Developing an Intelligent Job Recommendation System Using Semantic Retrieval and Explainable AI Techniques

The rapid growth of online recruitment platforms has created a need for job recommen- dation systems that can retrieve relevant opportunities from large and heterogeneous collections of job postings. Conventional keyword-based search remains efficient and trans- parent, but it often fails when equivalent job roles are expressed with different terms. This paper presents an intelligent job recommendation system that combines lexical retrieval, semantic retrieval, and explainable artificial intelligence techniques. The system is designed for a metadata-only setting and uses structured fields such as job title, company name, location, seniority level, job function, employment type, and industry. It does not rely on full job descriptions, user profiles, click logs, or application histories. The proposed pipeline builds a sparse lexical representation using TF-IDF and a dense semantic represen- tation using Sentence-BERT embeddings. Candidate jobs are retrieved through semantic nearest-neighbor search and then ranked using a weighted hybrid scoring function. An optional Cross-Encoder re-ranking stage is used to refine the top candidates. To improve transparency, the system reports matched keywords, applied filters, and metadata-based evidence. Experiments on a cleaned LinkedIn job posting dataset containing 31,262 records show that the best hybrid configuration achieved Precision@10 of 0.8032 and nDCG@10 of 0.9496. A second-stage Cross-Encoder improved Precision@10 from 0.7896 to 0.7948 and nDCG@10 from 0.9666 to 0.9739 under the internal evaluation protocol. These findings indicate that a carefully engineered combination of lexical matching, semantic retrieval, and explainable ranking can produce effective job recommendations even when only structured metadata is available.