PathFinder: A Graph-Based Career Discovery Platform
Year Long Project
CompSoc
Abstract
This project presents a web-based dual-sided recruitment recommendation platform that intelligently matches candidates with suitable job opportunities and recruiters with relevant candidate profiles using a temporal heterogeneous graph neural network architecture. Traditional recruitment systems rely heavily on keyword matching and manual filtering, which are inefficient and ineffective in handling the severe cold-start and temporal constraints inherent in recruitment data. To overcome these limitations, the proposed system models candidates, jobs, skills, companies, and interactions as a unified temporal heterogeneous graph containing multiple node and edge types. The architecture integrates GraphSAGE and Graph Attention Network (GAT) layers to capture both structural and semantic relationships within the graph, while temporal sampling and filtering mechanisms ensure time-aware recommendations without future data leakage. The platform further supports dynamic node instantiation, real-time graph augmentation, and incremental online learning to continuously adapt to new users and interactions. The system provides an efficient, scalable, and adaptive solution for modern recruitment recommendation tasks.
Abstract
This project presents a web-based dual-sided recruitment recommendation platform that intelligently matches candidates with suitable job opportunities and recruiters with relevant candidate profiles using a temporal heterogeneous graph neural network architecture. Traditional recruitment systems rely heavily on keyword matching and manual filtering, which are inefficient and ineffective in handling the severe cold-start and temporal constraints inherent in recruitment data. To overcome these limitations, the proposed system models candidates, jobs, skills, companies, and interactions as a unified temporal heterogeneous graph containing multiple node and edge types. The architecture integrates GraphSAGE and Graph Attention Network (GAT) layers to capture both structural and semantic relationships within the graph, while temporal sampling and filtering mechanisms ensure time-aware recommendations without future data leakage. The platform further supports dynamic node instantiation, real-time graph augmentation, and incremental online learning to continuously adapt to new users and interactions. The system provides an efficient, scalable, and adaptive solution for modern recruitment recommendation tasks.
Report Information
Team Members
Team Members
Report Details
Created: May 10, 2026, 4:38 p.m.
Approved by: None
Approval date: None
Report Details
Created: May 10, 2026, 4:38 p.m.
Approved by: None
Approval date: None