Litcius/Paper detail

Link Prediction using GraphSAGE

Sarthak Bhatkar, Purva Gosavi, Vishakha Shelke, John Kenny

202316 citationsDOI

Abstract

In most real-world networks, not all connections or relationships are known. Link prediction helps fill in the gaps, thus providing a more comprehensive understanding of the network. Link prediction is a crucial task in network analysis and plays a pivotal role in domains such as social networks and recommendation systems. In this article, we propose an enhanced link prediction model that leverages snscrape, a data scraping tool for near-real-time acquisition of raw real-world data, and employs GraphSAGE, a Graph Neural Network (GNN) framework which possesses the ability to learn node embeddings in large-scale graphs that capture the network’s structural features and forecast new links. For link classification, we adopt the ‘IP’ method, enhancing the accuracy and reliability of our link predictions. Furthermore, we optimize our model’s performance by utilizing the widely recognized ‘ADAM’ optimizer. The proposed model contributes significantly to the field of network analysis across diverse domains, addressing link prediction challenges through a data-driven approach. In real-world scenarios, our approach outperforms existing methods by effectively capturing structural features of large-scale networks and conducting comprehensive evaluations with diverse parameters, leading to enhanced accuracy and reliability of link predictions.

Topics & Concepts

Computer scienceReliability (semiconductor)Link (geometry)Field (mathematics)Data miningRaw dataTask (project management)Machine learningArtificial intelligenceNode (physics)Artificial neural networkComplex networkGraphTheoretical computer scienceComputer networkWorld Wide WebQuantum mechanicsPure mathematicsMathematicsPower (physics)EngineeringProgramming languagePhysicsStructural engineeringEconomicsManagementComplex Network Analysis TechniquesAdvanced Graph Neural NetworksBioinformatics and Genomic Networks