Year: 2026
Type: article
Abstract: The swift use of Artificial Intelligence (AI) in public communication systems has significantly facilitated the processing of vast data sets, predictive analytics, and automated decision-making. However, as we increasingly depend on complex black-box models, it raises significant concerns regarding transparency, accountability, and public trust. Explainable Artificial Intelligence (XAI) techniques such as SHAP and LIME provide feature-level explanations; nonetheless, these outputs are predominantly numerical, reliant on cases, and challenging for non-technical individuals to comprehend. This limitation is particularly challenging in public communication contexts, as interpretability must extend beyond technical clarity to encompass semantic coherence and collective comprehension. This research presents a Hybrid Explanation Framework utilizing Semantic Clustering for Public Communication Systems, integrating quantitative explanation methodologies with semantic aggregation tactics. The system features a multi-layer architecture that transforms prediction outputs into explanation vectors, projects them into an explanation space, and subsequently use semantic clustering to extract higher-order interpretability patterns. The proposed methodology integrates explanation vectors to identify the thematic structures underlying model decisions, distinguishing it from conventional XAI methods that emphasize singular explanations. This enables the transition from first-order interpretability ("why was this decision made?") to second-order interpretability ("what semantic patterns define these decisions?"). The study employed a qualitative-conceptual research methodology, utilizing thematic analysis, comparative model analysis, and conceptual synthesis to construct the framework. The results demonstrate that semantic clustering enhances interpretability consistency, facilitates user understanding, and strengthens perceived trust compared to instance-level explanations alone. The divergence between data clustering and explanation clustering demonstrates that statistical similarity does not necessarily imply similarity in decision-making, highlighting the importance of structural interpretability in AI governance. The suggested methodology improves XAI by integrating semantic and communicative components, offering a scalable solution for transparent AI-driven public discourse. This method enhances the accountability, comprehensibility, and reliability of AI systems in public spaces by linking mathematical models to human cognition.
DOI: 10.51699/24epym13
Institution: Sekolah Tinggi Multi Media "MMTC" Yogyakarta
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7219916697