Effective fake news video detection using domain knowledge and multimodal data fusion on youtube

In the digital age, numerous videos are being actively produced and uploaded online. Simultaneously, fake news videos to attract public attention are also on the rise. Therefore, intensive research is being conducted to detect them. Validating video content is critical for all users as the public is exposed to various fake news videos. This study proposes ways to detect fake news videos effectively using domain knowledge and multimodal data fusion. We use domain knowledge to perform learning by reflecting the potential meaning of comments, helping us detecting fake news videos. We also use the linear combination to efficiently adjust the encoding rate for each characteristic of the video and effectively detect fake news videos. In particular, the domain knowledge improves the model performance by approximately 3% for all test datasets. Consequently, we achieve an F1-score of 0.93, which is higher than those of other comparison models in all the test datasets.