Unified-Cyber-Threat-Detection: v1.0.0: Official Implementation - A Unified Framework for Heterogeneous Cyber Threat Detection

Summary This is the official v1.0.0 release of the source code associated with the research paper: "A Unified Framework for Heterogeneous Cyber Threat Detection: Integrating Text and Tabular Data via Semantic Serialization". This release contains all necessary scripts and configurations to reproduce the experimental results reported in the manuscript (Tables 1-6 and Figures 1-3). Key Components Included Preprocessing: Scripts to generate stratified splits for datasets T1-T4. Serialization: Implementation of the semantic serialization method for converting tabular data to text. Training Core: RoBERTa model fine-tuning loop using Low-Rank Adaptation (LoRA). Evaluation: Metrics calculation (AUPRC, F1, etc.) for all tasks. Analysis: SHAP interpretability scripts. Environment Includes environment_linux.yaml for strict reproduction (Linux) and environment.yml for cross-platform compatibility (Windows/macOS/Linux). Authors Yen-Chin Huang Chih-Hung Wang Chun-I Fan