A TWO-STAGE MACHINE LEARNING FRAMEWORK FOR AUTOMATIC JOB TITLE IDENTIFICATION FROM ONLINE RECRUITMENT ADVERTISEMENTS

Abstract

Online recruitment platforms host millions of job postings daily, making accurate job title identification critical for job matching, market analysis, and automated HR analytics. However, job titles in postings are often ambiguous, verbose, or embedded within unstructured descriptions, making extraction challenging. This study proposes a two-stage job title identification system that combines text preprocessing, semantic classification, and deep learning-based title normalization. In Stage 1, a contextual classifier identifies candidate title segments from raw advertisements. In Stage 2, a transformer-based normalization model maps extracted titles to standardized occupational taxonomies. Experimental evaluation demonstrates improved accuracy, reduced ambiguity, and enhanced generalization across diverse job categories compared to traditional rule-based or single-stage ML systems. This framework addresses inconsistencies in online job advertisements and provides scalable automation for recruitment analytics [1], [2], [3], [6], [14].