FakeIDCA: Fake news detection with incremental deep learning based concept drift adaption

Social media facilitates rapid information sharing, improving exposure, connections, and content promotion. However, it also poses the challenge of fake news, which can mislead and harm individuals physically, and mentally, and incite violence. Fake news is often known as incorrect or misleading inf...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Multimedia tools and applications Jg. 83; H. 10; S. 28579 - 28594
Hauptverfasser: Suryawanshi, Shubhangi, Goswami, Anurag, Patil, Pramod
Format: Journal Article
Sprache:Englisch
Veröffentlicht: New York Springer US 01.03.2024
Springer Nature B.V
Schlagworte:
Accuracy
Computer Communication Networks
Computer Science
Data Structures and Information Theory
Deep learning
Drift
Dynamic stability
Impact analysis
Machine learning
Multimedia Information Systems
Neural networks
News
Performance degradation
Special Purpose and Application-Based Systems
Deep learning
Incremental learning
Concept drift
Fake news
ISSN:1573-7721, 1380-7501, 1573-7721
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Social media facilitates rapid information sharing, improving exposure, connections, and content promotion. However, it also poses the challenge of fake news, which can mislead and harm individuals physically, and mentally, and incite violence. Fake news is often known as incorrect or misleading information. Prior research used Machine Learning (ML) and Deep Learning (DL) techniques for fake news detection. These studies predominantly relied on static offline models, overlooking the dynamic and evolving nature of news patterns, assuming their stability over time. Our paper proposes an incremental ensemble neural network for fake news detection that continuously learns from fake news streams, adapting to changes. It employs performance-based pruning to eliminate underperforming classifiers, improving overall performance. Additionally, the model detects concept drift in real-time and triggers adaptation strategies to maintain accuracy and robustness. The models undergo evaluation in two scenarios, utilizing consistent news patterns for training and testing, demonstrating consistent performance among all ML and incremental models. In the second scenario, the study analyzes the impact of news patterns over time, including concept drift due to significant events like the United States election. The analysis reveals that offline-trained methods are susceptible to performance degradation. However, the proposed model exhibits consistent performance with an accuracy of 97.90% and 99.76% on two fake news datasets, despite changes in the news pattern over time. The findings demonstrate how the evolution of the news pattern impacts the effectiveness of fake news detection models. The proposed model used for the experimentation indicates consistent performance even in the presence of drift.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:1573-7721
1380-7501
1573-7721
DOI:10.1007/s11042-023-16588-z