From Online Behaviours to Images: A Novel Approach to Social Bot Detection

Multimodal detection of epilepsy with deep neural networks

“…In [15], the authors proposed a new method for recognizing bots in Twitter based on image recognition. Specifically, they introduced a new approach for converting the digital DNA sequence into an image consisting of three channels.…”

“…To address the issue of the imbalanced dataset, we create a dataset consisting of 943 real users and 943 social spambots. This technique, which addresses the issue of imbalanced dataset by downsampling the set of real users, has been adopted by previous studies [15], [34].…”

“…After having created a digital DNA sequence per user, we adopt the method introduced in [15] for transforming the DNA sequence into an image consisting of three channels. First, we check whether the length of the longest DNA sequence is a perfect square.…”

“…The resulting image is a grayscale image in a (1, H, W) format. Similar to [15], we convert the images into a (3, H, W) format utilizing the grayscale transformation 1 .…”

“…First, motivated by [13], [14], we design the digital DNA per user, which indicates the sequence of actions that an account performs. Next, we adopt the methodology proposed by [15] for converting the DNA sequence into an image. After that, we fine-tune several pretrained models, including AlexNet, ResNet, VGG16, etc., and compare their performances.…”