An Enhanced MIBKA-CNN-BiLSTM Model for Fake Information Detection

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Fake information detection faces challenges due to information complexity and inefficient parameter optimization in existing models.
• Current detection technologies struggle with accuracy and efficiency when dealing with sophisticated fake information.

Purpose of the Study:

• To propose a hybrid detection model, MIBKA-CNN-BiLSTM, that enhances fake information detection accuracy and efficiency.
• To improve the Black Kite Optimization Algorithm (MIBKA) with enhanced strategies for better parameter optimization.
• To develop an optimized dual-channel deep learning architecture for adaptive fake information detection.

Main Methods:

• Implemented a triple-strategy enhanced Black Kite Optimization Algorithm (MIBKA) with circle chaotic mapping, DE/rand-to-best/1 mutation, and logarithmic spiral opposition-based learning (LSOBL).
• Constructed a CNN-BiLSTM dual-channel feature extraction network with MIBKA-optimized hyperparameters for adaptive model alignment.
• Created a high-quality fake information dataset using social media platforms, including CCTV.

Main Results:

• The MIBKA-CNN-BiLSTM model achieved the highest accuracy on a self-built dataset, outperforming the optimal hybrid model by 3.11%.
• On the Weibo21 dataset, the model demonstrated improved performance with a 1.52% increase in accuracy and a 1.71% increase in F1-score compared to baseline models.
• The enhanced MIBKA algorithm showed improved parameter space coverage, exploration-exploitation balance, and dynamic opposition solution space exploration.

Conclusions:

• The MIBKA-CNN-BiLSTM model offers a practical and effective solution for detecting lightweight and robust false information.
• The proposed enhancements to the MIBKA algorithm and the dual-channel deep learning architecture significantly boost detection performance.
• This research provides a valuable contribution to the field of fake information detection through improved algorithmic strategies and model optimization.