Otomatisasi Identifikasi Kesenjangan Keahlian Kerja Melalui Analisis Komparatif Data Lowongan Dan Profil Kandidat

Abstract

This study addresses the persistent skill mismatch faced by students and new graduates in the Indonesian labor market by proposing an integrated automation framework for skill gap identification, developed as part of the TalentIQ career analytics platform. The framework combines Natural Language Processing, semantic matching, and rule-based set-comparison techniques across three datasets: 10,766 job postings, 10,720 synthetic candidate profiles, and 5,663 online courses. After text normalization and skill standardization, all job and candidate records were encoded into 384-dimensional embeddings using Sentence-BERT (all-MiniLM-L6-v2). A Two-Tower Deep Neural Network was trained on cosine-similarity-based pseudo-labels to predict candidate-job match probability, while a rule-based module compared explicit skill sets using a canonical dictionary of over 250 skills to compute matched, missing, and extra skills. Both scores were combined into a Hybrid Readiness Score (HRS). Results show that Operations & Management (36.6%), Video/Content Creator (14.2%), and Web Developer (9.1%) dominate job demand, while English, Information Architecture, and Excel are the most requested skills. The Two-Tower DNN achieved 98.60% accuracy (MAE = 0.0153) on pseudo-labeled test data. Evaluation on 500 candidate-job pairs revealed a polarized readiness distribution: 44.2% "Not Ready," 31.4% "Fairly Strong," 23.8% "Very Strong," and only 0.6% "Needs Improvement," with an average rule-based skill coverage of 21.95%. The most frequent missing skills were information architecture, social media, English, teamwork, and content creation. These findings demonstrate that combining deep semantic matching with explicit rule-based comparison produces an interpretable and actionable readiness measure, offering practical guidance for job seekers and curriculum development in Indonesia's digital creative sector.