This paper focuses on the problem of inefficient supply and demand matching in the talent market under the background of digital economy and AI, and puts forward an optimization model and intelligent recommendation system based on multi-feature deep learning. The system integrates text, structural and behavioral data, constructs a deep neural network based on the double-tower structure to learn the high-order semantic matching relationship between job seekers and positions through the weighted features of attention mechanism, and introduces reinforcement learning (RL) to realize dynamic strategy adjustment and improve the dynamic adaptability of recommendation system. The intelligent recommendation system adopts hierarchical architecture design, covering data layer, feature engineering layer, model service layer and application layer. It uses Apache Flink real-time stream processing, BERT fine-tuning, TF Serving online reasoning and other technologies to achieve efficient and accurate job recommendation. The experiment was based on 2 million job-resume interaction records from Zhaopin.com. The results show that the system outperforms traditional methods and other baseline models in matching accuracy (AUC of 0.91, NDCG@5 of 0.72), system response latency, and business conversion rate. After being deployed in 10 companies, the system significantly improved recruitment efficiency—for example, the average resume screening time per job position decreased by 64.3%, the recall rate of high-quality resumes increased by 76.3%, the average job filling cycle was shortened by 44.7%, and HR satisfaction improved by 53.6%. In addition, the system enhances recommendation interpretability and user experience by using SHAP values to explain model predictions.
Jianguo et al. (Sun,) studied this question.