Empowering Job Recommendation with Large Language Models: A Generative and Attentive Approach

The escalating complexity of online recruitment demands next-generation job recommendation systems to overcome two fundamental limitations: (1) the contextual fragmentation in lengthy, noisy resumes/JDs, and (2) the temporal misalignment between static or unidirectional models and evolving user-employer intents. To address these challenges, we propose a novel B ehavior- R ich I nsights D riven job recommendation with G enerative E mbeddings (BRIDGE) framework, which energizes generative embedding refinement with intent-aware behavior modeling. Our key contributions lie in two aspects: (1) A generator-evaluator architecture with generative semantic alignment, which enhances holistic understanding of resumes/JDs and generates mutually enriched embeddings; (2) A co-evolutionary intent-aware attention mechanism which captures real-time preference shifts, ensuring adaptive recommendations. Finally, extensive comparative experiments and an online A/B test were conducted on the JiuYeJie online recruitment platform, validating BRIDGE's real-time effectiveness and superiority. The code is available at https://github.com/raochongzhi/BRIDGE/ .