SIA_Dataset / SIABench-main /Alert_Triaging_Dataset /Alert_Triaging_Dataset.md
SIABench's picture
Upload folder using huggingface_hub
f653531 verified

🎯 Overview

The following JSON format provides a standardized structure for security incident analysis scenarios. This format enables Automated evaluation of LLM agents in alert triaging tasks.

Structure Overview

{
  "scenarios": [
    {
      "metadata": {
        "scenario_name": "monday_pri_9"
      },
      "sia_components": {
        "scenario": "Scenario description with alert included",
        "alert": "Raw security alert from Snort",
        "tools_available": ["tshark", "python", "grep", "..."],
        "files_available": "capture.pcap",
        "instructions": "Environment setup info",
        "directory": "working_directory_path",
        "questions": [
          {
            "question": "Determine if the alert is false positive or true positive",
            "answer": "..."
          }
        ]
      }
    }
  ]
}

Field Descriptions

Metadata

  • scenario_name: Unique identifier for the scenario

SIA Components

  • scenario: Background context for the security incident
  • alert: Raw alert from IDS/IPS system
  • tools_available: List of command-line tools for analysis
  • files_available: Files provided for investigation
  • instructions: Environment and usage guidelines
  • directory: Working directory path
  • questions: Array of questions with expected answers

Usage

For LLM Input: Use all fields except questions[].answer and alert

For Evaluation: Compare LLM responses against questions[].answer

Artifacts: The PCAP files are given in the following CICIDS - link and TII-SRC - link. Each scenario name corresponds to its respective PCAP folder.