A BERT-Based Approach to Keyword Search for Digital Evidence Analysis

Existing forensic tools, especially those relying on traditional keyword searches, often fall short in efficiency and accuracy. These tools are vulnerable to human error, bias, and investigator fatigue, which can result in the omission of crucial evidence, thereby undermining the success of investigations. Keyword searching, although a widely used technique across various types of digital investigations, frequently proves inadequate when faced with complexities such as large data volumes, data variety, unstructured data, and context-dependent information of modern data sets. This research presents an innovative AI-powered approach to keyword searching designed to overcome the complexities of traditional methods. By incorporating advanced Bidirectional Encoder Representations from Transformers (BERT), a machine learning (ML) algorithm, and natural language processing (NLP) techniques, the proposed identifies relevant terms and entities within large and complex data sets. This approach accelerates the investigative process and has a final accuracy of approximately 92.5%, reducing the risk of overlooking critical evidence by leveraging techniques like semantic search, contextual understanding, and conceptual matching to identify relevant information even when exact keywords or variations are not used. The adoption of AI-driven keyword searching is expected to transform digital forensic investigations by providing a more reliable, efficient, and comprehensive tool for evidence discovery.