MVFNN: Multi-Vision Fusion Neural Network for Fake News Picture Detection
During this year’s Novel Coronavirus (2019-nCoV) outbreak, the spread of fake news has caused serious social panic. This fact necessitates a focus on fake news detection. Pictures could be viewed as fake news indicators and hence could be used to identify
- PDF / 2,908,043 Bytes
- 8 Pages / 439.37 x 666.142 pts Page_size
- 45 Downloads / 185 Views
2
School of Software, Zhengzhou University, Zhengzhou 450002, China [email protected], [email protected] Zhongyuan Network Security Research Institute, Zhengzhou 450002, China
Abstract. During this year’s Novel Coronavirus (2019-nCoV) outbreak, the spread of fake news has caused serious social panic. This fact necessitates a focus on fake news detection. Pictures could be viewed as fake news indicators and hence could be used to identify fake news effectively. However, fake news pictures detection is more challenging since fake news picture identification is more difficult than the fake picture recognition. This paper proposes a multi-vision fusion neural network (MVFNN) which consists of four main components: the visual modal module, the visual feature fusion module, the physical feature module and the ensemble module. The visual modal module is responsible for extracting image features from images pixel domain, frequency domain, and tamper detection. It cooperates with the visual features fusion module to detect fake news images from multi-vision fusion. And the ensemble module combines visual features and physical features to detect the fake news pictures. Experimental results show that our model could achieve better detection performance by at least 4.29% than the existing methods in benchmark datasets.
Keywords: Fake news pictures
1
· Deep learning · Multi-vision domain
Introduction
The rise of social platforms such as Weibo and Twitter not only brings convenience to users, but also provides soil for the breeding and dissemination of fake news. The frantic spread of fake news has had many negative effects. Take the Novel Coronavirus (2019-nCoV) in 2020 as an example, the spread of various fake news caused serious social panic during the virus outbreak. Fake news seriously harms the harmony and stability of society [1,2], which necessitates the effective automated fake news detection [3–5]. The work was supported by the National Key Research and Development Program of China: No. 2018******400, and the training plan of young backbone teachers in colleges and universities of Henan Province. c Springer Nature Switzerland AG 2020 F. Tian et al. (Eds.): CASA 2020, CCIS 1300, pp. 112–119, 2020. https://doi.org/10.1007/978-3-030-63426-1_12
MVFNN
113
Pictures are always important parts of the news. Studies have shown that the spread range of the news containing pictures is wider than the one without pictures by 11 times [6]. Fake news always use provocative pictures to attract and mislead readers as well. Therefore, an effective way to identify fake news pictures would help to detect the fake news. Actually, one important potential remedy for fake news recognition is to make use of the visual modal content of the news. Jin et al. found that the fake news pictures were statistically different from those of real news [6]. For example, the number of pictures illustrated in the news, the proportion of news containing hot pictures, and the proportion of special pictures (such as long pictures, chat screenshots, etc
Data Loading...
