Optimal $(\alpha,\beta)$-Dense Subgraph Search in Bipartite Graphs

TL;DR

This paper introduces BD-Index, an efficient and scalable indexing method for $(oldsymbol{ extalpha},oldsymbol{eta})$-dense subgraph queries in bipartite graphs, supporting dynamic updates with flexible trade-offs.

Contribution

The paper proposes BD-Index, a novel index structure with optimal query time and linear space, along with two strategies for dynamic maintenance in bipartite graphs.

Findings

BD-Index answers queries in optimal time with linear space.

Two maintenance strategies balance update efficiency and memory usage.

Experiments show high scalability on large real-world datasets.

Abstract

Dense subgraph search in bipartite graphs is a fundamental problem in graph analysis, with wide-ranging applications in fraud detection, recommendation systems, and social network analysis. The recently proposed $(\alpha, \beta)$-dense subgraph model has demonstrated superior capability in capturing the intrinsic density structure of bipartite graphs compared to existing alternatives. However, despite its modeling advantages, the $(\alpha, \beta)$-dense subgraph model lacks efficient support for query processing and dynamic updates, limiting its practical utility in large-scale applications. To address these limitations, we propose BD-Index, a novel index that answers $(\alpha, \beta)$-dense subgraph queries in optimal time while using only linear space $O(|E|)$, making it well-suited for real-world applications requiring both fast query processing and low memory consumption. We further develop two complementary maintenance strategies for dynamic bipartite graphs to support efficient updates to the BD-Index. The space-efficient strategy updates the index in time complexity of $O(p \cdot |E|^{1.5})$ per edge insertion or deletion, while maintaining a low space cost of $O(|E|)$ (the same as the index itself), where $p$ is typically a small constant in real-world graphs. In contrast, the time-efficient strategy significantly reduces the update time to $O(p \cdot |E|)$ per edge update by maintaining auxiliary orientation structures, at the cost of increased memory usage up to $O(p \cdot |E|)$. These two strategies provide flexible trade-offs between maintenance efficiency and memory usage, enabling BD-Index to adapt to diverse application requirements. Extensive experiments on 10 large-scale real-world datasets demonstrate high efficiency and scalability of our proposed solutions.