Explainable and Scalable Job Recommendation System using Transformer-based Text Summarization, Cross-Encoder Semantic Similarity, and Ensemble Learning

This research investigates the impact of combining BERT-based semantic similarity with ensemble classifiers and explainable AI (XAI) on improving recommendation accuracy and interpretability in job matching systems aligned with the Industry 4.0 paradigm. We propose a hybrid system integrating BERT-based summarization, Cross-Encoder semantic scoring, and ensemble classification (XGBoost, Random Forest, AdaBoost) to deliver accurate and transparent recommendations. Leveraging a large LinkedIn dataset (1.3M job postings), the system captures joint semantic relevance and structural features such as job seniority and skill density. Experimental results show that XGBoost achieves 93.4% accuracy with an RMSE of 0.237. Explainability is achieved through SHAP and LIME, ensuring accountability in AI-assisted hiring. The framework also takes into account the recommendation of emerging roles such as robotics engineers, IoT analysts, and digital twin specialists; critical to smart manufacturing and future industry needs. Furthermore, the proposed system addresses semantic and transparency gaps in large-scale, real-world recruitment scenarios.