Development and optimization of multidimensional data fusion and intelligent decision system for university students' employment market
TL;DR
An intelligent employment decision support system that integrates multi-source heterogeneous data with GNNs (Graph Neural Networks), which can effectively identify structurally mismatched groups, provide quantitative basis for professional adjustment and regional talent policies, and demonstrate good technical performance and institutional embedding potential is constructed.
Abstract
Faced with challenges such as diverse data sources, structural heterogeneity, and semantic fragmentation in the university student employment market, this article constructs an intelligent employment decision support system that integrates multi-source heterogeneous data with GNNs (Graph Neural Networks). The study first systematically sorted out the characteristic differences of four core data types, including university student enrollment, recruitment platforms, internship records, and behavior logs, and proposed a semantic normalization fusion framework based on the ability ontology library. Then design MEMN (Multimodal Employment Matching Network) to achieve fine-grained semantic alignment and interpretable scoring between students and positions through graph attention mechanism. Conduct multi scenario simulation experiments based on real data from 14327 graduates and 28954 job positions. The results showed that the model achieved a matching accuracy of 68.3% in Top-5, NDCG@10 In disciplines such as engineering and science, the similarity of weekly recommendations is consistently maintained within the range of 0.73-0.78, with a P95 (95th percentile delay) delay controlled at 238 milliseconds under 1000 concurrent requests. Its output can effectively identify structurally mismatched groups, provide quantitative basis for professional adjustment and regional talent policies, and demonstrate good technical performance and institutional embedding potential.
