A dual-layered artificial intelligence solution for classifying disinformation in socially oriented systems

The detection of fabricated information on interactive social platforms has gained significant academic and regulatory attention.During social instability, such disinformation presents substantial risks to individuals and society.Disinformation varies in impact, from harmless humor to content threatening societal stability.This study focuses on textual news content due to limitations in generating convincing visual forgeries.For text classification, three classical approaches are utilized: probabilistic models, neural networks, and polynomial models.Previous research has shown that hybrid recurrent-convolutional network (RCNN) offers superior binary classification performance, yet multi-classification across diverse disinformation categories remains unresolved.This paper establishes a classification framework categorizing content through the RCNN model into five classes: explicit satire, subtle humor, content targeting individuals, regionally harmful news, and globally impactful disinformation.Based on this framework, three data categorization models were developed using neural networks, naive Bayes classification, and polynomial algorithms.Experimental evaluation measured accuracy, processing efficiency, and data reduction requirements to achieve >80% accuracy.Results demonstrate that dual-layer implementations achieved approximately 20% improved effectiveness compared to standalone approaches.The RCNN-naive Bayes hybrid exhibited optimal accuracy and processing speed, showing considerable potential for highthroughput systems requiring swift responses to misinformation.These findings represent a significant advancement in automated information verification methodologies, establishing a foundation for future development in complex classification tasks.