Advanced artificial intelligence architectures for the proactive search and classification of manipulative information in digital discourse

DOI: 10.31673/2412-9070.2026.318102

Abstract

The subject of this research is the development of an automated framework for the detection and classification of manipulative information within digital ecosystems, leveraging hybrid machine learning architectures and explainable artificial intelligence (XAI). In an era characterized by the proliferation of computational propaganda and generative AI-driven disinformation, traditional reactive detection methods are increasingly insufficient. This article introduces a novel, multi-layered architecture — the Cognitive-Linguistic Manipulation Analysis Framework (CLMAF) — designed to identify non-transparent influence attempts by integrating linguistic pragmatics, stylometric profiling, and cross-modal consistency checking. The primary objective is to enhance the interpretability of detection models, thereby fostering user trust and enabling proactive moderation of harmful content. The methodology employs an ensemble of transformer-based models (specifically fine-tuned BERT and RoBERTa architectures) integrated with Graph Neural Networks (GNN) to analyze both the semantic content and the structural propagation patterns of potential manipulation. The scientific novelty lies in the synthesis of a stakeholder-centric, multimodal architecture that moves beyond binary veracity classification toward a nuanced identification of psychological manipulation techniques. Findings suggest that the integration of XAI not only improves the transparency of AI-driven decisions but also enhances the overall robustness of the system against adversarial attacks. The proposed framework effectively bridges the gap between high-performance "black-box" neural networks and the necessity for human-centric accountability in information security. Future research directions include the empirical validation of the CLMAF architecture against evolving generative threats and the refinement of cross-lingual manipulation markers. The text further elaborates on the mathematical foundations of the verification probability model and the similarity-based retrieval mechanism, providing a comprehensive blueprint for next-generation information defense systems.

Keywords: manipulative information, artificial intelligence, machine learning, explainable AI (XAI), computational linguistics, digital disinformation, transformer models, neural network architectures, software engineering, data processing.