All Articles
January 15, 2025 • 2 min read

Multimodal Information Credibility Classification Using CNN-BiGRU with Particle Swarm Optimization

An IEEE-published thesis proposing a multimodal hybrid deep learning approach (CNN-BiGRU + PSO) to classify information credibility on social media X (Twitter) by fusing tweet text, visual images, and user metadata.

Multimodal Information Credibility Classification Using CNN-BiGRU with Particle Swarm Optimization

Abstract

This research proposes a multimodal hybrid deep learning approach combining CNN-BiGRU (Convolutional Neural Network - Bidirectional Gated Recurrent Unit) with Particle Swarm Optimization (PSO) to classify information credibility on social media platform X (formerly Twitter). The approach fuses three modalities: tweet text features (TF-IDF & GloVe), visual image features (MobileNetV1), and user metadata for joint classification.

Research Context

The proliferation of misinformation on social media platforms poses significant challenges to public discourse and decision-making. Traditional text-only approaches for credibility classification often fail to capture the multimodal nature of social media content, where posts frequently combine text, images, and contextual metadata.

Technical Approach

Dataset

  • 23,564 annotated tweets with balanced 50:50 distribution (credible vs. non-credible)
  • Each tweet paired with associated images and user metadata
  • GloVe Corpus: 62,274 text records combining tweets with 38,710 IndoNews articles for domain-specific word embeddings

Multimodal Fusion Architecture

The proposed architecture integrates three feature extraction streams:

  1. Text Features: Dual representation using TF-IDF for statistical features and GloVe embeddings for semantic understanding
  2. Visual Features: MobileNetV1 for lightweight yet effective image feature extraction
  3. User Metadata: Structured features capturing account characteristics and behavioral patterns

PSO Hyperparameter Optimization

Particle Swarm Optimization was employed for automated hyperparameter tuning, optimizing:

  • Learning rate
  • Dense layer size
  • Dropout rate

Results

ModelAccuracy
CNN-BiGRU + PSO (Best)79.09%
BiGRU-CNN + PSO77.93%

Detailed Performance Metrics

  • Precision (Credible): 80.38%
  • Recall (Non-Credible): 80.97%
  • Multimodal integration improved accuracy by +4.95% compared to text-only baseline

Key Contributions

  1. Demonstrated that multimodal fusion significantly outperforms single-modality approaches for credibility classification
  2. Validated PSO as an effective hyperparameter optimization strategy for deep learning architectures
  3. Created domain-specific GloVe embeddings combining social media text with news articles for improved contextual understanding
  4. Published and peer-reviewed at IEEE International Conference on ICT for Smart Society (ICICyTA)

Publication Details

  • DOI: 10.1109/ICICyTA68677.2025.11362759
  • Conference: IEEE International Conference on ICT for Smart Society (ICICyTA)
  • University: Telkom University
  • Grade: Summa Cum Laude (GPA 3.96/4.00)

Tech Stack

Python, PyTorch, TensorFlow, CNN, BiGRU, GloVe, TF-IDF, MobileNetV1, Particle Swarm Optimization

Last updated on September 16, 2026 at 11:27 AM UTC+7. See Changelog