Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
80 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
7 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

FANG: Leveraging Social Context for Fake News Detection Using Graph Representation (2008.07939v2)

Published 18 Aug 2020 in cs.SI, cs.CL, cs.IR, and cs.LG

Abstract: We propose Factual News Graph (FANG), a novel graphical social context representation and learning framework for fake news detection. Unlike previous contextual models that have targeted performance, our focus is on representation learning. Compared to transductive models, FANG is scalable in training as it does not have to maintain all nodes, and it is efficient at inference time, without the need to re-process the entire graph. Our experimental results show that FANG is better at capturing the social context into a high fidelity representation, compared to recent graphical and non-graphical models. In particular, FANG yields significant improvements for the task of fake news detection, and it is robust in the case of limited training data. We further demonstrate that the representations learned by FANG generalize to related tasks, such as predicting the factuality of reporting of a news medium.

Overview of "FANG: Leveraging Social Context for Fake News Detection Using Graph Representation"

In the paper "FANG: Leveraging Social Context for Fake News Detection Using Graph Representation," the authors present a novel approach to detecting fake news by exploiting the social context in which the information is disseminated. Their proposed framework, Factual News Graph (FANG), utilizes a graph representation to model major social actors and their interactions in a manner that enhances the task of fake news detection. The work emphasizes the importance of representation learning and addresses several limitations found in previous approaches that relied heavily on targeted performance without adequately modeling the social context.

Framework and Methodology

The FANG model is built on a comprehensive graph that incorporates three major types of social entities: questionable news articles, news sources, and social users. It leverages seven types of interactions, including followership among users, news publication by sources, and users’ stance toward news articles. The framework stands out for its ability to capture temporal patterns and incorporate them into the graph learning process. This is achieved through a Bi-LSTM network that extracts temporal features from user engagement sequences, complemented by an attention mechanism that dynamically focuses on relevant user interactions.

In terms of representation learning, FANG integrates unsupervised proximity loss, self-supervised stance loss, and supervised fake news detection loss. This holistic approach ensures that the resulting embeddings are not only effective for detecting fake news but also generalizable for related tasks, such as assessing media source credibility.

Experimental Evaluation

The authors conducted extensive experimentation on a dataset comprising various social media engagements, factuality-checking website annotations, and citation data. The results demonstrated that FANG significantly outperforms both Euclidean models, such as Support Vector Machines, and other graph-based approaches, like Graph Convolutional Networks, in fake news detection, especially under limited training data conditions. Notably, FANG's use of engagement temporality proved crucial, allowing the model to differentiate between fake and real news patterns effectively.

Implications and Future Directions

The implications of this research are multifaceted. Practically, FANG provides a robust tool for enhancing the accuracy of automated fake news detection systems deployed on social media platforms, which are inundated with rapidly disseminated information. Theoretically, this work advances the understanding of how social context can be effectively encoded into graph models, opening pathways for further investigations into social behavior modeling and network-based learning.

FANG's adaptability to minimal supervision and its inductive nature in handling unseen nodes offer scalability advantages, critical in dynamic social network environments. Future research avenues include refining the stance detection process, expanding the methodology to integrate multimodal data (e.g., images and videos), and further developing the framework to address multiple related tasks through multi-task learning paradigms. This could involve joint optimization for a range of social network analysis challenges, such as echo chamber detection and misinformation propagation analysis.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (4)
  1. Van-Hoang Nguyen (1 paper)
  2. Kazunari Sugiyama (14 papers)
  3. Preslav Nakov (253 papers)
  4. Min-Yen Kan (92 papers)
Citations (262)