Abstract

In today's turbulent international situation, regional conflicts often occur, and the game between major countries has been continuing, which is not only reflected in the military strength, but also the war by means of information technology, especially the turmoil caused by the spread fake news. It is common to forge the statements of influential leaders or entrepreneurs called key populations forgery and spread them on the Internet, which causes bad influence. The number of key populations forgery belongs to a small group in the huge deep forgery number, and the conventional deep forgery detection methods have a great gap in the detection efficiency and performance of small group samples. Therefore, we use the forged information of a large number group to judge the forged information of a small number group, and propose a deep forgery detection method of information construction and comparison under information measurement. Firstly, we propose a sample coverage analysis method for the sample library boundary to explore the difference in the feature space between the small number of key population samples and the large number samples, so as to determine the sample boundary of small number groups from the information level. Secondly, we construct information of key groups and expand the sample scale to meet the data needs of improving the detection accuracy. Finally, we compare the forged features of the large number group with the feature of the expanded small number group, and measure the information coincidence of the two types features by the complexity of feature clusters. When the coincidence degree is high, it is considered that the small number group samples have gone through the forgery process and carried forged information. We have analyzed and tested the current mainstream deep forgery databases and collected a small number of population sample, and compare with some popular methods. The experimental results show that our proposed method achieves the state-of-the-art (SOTA).

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call