COL-Trees: Efficient Hierarchical Object Search in Road Networks

Location-based services rely heavily on efficient methods that search for relevant points-of-interest (POIs) near a given location. A k Nearest Neighbor (kNN) query is one such example that finds the k closest POIs from an agent’s location. While most existing techniques focus on retrieving nearby POIs for a single agent, these search heuristics do not translate to many other applications. For example, Aggregate k Nearest Neighbor (AkNN) queries require POIs that are close to multiple agents. k Farthest Neighbor (kFN) queries require POIs that are the antithesis of nearest. Such problems naturally benefit from a hierarchical approach, but existing methods rely on Euclidean-based heuristics, which have diminished effectiveness in graphs such as road networks. We propose a novel data structure, COL-Tree (Compacted Object-Landmark Tree), to address this gap by enabling efficient hierarchical graph traversal using a more accurate landmark-based heuristic. We then present query algorithms that utilize COL-Trees to efficiently answer AkNN, kFN, and other queries. In our experiments on real-world and synthetic datasets, we demonstrate that our techniques significantly outperform existing approaches, achieving up to 4 orders of magnitude improvement. Moreover, this comes at a small pre-processing overhead in both theory and practice.

💡 Research Summary

The paper addresses the problem of efficiently retrieving points‑of‑interest (POIs) on road‑network graphs for a variety of queries that go beyond the classic single‑source k‑nearest‑neighbor (k‑NN) problem. While existing methods rely on Euclidean distance heuristics or simple landmark lower‑bounds, they either provide very loose estimates on road‑network distances or cannot support aggregate‑distance (AkNN) and farthest‑neighbor (k‑FN) queries. To close this gap, the authors introduce two tightly coupled data structures:

1. Subgraph‑Landmark Tree (SUL‑Tree) – a lightweight index that recursively partitions the road graph into balanced subgraphs of size ≤ α, assigning a representative landmark to each subgraph. This partitioning is performed with a novel subgraph‑reordering technique that dramatically reduces the preprocessing time compared with naïve Dijkstra‑based landmark computation.

2. Compacted Object‑Landmark Tree (COL‑Tree) – the main hierarchical index built on top of the SUL‑Tree. Each internal node stores a “subgraph‑landmark distance list” (SDL) containing distances from every vertex of the subgraph to all landmarks; each leaf stores an “object‑landmark distance list” (ODL) for the POIs it contains.

The key insight is that for any monotone aggregate function (sum, max, etc.) the aggregation of landmark lower‑bounds (LLBs) yields a valid lower‑bound on the true aggregate distance (Lemma 1). Consequently, during query processing the algorithm can compute a lower‑bound aggregate distance (LB_agg) for every node and prune any node whose LB_agg exceeds the current best‑k threshold. For k‑FN queries, the authors also exploit landmark upper‑bounds (LUBs) to maintain a tight upper bound on the farthest distance seen so far, allowing early termination of branches whose LUB is smaller than the current kth farthest distance. Range queries are handled by simply discarding nodes whose LB exceeds the query radius.

The query algorithm proceeds as a best‑first search on the COL‑Tree: the root is inserted into a priority queue keyed by its LB_agg; the node with the smallest bound is expanded, its children’s bounds are recomputed, and leaves are evaluated with exact network distances only when necessary. The hierarchical nature of COL‑Tree ensures that only a tiny fraction of POIs are examined exactly, while the landmark‑based bounds guide the search efficiently.

Experimental evaluation on real‑world US road networks (hundreds of thousands of vertices, millions of edges) and synthetic graphs shows dramatic speedups. For AkNN, k‑FN, and range queries, COL‑Tree achieves up to 10⁴‑fold reduction in query time compared with Euclidean‑R‑Tree, landmark‑based k‑NN, and other recent graph indexes. Preprocessing time remains modest (tens of seconds) and scales linearly with the number of landmarks; memory consumption stays comparable to prior landmark approaches even when using 64–128 landmarks. The SUL‑Tree construction is 2–3× faster than the baseline landmark index, confirming its practicality for large‑scale deployment.

In summary, the contributions are:

• A landmark‑based hierarchical index (COL‑Tree) that supports tight lower‑ and upper‑bounds for both aggregate‑distance and farthest‑distance queries.
• Formal proof that monotone aggregates preserve lower‑bound validity, enabling optimal pruning for AkNN.
• The first optimal branch‑and‑bound algorithm for k‑FN on road networks using landmark upper‑bounds.
• The SUL‑Tree auxiliary structure that dramatically reduces preprocessing overhead while allowing COL‑Tree to compactly represent subgraphs.

Future work suggested includes dynamic updates (road closures, new POIs), extensions to non‑monotone aggregates, application to other graph domains (social, sensor networks), and adaptive landmark selection based on real‑time traffic conditions. Overall, COL‑Tree and SUL‑Tree together provide a powerful framework for high‑performance, multi‑criteria POI search in road‑network‑based location services.