Authors: Hande Pooja Vinod, Professor Bangar Abhishek P., Professor Karad Vandana A., Professor Bhosale Sachin B.
Abstract: Job matching is increasingly supported by computational systems that estimate employment opportunities, interpret unstructured career information, and allocate scarce positions or resources. This review synthesises three supplied research papers that address complementary parts of this problem: large-scale machine learning for labour-market matching, strategyproof matching when project capacities depend on resource allocation, and CareerBERT for semantic resume-to-job recommendation. The first study analyses approximately 55 million German employment spells from 2012–2018 and reports that random forests outperform logit, ordinary least squares, and k-nearest neighbours for occupation-level prediction. The second formalises a student-project-resource problem and shows that strategyproofness cannot be combined with fairness and weak efficiency in full generality; it introduces Sample and Deferred Acceptance (SDA) mechanisms to balance these objectives. The third combines ESCO occupational taxonomy entries with EURES job advertisements and fine-tunes a BERT-based Siamese network using Multiple Negatives Ranking loss to place resumes and jobs in a shared semantic embedding space. Across the papers, matching quality depends on the representation of preferences and profiles, the scale and quality of data, the way capacities are modelled, and the objectives used to evaluate outcomes. Predictive accuracy and semantic relevance can improve recommendations, but they do not by themselves guarantee fairness, truthful reporting, or efficient resource use. The review therefore proposes a layered design that combines data-driven ranking with transparent allocation rules, human oversight, and explicit auditing of outcomes. Since the review is limited to three supplied studies with different tasks and evaluation settings, reported metrics are compared descriptively rather than treated as directly equivalent.
