An Intelligent Learning Framework for Predictive Modelling in AI-Driven Career Guidance

Abstract

Traditional career guidance systems often rely on static keyword matching and rigid rule-based filters, which are limited in their ability to capture the nuanced aspirations, evolving skill sets, contextual preferences, and emotional states of users. Such approaches often produce generic recommendations that lack personalization and adaptability. To address these limitations, this paper proposes an Intelligent Learning Framework for predictive modelling in AI-driven career guidance, designed to deliver highly personalized, context-aware, and continuously adaptive career recommendations. The proposed framework adopts a hybrid multi-agent architecture that integrates multiple intelligent components to enhance decision-making. Semantic understanding is achieved through Sentence-BERT (SBERT) combined with Cosine Similarity, enabling deep contextual matching between user profiles and career domains. A Rule-Based Engine is incorporated to enforce logical constraints, domain knowledge, and eligibility criteria, ensuring the validity and reliability of recommendations. Furthermore, Reinforcement Learning using Q-Learning is employed to dynamically refine the system’s predictions based on user feedback, interaction patterns, and long-term learning objectives. To further enhance user engagement, the system incorporates emotion-aware interaction using VADER Sentiment Analysis, allowing it to interpret user sentiments from textual inputs and adjust recommendations accordingly. This ensures a more empathetic and user-centric guidance experience. The system is implemented using a Flask-based backend integrated with a MySQL database, supporting scalable deployment, efficient data handling, and modular system design. Additionally, the framework emphasizes explainability by providing transparent reasoning behind each recommendation, thereby improving user trust and interpretability. The model considers multiple input dimensions, including user interests, academic background, skills, and preferences, to generate holistic career recommendations. Experimental results show that the proposed hybrid approach outperforms traditional systems in accuracy, adaptability, and user satisfaction, achieving a Top-3 Hit Rate of 90.5%. The integration of Q-Learning enables continuous improvement by adapting to user feedback and evolving trends. Overall, the framework is robust, scalable, and suitable for real-world deployment in educational and career guidance platforms.