| Abstract: |
The exponential proliferation of large-scale interconnected data across domains such as social networks, knowledge graphs, biological systems, and recommendation engines has rendered scalable graph algorithms a cornerstone of modern computational intelligence. This review paper presents a comprehensive meta-analysis of scholarly contributions spanning the past two decades, synthesizing foundational and contemporary developments in graph algorithm optimization for large-scale data processing and graph-based machine learning (GraphML). The paper systematically surveys distributed graph computation frameworks, graph neural network (GNN) architectures, sampling-based scalability strategies, graph partitioning heuristics, and hardware-accelerated graph processing systems. Through critical meta-analytic evaluation of over thirty seminal and recent studies, recurring performance bottlenecks, algorithmic trade-offs, and scalability thresholds are identified and discussed. The review further examines how optimization paradigms including approximation algorithms, asynchronous processing, and mini-batch training have redefined the computational boundary of graph workloads. Findings indicate that while significant progress has been achieved in memory efficiency and distributed throughput, open challenges persist in dynamic graph handling, heterogeneous graph representation, and end-to-end scalable GraphML pipelines. This work serves as a structured reference for researchers and practitioners designing next-generation graph-based intelligent systems. |
