Computational Propaganda Detection in Social Media

Computational propaganda detection in social media encompasses the automated identification and analysis of misleading or manipulative messages disseminated through digital platforms. By leveraging advances in natural language processing and machine learning, researchers build systems capable of flagging content that aims to influence public opinion, polarise communities or propagate ideological narratives. Core methodologies include feature‐based approaches, which extract linguistic, syntactic and sentiment indicators; deep learning frameworks that employ contextual word embeddings; and hybrid models combining both paradigms. Addressing multilingual settings and resource‐scarce languages has become a priority, as has enhancing model transparency to ensure accountability in moderation. Key challenges include the dynamic evolution of adversarial tactics, class imbalance in annotated corpora and the need for multimodal analysis of text, images and video. Globally, effective detection tools support democratic resilience by enabling platforms, regulators and civil society to monitor disinformation campaigns, uphold content integrity and foster informed public discourse.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Computational Propaganda Detection in Social Media publication trend

The graph below shows the total number of articles in computational propaganda detection in social media across all publications each year (not limited to Nature Index journals).

Technical terms

Computational propaganda: The use of automated methods and algorithms to create, distribute or amplify misleading information with intent to influence public opinion.

Transformer: A deep learning architecture that models sequential data using self-attention mechanisms to capture contextual relationships in text.

Feature engineering: The process of selecting, transforming and combining raw data attributes into informative input variables for machine learning models.

Class imbalance: A situation in classification tasks where one category (e.g., propaganda) is under-represented relative to others, potentially biasing model performance.

Binary classification: A supervised learning task in which the model predicts one of two possible labels, such as “propaganda” or “non-propaganda.”

References

1. How to detect propaganda from social media? Exploitation of semantic and fine-tuned language models. PeerJ Computer Science (2023).
2. Automated Multilingual Detection of Pro-Kremlin Propaganda in Newspapers and Telegram Posts. Datenbank-Spektrum (2023).
3. HAPI: An efficient Hybrid Feature Engineering-based Approach for Propaganda Identification in social media. PLOS ONE (2024).