A Multimodal Adaptive Graph-based Intelligent Classification Model for Fake News

Junhao

2024-11-09

Abstract:Numerous studies have been proposed to detect fake news focusing on multi-modalities based on machine and/or deep learning. However, studies focusing on graph-based structures using geometric deep learning are lacking. To address this challenge, we introduce the Multimodal Adaptive Graph-based Intelligent Classification (aptly referred to as MAGIC) for fake news detection. Specifically, the Encoder Representations from Transformers was used for text vectorization whilst ResNet50 was used for images. A comprehensive information interaction graph was built using the adaptive Graph Attention Network before classifying the multimodal input through the Softmax function. MAGIC was trained and tested on two fake news datasets, that is, Fakeddit (English) and Multimodal Fake News Detection (Chinese), with the model achieving an accuracy of 98.8\% and 86.3\%, respectively. Ablation experiments also revealed MAGIC to yield superior performance across both the datasets. Findings show that a graph-based deep learning adaptive model is effective in detecting multimodal fake news, surpassing state-of-the-art methods.

Artificial Intelligence

What problem does this paper attempt to address?

### What problem does this paper attempt to solve?

This paper aims to solve the problem of multimodal fake news detection, especially by using graph - structure - based geometric deep learning methods to improve the detection effect. Specifically, the paper proposes a model named **Multimodal Adaptive Graph - based Intelligent Classification (MAGIC)** for detecting fake news on social media.

#### Background and Challenges

With the rapid development of online social networks (OSNs), the speed of information generation and dissemination has been greatly accelerated, which enables fake news to spread rapidly. Fake news refers to deliberately spread misleading, false or incorrect information, including rumors, scams and conspiracy theories, etc. These contents usually rely on unverified and deliberately misleading materials, and have a complex impact on social norms and personal lives.

Although a large number of studies have used machine learning and deep learning methods to detect fake news, most of the studies mainly focus on unimodal data (such as text or image), or only use graph structures for analysis. However, methods that combine multimodal inputs (text and image) and utilize graph structures are still relatively few. Among the existing studies, only a few works have attempted to combine multimodal inputs with graph - structure mechanisms, such as the heterogeneous attention network model proposed by Bi et al. [21].

#### Research Objectives

To fill this gap, this paper proposes the MAGIC model, which improves fake news detection in the following ways:

1. **Multimodal input processing**: Process text (including the original post and its comments) and image simultaneously.
2. **Graph - structure construction**: Use the Adaptive Graph Attention Network (GAT) to construct a comprehensive information interaction graph.
3. **Classification and reasoning**: Classify multimodal inputs through the Softmax function.

#### Main Contributions

- Propose a new multimodal adaptive graph - based intelligent classification model (MAGIC), achieving accuracies of 98.8% and 86.3% on two fake news datasets (Fakeddit and Multimodal Fake News Detection), respectively.
- Verify the superior performance of the MAGIC model on different datasets through ablation experiments.
- The results show that the graph - structure - based deep learning adaptive model performs excellently in detecting multimodal fake news, surpassing existing methods.

In conclusion, this paper is committed to developing a more effective multimodal fake news detection method to deal with the increasingly complex problem of false information dissemination.