DIGITAL TRACE DETECTION SYSTEM FOR CROSS-DEVICE TEXT FILE

Abstract

With the rapid growth of new digital devices and disparate platforms, computing and cybercrime investigations have become highly complicated. Traditional forensic tools encounter many problems correlating fragmented evidence when they're spread across various forms of data, such as text, image, and audio, and across plenty of devices, cloud storage, and/or local data repositories. This paper promulgates the development and evaluation of a fully automated AI-based system overcoming these difficulties to aid in faster and more accurate digital forensic analysis. The research project proposes the Forensic Digital Analyzer, which comprises modular architecture using state-of-the-art deep learning algorithms. The key algorithms include Sentence Transformers for textual embeddings, convolutional neural networks for fine-grain image analysis, and automatic speech recognition using Whisper models for audio. The interactive web user interface offers ubiquitous visualization of digital file structures, similarity detection, and complex evidence-related graphs. The system also offers scalable pipelines for entity recognition, metadata extraction, and AI-aided reporting to further expedite forensic workflows. The experimental results applied on datasets mirroring real forensic scenarios exhibited a tremendous increase in precision values, along with recall and semantic robustness offered by conventional baseline methods. These systems may be computationally demanding at times, generating false positives for noisy modalities. Nevertheless, it manages to reliably link paraphrased texts, visually modified images, and compressed audio files. Therefore embedding-based automation presents a significant leap in the advancement for digital forensic investigation, showing enormous prospects for deployment in the real world. Enhancements in these capabilities might come with multilingual processing, integration of localized large language models, explainable AI frameworks, and disease-specific model optimization.