A Review of Web Infodemic Analysis and Detection Trends across Multi-modalities using Deep Neural Networks

Chahat Raj, Priyanka MeelarXiv 2021 · OpenReview

TL;DR

This paper reviews the rapidly evolving field of multi-modal fake news detection, highlighting the increasing sophistication of deep learning techniques that combine text, images, and videos. It identifies critical research gaps, particularly in the availability of robust datasets and the need for real-time, ubiquitous detection systems, paving the way for future advancements in combating online misinformation.

What Problem Does This Paper Address?

The internet and social media have become fertile ground for the rapid spread of "fake news" and misinformation, a phenomenon amplified by the COVID-19 pandemic and termed an "infodemic." While early research focused on detecting fake news solely through text, the reality is that misinformation is increasingly disseminated across multiple modalities: text, images, and videos, often in combination. This multi-modal nature makes detection significantly more challenging, as a single piece of fake news can leverage the visual impact of an image or video to reinforce a false textual narrative. Existing surveys often focus on single modalities, leaving a gap in a comprehensive understanding of multi-modal fake news detection techniques, their limitations, and future research directions.

Methods

This review categorizes fake news detection methods into several key areas:

  • Deep Learning: Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), including LSTMs, are highlighted as dominant techniques for analyzing visual and sequential data, respectively. These models are often pre-trained on large datasets like ImageNet and then fine-tuned for fake news detection.
  • Image/Video Forensics: Techniques that analyze the integrity of visual media, such as detecting tampered images (e.g., copy-move forgery, splicing) and manipulated videos (e.g., deepfakes), are crucial.
  • Explicit Features: This category encompasses a wide range of features derived from content and user behavior, including statistical features (likes, shares), semantic analysis, user profile information, propagation patterns, and geolocation.
  • Reverse Engineering: Primarily through reverse image search, this method verifies the authenticity and origin of images by matching them with existing online occurrences.
  • Data Forensics: This involves techniques to detect manipulations in images and videos without requiring original versions, covering forgery, splicing, face-swapping, and more.

The paper emphasizes that combining these methods, particularly through multi-modal approaches that fuse information from different modalities, leads to more effective detection.

Datasets

The paper identifies a significant challenge: the lack of large-scale, high-quality multi-modal datasets. While text-based datasets are more numerous, those incorporating text, images, and videos are limited and often of poor quality. The review lists several benchmark datasets used in the reviewed articles, including:

  • MediaEval: Focuses on tweet and image data related to specific events.
  • FakeNewsNet: Includes news content, social context, and article URLs.
  • Fakeddit: A Reddit dataset with text, image, and metadata.
  • Newsbag/Newsbag++: Datasets containing news articles and images, with Newsbag++ utilizing data augmentation.
  • FVC4: A corpus specifically for fake video detection.
  • Twitter, Sina Weibo, Politifact, GossipCop, CrawlerNews: Datasets derived from social media platforms and news sources, often containing text and images.
  • WhatsApp Dataset: Contains fake images extracted from WhatsApp groups.

The authors note that video datasets are particularly rare and often contain limited samples.

Key Insights and Ideas

  • The "Infodemic" is Multi-modal: Fake news is no longer just text; it's a sophisticated blend of text, images, and videos designed to deceive. This necessitates multi-modal detection approaches.
  • Deep Learning Dominance: CNNs and RNNs are the workhorses for analyzing visual and sequential data, respectively, and their combination is proving highly effective.
  • Dataset Scarcity is a Bottleneck: The lack of comprehensive, real-world multi-modal datasets is a major impediment to advancing fake news detection research.
  • Forensics is Crucial for Visuals: Detecting image and video manipulation is a critical component of multi-modal fake news detection, as visuals are often used to lend credibility to false narratives.
  • User Behavior Matters: Features derived from user profiles and propagation patterns can provide valuable signals for identifying fake news.
  • Real-time Detection is the Holy Grail: Current systems often detect fake news after it has spread widely. The future lies in developing real-time detection mechanisms.

Key Results

The paper is a survey, so it doesn't present new experimental results but rather synthesizes findings from 80 reviewed articles. Key takeaways from the literature include:

  • Performance Trends: The review shows a rising trend in research on visual and multi-modal fake news detection, particularly after 2016.
  • Accuracy Ranges: Performance metrics like F1-scores for combined textual and visual features often range from 80-95%. Accuracy for fake news classification tasks generally falls between 70-100%, with an average around 85%.
  • Methodology Distribution: CNNs and their combinations with other deep learning models (like RNNs) are prevalent, accounting for over 40% of methodologies.
  • Forensic Techniques: Image and video forensics methods are showing high accuracy in detecting manipulations, with some achieving over 99% accuracy.

Limitations and Future Work

Limitations:

  • Dataset Scarcity: A significant limitation is the lack of large-scale, high-quality multi-modal datasets, especially for video.
  • Limited Literature on Non-Text Modalities: While text-based detection is well-researched, work on audio, embedded content, and other modalities is less explored.
  • Real-time Detection Gap: Most existing systems detect fake news after it has spread, highlighting a need for real-time detection.
  • Ubiquitous Detection: Current models often focus on single platforms, while fake news spreads across multiple social networks.

Future Work:

  • Develop Large-Scale Multi-modal Datasets: The collection and curation of comprehensive datasets are paramount.
  • Real-time Detection Models: Building models that can detect fake news as it emerges, using fact-checked articles for training.
  • Early Detection Mechanisms: Developing systems that can identify and flag fake news before it gains widespread traction.
  • Ubiquitous Detection Models: Creating cross-platform systems that can detect fake content across various social media networks.
  • Data-oriented and Feature-oriented Detection: Developing systems that can verify content across all modalities and leverage a comprehensive set of features.
  • Integrity Assessment: Focusing on semantic integrity to detect unrelated but co-occurring modalities.
  • Embedded Fake News Detection: Addressing the growing spread of fake news within embedded content like memes.
  • Multilingual Detection: Expanding detection capabilities to regional languages.
  • Merging Data Manipulation and Fake News Detection: Integrating techniques from both fields for more robust detection.
  • User-Friendly Tools: Developing browser plugins, applications, and software for easier user-based detection.

Summary updated

Abstract

Fake news and misinformation are a matter of concern for people around the globe. Users of the internet and social media sites encounter content with false information much frequently. Fake news detection is one of the most analyzed and prominent areas of research. These detection techniques apply popular machine learning and deep learning algorithms. Previous work in this domain covers fake news detection vastly among text circulating online. Platforms that have extensively been observed and analyzed include news websites and Twitter. Facebook, Reddit, WhatsApp, YouTube, and other social applications are gradually gaining attention in this emerging field. Researchers are analyzing online data based on multiple modalities composed of text, image, video, speech, and other contributing factors. The combination of various modalities has resulted in efficient fake news detection. At present, there is an abundance of surveys consolidating textual fake news detection algorithms. This review primarily deals with multi-modal fake news detection techniques that include images, videos, and their combinations with text. We provide a comprehensive literature survey of eighty articles presenting state-of-the-art detection techniques, thereby identifying research gaps and building a pathway for researchers to further advance this domain.
BibTeX
@misc{raj2021review,
  author = {Chahat Raj and Priyanka Meel},
  title = {{A Review of Web Infodemic Analysis and Detection Trends across Multi-modalities using Deep Neural Networks}},
  year = {2021},
  eprint = {2112.00803},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2112.00803},
}
Figure