Application of CNN-BiGRU-MHSA: A Self-Attention Mechanism Based Detection Method in Electronic Data Forensics

Abstract

In response to the shortcomings of the current file fragmentation detection technology in terms of accuracy and efficiency, this study proposes a file fragmentation detection method based on the multihead self-attention mechanism. First, the traditional self-attention mechanism is optimized by introducing the concept of multi-dimensional features and multi-head feature detection. Second, a file fragmentation detection model is constructed by combining the optimized feature extraction method, bidirectional gated recurrent units, and convolutional neural networks. This model achieved classification accuracy as high as 99% and 98.9% on the GovDocs1 dataset and the Enron email dataset, respectively, with a mean square error as low as 0.08. In practical applications, the model achieved a high classification accuracy of 99.2%, a low classification time of 0.09 seconds, and a low false detection rate of 0.02%, demonstrating excellent detection performance. The detection algorithms and models designed in this study outperformed existing methods in both performance and practical application effectiveness. This integrated approach not only circumvents the constraints of conventional self-attention mechanisms in one-dimensional feature extraction but also augments the model's capacity to discern and extract multi-dimensional features in an efficacious manner. The results indicate that the model effectively improves the work efficiency in actual electronic data forensics, providing a new reference method for the field.