Short text classification approach to identify child sexual exploitation material.

Area of Science:

• Digital Forensics
• Cybercrime Investigation
• Machine Learning Applications

Background:

• Child Sexual Exploitation Material (CSEM) is a serious crime requiring efficient Law Enforcement Agency (LEA) investigation.
• Manual analysis of seized digital devices for CSEM evidence is time-consuming, especially under Spanish warrant limitations.
• Obfuscated file names and naming patterns present challenges in identifying CSEM-related files.

Purpose of the Study:

• To develop and evaluate a method for accelerating CSEM identification by analyzing file names and paths, bypassing content analysis.
• To address the challenges posed by distorted and sparse text in file names and paths.
• To benchmark machine learning and convolutional neural network models for CSEM file identification.

Main Methods:

• Two approaches were developed: one combining independent file name and path classifiers, and another using a file name classifier on absolute paths.
• Both methods utilize character n-gram analysis.
• Novel binary and orthographic features were introduced to enhance text representation.

Main Results:

• Six classification models, including machine learning and convolutional neural networks, were benchmarked.
• The proposed classifier achieved a high F1 score of 0.988.
• The approach demonstrates significant potential for improving the efficiency of CSEM investigations.

Conclusions:

• Analyzing file names and absolute paths is an effective strategy for rapid CSEM identification.
• The developed classifier offers a promising tool for Law Enforcement Agencies (LEAs) to expedite digital forensic investigations.
• This method can help LEAs combat the production and sharing of CSEM more efficiently.