DMIS Laboratory - Korea University

Official implementation of "ASH: A Systematic Automated Evaluation Framework for Creative Text Generation in Cuisine Transfer Tasks" (Submitted to IEEE Access). This repo includes the ASH benchmark data, recipe generation codes, and prompt engineering experiments for evaluating LLM-as-a-judge reliability.

Official Code for "Do All Visual Tokens Matter Equally? Object-Evidence Preserving Token Merging for Vision-Language Retrieval"

[ACL 2026 Findings] ToxReason: A Benchmark for Mechanistic Chemical Toxicity Reasoning via Adverse Outcome Pathway

[NAACL 2025] Rationale-Guided Retrieval Augmented Generation for Medical Question Answering