In order to meet the need for scalability in academic advising in higher education settings while ensuring compliance, fairness, and privacy, we propose a governance-ready multi-agent reinforcement learning framework that aggregates the advantages of four highly specialized agents, namely, the Plan Validator, Risk Detector, Resource Recommender, and Conversational Advisor, which function collaboratively with a policy learner and under the supervision of a human. In a quasi-experimental mixed-methods study involving n = 345 participants from different departments, the system yielded a substantial improvement over the baseline in terms of efficiency and quality, where the turnaround time in academic advising was reduced by ≈60% or from 48.6 to 19.3 hours, while compliance was boosted by +11.3% and risk detection recall was enhanced by +0.12 or from 0.67 to 0.79. The stability of the learning process was satisfactory, with a normalized cumulative reward of +0.35 and around 1,200 episodes to converge. For the governance control, the effectiveness of the proofs was satisfactory, as the fairness gap reduced from 7.8% to 3.2%, while the privacy incidents remained zero even with continuous audits. In terms of evaluation, a mixed methods evaluation design was used, which included the performance records of the RL, ANOVA subgroup, and trust judgments to establish validity. From the findings, the MARL solution is technically viable, ethical, and feasible as an advisory mechanism in HEIs. Suggestions are made for future directions to investigate the solution's scalability and dynamic reward shaping for various policies.

Traditional folktales are vital carriers of cultural identity, yet preserving and analyzing them in the digital age poses challenges. In the Philippines, Biliran folk narratives embody deep cultural insights that demand linguistically and culturally sensitive approaches. This study explores deep learning techniques for analyzing thematic elements in these narratives while maintaining authenticity. Specifically, it examines how multilingual language models can be adapted to identify, interpret, and correlate themes within culturally specific texts. The research employed a progressive fine-tuning strategy in four stages, adapting mBERT, XLM-RoBERTa, and GPT models to Biliran narratives. The methodology included: (1) constructing a corpus of 232 paragraphs, (2) thematic annotation by three trained annotators (κ = 0.78), (3) designing model architectures that integrate cultural knowledge, and (4) evaluating performance through automated metrics and expert assessments. The ensemble approach outperformed baselines, achieving an F1-score of 0.86 and reducing perplexity by 59%. Thematic classification identified Cultural Values as the most dominant theme (f1 = 0.90). Statistical analysis revealed significant correlations, particularly between Cultural Values and Family Relationships (r = 0.73, p < 0.01). Cultural experts validated the models’ effectiveness, rating authenticity at 4.5/5.0 and thematic coherence at 4.6/5.0. The study contributes: (1) a methodological framework for culturally sensitive AI analysis, (2) empirical proof that progressive fine-tuning enhances model performance, (3) statistical insights into thematic relationships reflecting Biliran cultural systems, and (4) practical methods for adapting multilingual models to low-resource contexts. These findings advance cultural heritage preservation and AI applications in diverse settings.

This paper presents an innovative ensemble method for developing a student policy chatbot that leverages advanced natural language processing techniques, specifically an extractive question-answering approach combined with a large language model (LLM). The study aims to enhance the chatbot’s ability to accurately understand and respond to student inquiries regarding university policies and procedures. Utilizing a dataset derived from the University Handbook and Code of Decorum for Genderbased Sexual Harassment, the chatbot architecture integrates the RoBERTa model for extractive answering and GPT-2 for natural language generation. Evaluation metrics, including automatic assessments and human evaluations, reveal the chatbot’s strengths in relevance and coverage, while also identifying areas for improvement in precision and coherence. The findings underscore the potential of AI-driven chatbots in higher education to enhance student engagement and support, while recommendations for future enhancements are necessary to ensure the chatbot’s effectiveness in addressing the diverse needs of students.

This study applies machine learning (ML) algorithms within the Knowledge Discovery in Databases (KDD) framework to analyze Bachelor of Science in Information Systems (BSIS) graduate tracer study data. Two core tasks were addressed: (1) predicting structured graduate outcomes, such as employment status, and (2) classifying sentiments from textual curriculum feedback. Data selection, preprocessing, and transformation prepared structured and unstructured inputs for supervised ML experiments conducted in Python 3.x using Google Colab, with scikit-learn supporting classical algorithms and TensorFlow/Keras enabling deep learning deployment. Algorithms evaluated included Logistic Regression, Random Forest, Gradient Boosting, Support Vector Machines, K-Nearest Neighbors, Naive Bayes, and Long Short-Term Memory (LSTM) networks. Results showed that Logistic Regression achieved the best performance for predicting employment status, with 74.19% accuracy, precision of 0.82, and the highest Kappa score (0.54), outperforming all other models. Gradient Boosting emerged as the most effective approach for sentiment analysis, attaining balanced results across recall (0.497), F1-score (0.481), and Kappa (0.441). Insights from confusion matrices revealed that Logistic Regression excelled at distinguishing employed graduates but struggled with underemployed and unemployed categories, while Gradient Boosting strongly identified neutral sentiments but faced challenges in detecting negative feedback. The findings highlight that algorithm selection should be task-specific: Gradient Boosting is more effective for sentiment analysis, whereas classical models like Logistic Regression are better suited for structured prediction. A hybrid framework combining both is recommended to enhance tracer study systems, enabling richer, data-driven insights for curriculum evaluation, institutional planning, and graduate outcome monitoring.

The growing adoption of open and distance learning environments has resulted in the massive generation of educational data that require systematic and interpretable analysis. This study applies the Knowledge Discovery in Databases (KDD) framework to predict student academic performance using public datasets from Kaggle and the UCI Machine Learning Repository. Through the structured phases of KDD—data selection, preprocessing, transformation, data mining, and interpretation—the study demonstrates how methodical knowledge discovery enhances both predictive accuracy and explainability. Machine learning algorithms such as Random Forest (RF), Decision Tree (DT), Logistic Regression (LR), and Neural Network (NN) were tested and compared. Results show that the Random Forest model achieved the highest accuracy (91.4%) and F1-score (0.89). The findings confirm that KDD is not merely a preprocessing mechanism but a methodological framework that transforms raw data into reliable, interpretable, and actionable educational insights.