2026-10-12T13:46:50Z https://api.figshare.com/v2/oai

oai:figshare.com:article/32826302 2026-10-01T16:09:06Z portal_1172 item_type_8 month_year_10_2026

Accelerating subgraph isomorphism search using graphlets Syed Ibtisam Tauhidi (24304541) PUREID: 684987719 Graph analytics subgraph isomorphism graphlets orbits machine Learning heuristic algorithm parallel algorithm Graphs are ubiquitous data structures for modelling complex relationships across diverse domains. Within graph analytics, a critical challenge is the Subgraph Isomorphism (SI) search problem. This NP-complete problem identifies all exact embeddings of a smaller pattern graph within a massive data graph. While numerous heuristic algorithms have been proposed to accelerate this search, no single approach consistently dominates across varying datasets. To overcome this limitation, this thesis presents a framework of solutions using machine learning, heuristic development, and parallel computing.<br><br>The first contribution is a machine learning-based framework for per-instance algorithm selection. Recognising the variability in graph semantics, this system extracts graph-specific features, such as summary statistics, label distributions, and local topological information, to dynamically predict the optimal SI heuristic per pattern-data graph pair. By transitioning from static heuristics to dynamic adaptability, this approach achieves runtime improvements of up to 1.54 times across diverse datasets. The second contribution introduces OrbitSI, a heuristic algorithm. OrbitSI accelerates the SI search process by exploiting the local topological properties of both the data and pattern graphs during preprocessing. By enforcing stricter preprocessing constraints based on local structures, OrbitSI drastically reduces the viable search space and eliminates redundant computations. Evaluations demonstrate that OrbitSI outperforms state-of-the-art baselines, delivering speedups ranging from 2.69 to 11.49 times on various datasets. The final contribution addresses the need for rapid topological profiling by introducing the NI-ORCA algorithm. Profiling the local structures around vertices is a prerequisite for many graph analytics algorithms, including OrbitSI. NI-ORCA studies parallelisation techniques to efficiently distribute computational workloads across multiple processing units. This yields massive improvements in scalability and execution time, achieving speedups of up to 60 times.<br><br>Together, these three contributions form a framework that improves the efficiency, scalability, and adaptability of Subgraph Isomorphism search in modern graph analytics. 2026-10-01T16:09:06Z Text Thesis 10.17034/32826302.v1 https://figshare.com/articles/thesis/Accelerating_subgraph_isomorphism_search_using_graphlets/32826302 All Rights Reserved