Core Concepts
The authors propose efficient linear-time algorithms to solve the earliest arrival time (EAT) and fastest path duration (FPD) problems in public transportation networks, achieving substantial speedups over state-of-the-art algorithms.
Abstract
The authors focus on designing efficient algorithms for two fundamental path problems in public transport networks: the earliest arrival time (EAT) and the fastest path duration (FPD).
Key highlights:
The authors introduce the notion of "useful dominating paths" and show a one-to-one mapping between these paths in the original temporal graph and paths in the transformed edge-scan-dependency (ESD) graph. This helps eliminate the traversal of many unnecessary paths.
The authors leverage the topology of the ESD graph to avoid time validation computations during the traversal, as every path in the ESD graph corresponds to a time-respecting path in the original graph.
The authors devise linear-time algorithms (O(m+n)) to solve the EAT and FPD problems, ensuring that each edge is processed at most once.
Experimental evaluation on 9 real-world public transportation datasets shows that the authors' EAT algorithm achieves up to 183x speedup and the FPD algorithm achieves up to 34x speedup compared to state-of-the-art algorithms.
Stats
The average out-degree of vertices in real-world public transport networks is around 3, with a maximum of 61.
The temporal out-degree of vertices is much higher than the static out-degree, as shown in Table 1.
Quotes
"Around 45% of the total running time is spent towards processing chain edges, which is a bottleneck."
"Whenever a vertex u is visited at time t, processing (u, v, t', λ') is not required, if there exist another edge (u, v, t'', λ'') such that t' + λ' > t'' + λ'' or t' < t."