Querying cohesive subgraphs on temporal graphs just got a 10,000x speedup—literally 2 to 4 orders of magnitude faster than index-free approaches, according to a new paper on (k,δ)-truss indexes. The work generalizes the classic static k-truss by adding a temporal constraint: triangles forming the truss must exist within a time window δ. When δ = ∞, the new structure collapses back to the familiar k-truss, but for real-world temporal graphs, the δ constraint is what matters.
The (k,δ)-Truss: Temporal Cohesion in a Triangle
A k-truss in a static graph requires every edge to be part of at least k−2 triangles. The new (k,δ)-truss extends that to temporal graphs: not only must each edge participate in enough triangles, but those triangles must all fit within a span of δ time units. This catches time-localized clusters that a purely structural truss would miss—think bursty interactions in social networks, financial transaction spikes, or communication patterns in sensor logs. The paper defines the concept formally and shows that the original k-truss is a special case when δ = ∞.
Index Design: Lossless Compression Down to 10^{-4}
To answer queries like “find all edges belonging to the (3, 10-minute)-truss,” the authors propose both index-free and index-based approaches. The trick is a dual containment relation among (k,δ)-trusses: as k or δ changes, trusses nest inside each other. The index exploits this to losslessly compress the entire family of (k,δ)-trusses into a map or tree structure. Compression ratios hit 10^{-4}—that’s 0.01% of the original data volume. Queries become near-instant: the index-based method returns results in interactive time while the index-free baseline is 2 to 4 orders of magnitude slower. For a graph with billions of temporal edges, that’s the difference between a second and an hour.
Dynamic Maintenance Without Rebuilding
Temporal graphs change constantly. Adding or removing edges requires the index to stay fresh. The paper offers two construction algorithms: one based on full truss decomposition, the other on incremental truss maintenance. The latter substantially reduces redundant computations. More importantly, the index supports dynamic updates without a full rebuild—edge insertions and deletions propagate through the structure efficiently. The experimental evaluation confirms that updating the index is far cheaper than recomputing from scratch, making the approach practical for streaming temporal data.
What this unlocks: real-time cohesive subgraph queries on any temporal graph where time-constrained triangles signal something worth knowing. The index makes it cheap enough to deploy at scale, and the dynamic maintenance means you don’t have to freeze the graph to keep answers accurate.
Source: Querying Cohesive Subgraph regarding Span-Constrained Triangles on Temporal Graphs with Dynamic Index Maintenance
Domain: arxiv.org
Comments load interactively on the live page.