Zero-Shot Entity Recognition on Forensic Timeline
Resky Ayu Dewi Talasari, Karina Fitriwulandari Ilham, Hudan Studiawan · 2024
Forensic timeline contains standardized entities such as date, time, and host which are essential in a forensic investigation setting. These components need to be analyzed to assist an investigator in analyzing forensic evidence and artifacts. However, traditional entity recognition models often require extensive labeled data for each entity of interest. This becomes challenging in forensic scenarios where new and unseen entities constantly emerge, and labeled data for those entities is non-existent. This paper introduces a method for entity recognition in forensic timeline using zero-shot learning (ZSL) technique by employing the widely used large language models (LLMs), such as ChatGPT and Claude. In this paper, three publicly available different types of datasets downloaded from Digital Corpora namely, 2010-nps-email, nps-2009-casper-rw, and nps-2009-canon-rw, are used to test the proposed approach. Experimental results show that Claude's ZSL model is more consistent than ChatGPT in recognizing entities based on finetuned prompts.