A DUAL-STAGE SYSTEM FOR ACCURATE JOB TITLE IDENTIFICATION IN ONLINE JOB ADVERTISEMENTS

Abstract

Job advertisements contain valuable information about employment opportunities, skills, and
industry trends. However, extracting job titles accurately from unstructured job postings remains
a challenge due to variations in terminology, abbreviations, and inconsistencies in formatting.
This paper proposes a dual-stage system that enhances job title identification accuracy by
combining natural language processing (NLP) techniques and machine learning-based
classification. In the first stage, a text preprocessing module is employed to clean and normalize
job descriptions by removing stopwords, stemming, and handling variations. The second stage
utilizes deep learning models such as BERT and LSTM to classify and extract job titles with
high precision. Experimental results on real-world job postings demonstrate that the proposed
system significantly outperforms traditional keyword-based approaches, achieving improved F1-
score and recall.