The Effect of the Visualization Type of Data Received from the X Platform on Manipulation Detection with Deep Learning

Deep learning can provide successful results in detecting manipulations made with bot accounts on the X (Twitter) platform and preventing disinformation. As a method for applying deep learning algorithms, numerical values in the data set can be visualized and classified by making sense of them. The methodology used in the data visualization process and the structure of the shape created are critical for deep learning and classification performance. In this article, the effect of the format used to visualize the data in the dataset to be applied deep learning on the classification performance has been demonstrated with more than 400 experimental studies. From visualization methods; The effect of deep learning applied to images created with formats such as area graph, qrcode, spectrogram graph, color map, histogram and distribution on the classification performance with Convolutional Neural Network algorithms has been demonstrated comparatively. As a result of experimental studies, it was observed that while the highest performance was achieved with 98.67% depending on the visual and algorithm used, the performance decreased to 23.70% when different visuals and algorithms were used. In this regard, it has been determined that the methodology applied in revealing the profile confidence index, the effect of the variables that make up the data set and the algorithms used are very important, but the data visualization method used in creating the figure is important. has a critical impact on performance.