Analysis of METIS graph partitioning algorithms for trust and recommendation systems

14 pages•Published: August 6, 2024

Mohith Siva Sai Bayana, Liz Maria Liyons and Geetha M

Abstract

In the era where social media and technology intersect, the vast user base of social networks presents a challenge in handling massive data. The issue intensifies when making user suggestions amidst the overwhelming data flow. This analysis addresses the complexities arising from the abundance of data in social networks and proposes a solu- tion through advanced graph partitioning techniques, focusing on algorithms from promi-nent libraries like DGL and PyTorch. This analysis compares three graph partitioning algorithms for social network analysis: DGL METIS (edge-balanced and node-balanced), and PyG METIS. We analyze their performance on the Epinions social recommendation dataset, focusing on edge based and node based metrics and visualization of partitions.Our findings reveal: PYG METIS consistently exhibited suboptimal performance across various evaluation metrics, with the exception of achieving satisfactory results in node balance. Conversely, DGL Node Balanced METIS demonstrated marginally superior outcomes compared to DGL Edge Balanced METIS in terms of edge loss and average edges per partition and surpassed it in node balance.

Keyphrases: gnn, graph partitioning, metis, trust and recommendation system

In: Rajakumar G (editor). Proceedings of 6th International Conference on Smart Systems and Inventive Technology, vol 19, pages 27-40.

Links:	https://easychair.org/publications/paper/qR3g
	https://doi.org/10.29007/bw69

BibTeX entry

@inproceedings{ICSSIT2024:Analysis_METIS_graph_partitioning,
  author    = {Mohith Siva Sai Bayana and Liz Maria Liyons and Geetha M},
  title     = {Analysis of METIS graph partitioning algorithms for trust and recommendation systems},
  booktitle = {Proceedings of 6th International Conference on Smart Systems and Inventive Technology},
  editor    = {Rajakumar G},
  series    = {Kalpa Publications in Computing},
  volume    = {19},
  publisher = {EasyChair},
  bibsource = {EasyChair, https://easychair.org},
  issn      = {2515-1762},
  url       = {/publications/paper/qR3g},
  doi       = {10.29007/bw69},
  pages     = {27-40},
  year      = {2024}}

Download PDF Open PDF in browser