[Paper Review] TI-CNN: Convolutional Neural Networks for Fake News Detection

Yang Yang, Lei Zheng+4 more|arXiv (Cornell University)|Jun 3, 2018

Misinformation and Its Impacts28 references234 citations

TL;DR

TI-CNN combines explicit and latent text and image features via dual CNN branches to detect fake news, outperforming baselines on a real-world dataset.

ABSTRACT

With the development of social networks, fake news for various commercial and political purposes has been appearing in large numbers and gotten widespread in the online world. With deceptive words, people can get infected by the fake news very easily and will share them without any fact-checking. For instance, during the 2016 US president election, various kinds of fake news about the candidates widely spread through both official news media and the online social networks. These fake news is usually released to either smear the opponents or support the candidate on their side. The erroneous information in the fake news is usually written to motivate the voters' irrational emotion and enthusiasm. Such kinds of fake news sometimes can bring about devastating effects, and an important goal in improving the credibility of online social networks is to identify the fake news timely. In this paper, we propose to study the fake news detection problem. Automatic fake news identification is extremely hard, since pure model based fact-checking for news is still an open problem, and few existing models can be applied to solve the problem. With a thorough investigation of a fake news data, lots of useful explicit features are identified from both the text words and images used in the fake news. Besides the explicit features, there also exist some hidden patterns in the words and images used in fake news, which can be captured with a set of latent features extracted via the multiple convolutional layers in our model. A model named as TI-CNN (Text and Image information based Convolutinal Neural Network) is proposed in this paper. By projecting the explicit and latent features into a unified feature space, TI-CNN is trained with both the text and image information simultaneously. Extensive experiments carried on the real-world fake news datasets have demonstrate the effectiveness of TI-CNN.

Motivation & Objective

• Motivate the study of fake news detection in social networks and its impact on information credibility.
• Investigate whether both text and image information improve fake news detection.
• Develop a unified model (TI-CNN) that fuses explicit and latent text and image features.
• Evaluate TI-CNN against baseline methods on a real-world dataset of fake and real news.

Proposed method

• Propose TI-CNN with two parallel branches for text and image information.
• Extract explicit features from text (statistics like word counts, punctuation, capitalization, etc.) and image (resolution, facial content).
• Learn latent features using convolutional neural networks on text (CNN over word embeddings) and images (CNN over image patches).
• Project explicit and latent features into a unified space and fuse text and image representations for final prediction.
• Train end-to-end using negative log-likelihood with RMSprop and standard regularization (dropout, L2, early stopping).

Experimental results

Research questions

• RQ1 Can explicit textual/image features be effectively combined with latent CNN-derived features for fake news detection?
• RQ2 Does coupling text and image information improve detection performance over using either modality alone?
• RQ3 What is the contribution of latent features learned by CNNs versus hand-crafted explicit features?
• RQ4 How does TI-CNN perform relative to traditional text-only models and image-only models on real-world data?

Key findings

• The dataset contains 20,015 articles: 11,941 fake and 8,074 real.
• TI-CNN outperforms baseline methods when using both text and image information.
• Image information alone is insufficient for reliable fake news detection.
• Text-based methods (e.g., logistic regression) are weaker on this data, while deep text models (GRU/LSTM) have limitations on long sequences.
• The model blends explicit and latent features from both modalities into a unified representation, achieving superior performance over baselines.