Intelligent Information Service Research Lab (IISR)
Bridging LLMs, Data-centric AI, and Humanities for Practical Solutions
We develop practical and verifiable AI systems, combining Retrieval (RAG) and Cross-domain Representation Learning to solve challenges in biomedical texts, low-resource languages, and historical archives.
|
|
🇹🇼
Localized LLMs
Efficient alignment for new languages, cultural benchmarks, and low-resource adaptation.
📄 ACL / EMNLP
|
🧬
Biomedical NLP
Data-centric, privacy-aware clinical text mining and evidence-grounded QA.
🏆 BioASQ Champion (6 yrs)
|
🗺️
Digital Humanities
Historical GIS, context-aware representation, and label-free map understanding.
📄 IJGIS / EMNLP
|
| ACL 2024 | Chat Vector: A parameter-efficient alignment strategy to equip LLMs with instruction-following capabilities in new languages. |
| EMNLP 2024 | TWBias: The first Taiwan cultural lens benchmark for detecting social bias in Traditional Chinese LLMs. |
| IJGIS 2025 | Label-free Map Understanding: Unsupervised domain adaptation for cross-style/cross-year land-use understanding. |
| npj DigMed | Clinical De-ID: Discovery of inverse scaling in de-identification beyond 6B parameters and effective PEFT strategies (2025). |
| CHAMPION | BioASQ (2020–2025): Six-year consecutive champion in biomedical semantic QA (Retrieval + Evidence-grounding). |
We collaborate with top-tier industry leaders and public sectors to deploy AI in real-world scenarios.
🏭 Industry
|
🏛️ Public Sector
|
🏥 Healthcare
|