A Human AI Collaborative Knowledge Mining Framework for Enterprise Risk Detection and Managerial Decision Intelligence

Abstract

Enterprise risk detection increasingly requires analytical frameworks capable of integrating structured records, semi-structured logs, and unstructured organizational text into a unified managerial intelligence process. This study proposes a hybrid knowledge mining framework designed to detect enterprise risk patterns and transform them into prioritized, explanation-ready decision support for managers. The framework combines data preprocessing, semantic knowledge extraction, predictive risk scoring, hybrid evidence fusion, and managerial prioritization within one architecture. Experimental results showed that the proposed framework achieved an accuracy of 0.91, precision of 0.89, recall of 0.87, and F1-score of 0.88, outperforming the structured-only model with 0.82 accuracy and 0.78 F1-score, the text-only model with 0.79 accuracy and 0.75 F1-score, and the log-only model with 0.81 accuracy and 0.78 F1-score. Incremental fusion analysis further demonstrated that the full hybrid configuration produced the strongest performance, exceeding structured-only by 0.13 points in F1-score and surpassing pairwise combinations such as structured plus logs at 0.85 and structured plus text at 0.86. Class-level evaluation showed that the framework performed best on financial irregularity with precision 0.92, recall 0.89, and F1-score 0.90, followed by compliance deviation with F1-score 0.89 and operational delay with F1-score 0.86, while still maintaining reliable performance for supplier instability at 0.82 and communication risk at 0.81. The managerial prioritization layer generated differentiated action recommendations, distributing 44.3% of cases to monitoring, 29.5% to mitigation, 16.2% to escalation, and 10.0% to review, indicating that the framework can triage organizational risks without inflating urgency. Expert-oriented evaluation of explanation quality also showed strong operational usefulness, with average scores of 4.5 for clarity, 4.4 for traceability, 4.6 for relevance, 4.3 for actionability, and 4.4 for trust on a five-point scale. These findings indicate that the proposed framework extends enterprise risk analytics beyond standalone prediction by combining heterogeneous evidence fusion, contextual reasoning, and managerial decision intelligence in a single system. The study contributes a practical and scalable architecture for organizations seeking to convert fragmented enterprise data into actionable and interpretable risk governance.