# IBKA-MSM: A Novel Multimodal Fake News Detection Model Based on Improved Swarm Intelligence Optimization Algorithm, Loop-Verified Semantic Alignment and Confidence-Aware Fusion

## TL;DR

This paper introduces a new fake news detection model that uses improved swarm intelligence and semantic alignment to achieve high accuracy.

## Contribution

A novel multimodal fake news detection framework combining improved swarm intelligence optimization with semantic alignment and fusion techniques.

## Key findings

- IBKA-MSM achieves 95.80% accuracy in fake news detection.
- The model improves F1 score by 2.8% over PSO and 1.6% over BKA.
- The framework effectively maintains multimodal semantic consistency.

## Abstract

With the proliferation of social media platforms, misinformation has evolved toward more diverse modalities and complex cross-semantic correlations. Accurately detecting such content, particularly under conditions of semantic inconsistency and uneven modality dependency, remains a critical challenge. To address this issue, we propose a multimodal semantic representation framework named IBKA-MSM, which integrates swarm-intelligence-based optimization with deep neural modeling. The framework first employs an Improved Black-Winged Kite Algorithm (IBKA) for discriminative feature selection, incorporating adaptive step-size control, an elite-memory mechanism enhanced by opposition perturbation, Gaussian-based local exploitation, and population diversity regulation through reinitialization. In addition, a Modality-Generated Loop Verification (MGLV) mechanism is designed to enhance semantic alignment, and a Semantic Confidence Matrix with Modality-Coupled Interaction (SCM-MCI) is introduced to achieve adaptive multimodal fusion. Experimental results demonstrate that IBKA-MSM achieves an accuracy of 95.80%, outperforming mainstream hybrid models. The F1 score is improved by approximately 2.8% compared to PSO and by 1.6% compared to BKA, validating the robustness and strong capability of the proposed framework in maintaining multimodal semantic consistency for fake news detection.