the dominant mechanisms driving the formation of future small groups in social networks?
• RQ2 How can the combinatorial explosion of possible group formations be reduced by focusing on specific accretion patterns?
• RQ3 Can topology-based models that extend dyadic link prediction to higher-order structures effectively predict incremental and subgroup accretion?
• RQ4 How do different modeling approaches—Katz-style, network alignment, and label propagation—compare in predicting future group memberships?
• RQ5 What are the performance trade-offs between per-group and global evaluation metrics in small group prediction?

Key findings

• The proposed models achieve an average recall of 0.605 for incremental accretion (IA) and 0.5135 for subgroup accretion (SA) at top 100 predictions per group, indicating strong coverage of true future groups.
• The label propagation-based method (BRWS) achieves the highest average precision (0.0355) and recall (0.6050) for IA, outperforming the Katz-style (GKS) and global label propagation (GLPS) models.
• For global metrics, the BRWS model achieves a precision of 0.0134 and recall of 0.0962 for IA, and 0.0327 precision and 0.0614 recall for SA, showing improved performance despite high sparsity.
• The high recall values (up to 0.605) suggest that the models effectively capture a significant portion of actual future group formations, even when precision is low due to the vast number of candidate groups.
• Statistical analysis confirms that IA and SA account for a large majority of future group formations, justifying the focus on these two mechanisms and reducing the search space from $n \choose r$ to manageable subsets.
• The results demonstrate that topology-based models leveraging hypergraph structures and bi-random walks are effective for higher-order link prediction in evolving social networks.