SpotFake+: A Multimodal Framework for Fake News Detection via Transfer Learning (Student Abstract)

Abstract

In recent years, there has been a substantial rise in the consumption of news via online platforms. The ease of publication and lack of editorial rigour in some of these platforms have further led to the proliferation of fake news. In this paper, we study the problem of detecting fake news on the FakeNewsNet repository, a collection of full length articles along with associated images. We present SpotFake+, a multimodal approach that leverages transfer learning to capture semantic and contextual information from the news articles and its associated images and achieves the better accuracy for fake news detection. To the best of our knowledge, this is the first work that performs a multimodal approach for fake news detection on a dataset that consists of full length articles. It outperforms the performance shown by both single modality and multiple-modality models. We also release the pretrained model for the benefit of the community.

CB-Fake: A multimodal deep learning framework for automatic fake news detection using capsule neural network and BERT

“…Furthermore, existing research is divided into two categories: unimodal and multimodal fake news identification. To detect fake news, the former uses either one of the content-based [ 1 , 2 , 5 , 7 , 9 , 18 , 22 , 25 , 30 , 31 , 33 , 41 , 47 , 54 , 55 ] or social-context-based features [ 13 , 23 , 24 , 26 , 27 , 38 , 50 , 51 ], while the latter uses a combination of any single modality feature [ 20 , 43 , 44 , 49 , 53 ].…”

“…Deep neural networks have been widely used for different multimodal data dependent tasks such as visual question answering [ 4 ], image captioning [ 19 ] and fake news detection [ 20 , 43 , 44 , 49 , 53 ]. Table 3 summarizes the existing multimodal fake news detection models.…”

“…Existing multimodal FND solutions have failed to improve the performance of the FND problem because they are unable to generate enhanced feature representation from the visual information of the news report. A convolution neural network (CNN) model called VGG-19 (Visual Geometry Group) has been used to capture visual features in several studies [ 20 , 43 , 44 , 49 , 53 ]. The challenges of CNN model are as follows: i) It requires lot of training data to improve the generalization of the model ii) It takes more training time iii) Due to the pooling operation, important information is lost.…”

Unsupervised Fake News Detection

“…Towards this direction, experiments were conducted by extracting features from two different modalities such as text and image. Works on that direction include EANN [37], MVAE [17], SpotFake [36] and SpotFake+ [35]. The EANN model, short for event adversarial neural networks, for multimodal fake news detection proposed by [37] consists of three sub-modules namely, textual feature extractor, visual feature extractor and an event discriminator module that when combined together is successful in detecting fake news.…”

“…This architecture too consists of three sub modules. Later methods for the task include SpotFake [36] and SpotFake+ [35]. SpotFake extracted features from both the text and image modality.…”