nlp-pipeline

What problem does it solve?

This NLP pipeline analyzes Korean traffic accident descriptions to convert unstructured text into structured fault-ratio data, enabling automated analysis and standardized reporting.

Core Features & Use Cases

• Text preprocessing rules tailored to Korean accident descriptions, including Unicode normalization, whitespace handling, and Korean-specific normalization
• Accurate accident type classification (차대차, 차대보행자, 단독사고, 차대건축물, 다중충돌, 주차차량) with sub-types and confidence scoring
• Named Entity Recognition (NER) to extract location, time, and vehicle/drivers entities, while preserving key terminology
• Violation and mitigating-factor extraction to support liability assessment with evidence flags
• Output in a structured JSON schema suitable for APIs, legal workflows, and dispute resolution

Quick Start

Provide a structured fault-ratio analysis for a given Korean traffic accident description.