Finding similarity in digital forensics investigations can be assisted with the use of Approximate Matching (AM) functions. These algorithms create small and compact representations of objects (similar to hashes) which can be compared to identify similarity. However, often results are biased due to common blocks (data structures found in many different files regardless of content). In this paper, we evaluate the precision and recall metrics for AM functions when removing common blocks. In detail, we analyze how the similarity score changes and impacts different investigation scenarios. Results show that many irrelevant matches can be filtered out and that a new interpretation of the score allows a better similarity detection.
Digital Forensic Investigations: Issues of Intangibility, Complications and Inconsistencies in Cyber-crimes
