Research Area: Machine Learning
Recent years, with the rapid development of the Internet, a large amount of false information has been widely disseminated on public social media platforms, thus misleading readers and causing serious negative effects. How to accurately identify and detect false information has become a major concern of the public and government. The existing false information detection methods focus on the simple splicing of text and image features, without considering the interaction between different modalities and the rich deep semantic information behind the text. In addition, most false information detection models do not work well on emerging events. Therefore, this paper proposes a false information detection method based on multi-modal event memory network(MEMN). Specifically, text features are extracted through BERT and Text-CNN, image features are extracted using the VGG-19 network, feature interactions of multiple modalities are captured through an attention mechanism, and finally an event memory network is used to mine connections between false information and events. A large number of experiments are carried out on the public microblog dataset. The experimental results show that the accuracy of this method is about 7 percentage points higher than that of other single-modal and multi-modal baseline methods such as EANN, and the generalization ability of the model is improved.
Keywords:
Visualization
Social networking (online)
Splicing
Blogs
Semantics
Government
Feature extraction
Author(s) Name: Jinzhong Xu; Hailong Zhao; Weiguang Liu; Xinyang Ding
Journal name:
Conferrence name: 2023 3rd International Conference on Consumer Electronics and Computer Engineering
Publisher name: IEEE
DOI: 10.1109/ICCECE58074.2023.10135191
Volume Information:
