IISR-Lab.github.io/research.html at main · IISR-Lab/IISR-Lab.github.io · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
<!DOCTYPE html>
<html lang="en">
<head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">

    <!-- SEO Meta Tags -->
    <title>IISR Lab - Richard Tzong-Han Tsai</title>
    <meta name="description" content="Research Lab at National Central University, Taiwan.">
    <meta name="keywords" content="AI, NLP, Machine Learning">
    <meta name="author" content="Richard Tzong-Han Tsai">

    <!-- Open Graph -->
    <meta property="og:title" content="IISR Lab - Richard Tzong-Han Tsai">
    <meta property="og:description" content="Research Lab at National Central University, Taiwan.">
    <meta property="og:type" content="website">


<style>
    body { font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Helvetica, Arial, sans-serif; margin: 0; padding: 0; background-color: #f9f9f9; }
    .markdown-body { max-width: 100% !important; padding: 0 !important; }
</style>

</head>
<body>
    <!-- 全域樣式與重置 -->
<style>
/* 強制重置 HackMD 預設樣式，確保全寬 */
.container-fluid, #doc, .markdown-body {
    max-width: 100% !important;
    padding: 0 !important;
    margin: 0 !important;
}
.custom-body {
    font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Helvetica, Arial, sans-serif;
    line-height: 1.6;
    color: #333;
    background-color: #f9f9f9;
    width: 100%;
    margin: 0;
    padding: 0;
}

/* 導覽列 */
.nav-bar {
    background: #fff;
    text-align: center;
    padding: 15px 20px;
    box-shadow: 0 2px 4px rgba(0,0,0,0.05);
    font-size: 16px;
    white-space: nowrap;
    position: sticky;
    top: 0;
    z-index: 1000;
    overflow-x: auto;
}
.nav-bar a {
    color: #0366d6;
    text-decoration: none;
    margin: 0 8px;
    font-weight: 500;
}
.nav-bar a:hover {
    background-color: #f0f7ff;
    border-radius: 4px;
}
.nav-sep { color: #ccc; font-size: 14px; }

/* 內容容器 */
.content-wrapper {
    max-width: 1000px;
    margin: 0 auto;
    padding: 40px 20px;
}

/* 標題樣式 */
.section-title {
    text-align: center;
    margin: 60px 0 30px 0;
    font-size: 1.8em;
    color: #2c3e50;
    font-weight: bold;
    position: relative;
}
.section-title::after {
    content: "";
    display: block;
    width: 60px;
    height: 3px;
    background: #c0392b;
    margin: 15px auto 0;
}

/* Thrust 標題樣式 */
.thrust-title {
    font-size: 1.5em;
    color: #2c3e50;
    margin-top: 60px;
    margin-bottom: 15px;
    border-bottom: 2px solid #eee;
    padding-bottom: 10px;
    font-weight: bold;
}

/* 標籤樣式 */
.tag-container {
    margin-bottom: 15px;
}
.tag {
    padding: 4px 10px;
    border-radius: 4px;
    font-weight: bold;
    font-size: 0.85em;
    margin-right: 5px;
    display: inline-block;
    margin-bottom: 5px;
}
.tag-blue { background: #e8f4fd; color: #0366d6; }
.tag-green { background: #eafaf1; color: #27ae60; }
.tag-red { background: #fff5f5; color: #c0392b; }
.tag-gray { background: #f2f2f2; color: #555; }

/* Impact Box 樣式 */
.impact-box {
    padding: 15px;
    border-radius: 4px;
    margin-top: 20px;
    font-size: 0.95em;
}
.impact-blue { background: #f0f7ff; border-left: 4px solid #0366d6; }
.impact-green { background: #f0fdf4; border-left: 4px solid #27ae60; }
.impact-red { background: #fff5f5; border-left: 4px solid #c0392b; }

/* 快速導覽按鈕 */
.quick-nav a {
    background: #fff;
    border: 1px solid #ddd;
    padding: 8px 15px;
    border-radius: 20px;
    color: #333;
    text-decoration: none;
    font-size: 0.9em;
    transition: all 0.2s;
    box-shadow: 0 2px 4px rgba(0,0,0,0.05);
}
.quick-nav a:hover {
    background: #f0f7ff;
    border-color: #0366d6;
    color: #0366d6;
    transform: translateY(-2px);
}

/* 列表樣式優化 */
ul { padding-left: 20px; margin-top: 10px; }
li { margin-bottom: 10px; }
</style>

<div class="custom-body">

<!-- 導覽列 -->
<div class="nav-bar">
<a href="index.html">Home</a> <span class="nav-sep">|</span>
<a href="pi.html">PI</a> <span class="nav-sep">|</span>
<a href="people.html">People</a> <span class="nav-sep">|</span>
<a href="research.html">Research</a> <span class="nav-sep">|</span>
<a href="impact.html">Impact</a> <span class="nav-sep">|</span>
<a href="collaboration.html">Collab</a> <span class="nav-sep">|</span>
<a href="publications.html">Pubs</a> <span class="nav-sep">|</span>
<a href="news.html">News</a> <span class="nav-sep">|</span>
<a href="opportunities.html">Opportunities</a>
</div>

<div class="content-wrapper">

<h1 style="text-align:center; color:#2c3e50; margin-bottom:10px;">Research Areas</h1>
<p style="text-align:center; color:#666; max-width:800px; margin:0 auto 30px auto;">
IISR advances intelligent information services through three major thrusts (2021–2025), integrating <b>LLMs</b>, <b>Retrieval (RAG)</b>, <b>Data-centric AI</b>, and <b>Cross-domain Representation Learning</b>.
</p>

<!-- 快速導覽按鈕 -->
<div class="quick-nav" style="display:flex; gap:10px; margin-bottom:40px; flex-wrap:wrap; justify-content:center;">
<a href="#thrust-1">🇹🇼 Localized LLMs</a>
<a href="#thrust-2">🧬 Biomedical NLP</a>
<a href="#thrust-3">🗺️ Digital Humanities</a>
</div>

<hr style="border:0; border-top:1px solid #eee; margin:40px 0;">

<!-- Thrust 1 -->
<a id="thrust-1"></a>
<h2 class="thrust-title">Thrust 1 — Taiwan-Localized LLMs & Low-Resource Languages</h2>

<div class="tag-container">
<span class="tag tag-blue">ACL 2024</span>
<span class="tag tag-blue">EMNLP 2024</span>
<span class="tag tag-blue">NAACL 2025</span>
<span class="tag tag-blue">COLING 2024</span>
</div>

<p><strong>Motivation:</strong> Mainstream LLMs often lack <strong>Traditional Chinese (Taiwan) cultural grounding</strong> and have limited coverage for local languages (e.g., Taiwanese Hokkien).</p>

<h3>🔑 Key Contributions</h3>
<ul>
<li>
<strong>Chat Vector (ACL 2024)</strong> <a href="https://aclanthology.org/2024.acl-long.590/" target="_blank">[📄 Paper]</a> <a href="https://github.com/aqweteddy/ChatVector" target="_blank">[💻 Code]</a><br>
A lightweight, economical alignment technique that transfers instruction-following ability using weight-space additivity—reducing dependence on expensive RLHF.
</li>
<li>
<strong>TWBias (EMNLP 2024)</strong> <a href="https://aclanthology.org/2024.findings-emnlp.507/" target="_blank">[📄 Paper]</a> <a href="https://github.com/hsinmosyi/TWBias" target="_blank">[💻 Code]</a><br>
The first benchmark for assessing social bias in Traditional Chinese LLMs <strong>through a Taiwan cultural lens</strong>, supporting trustworthy evaluation for Taiwan-localized models.
</li>
<li>
<strong>Hokkien NLP & Multimodal Learning</strong>
<ul>
<li>Standardizing and bridging multiple writing systems (<em>COLING 2024</em>). <a href="https://aclanthology.org/2024.lrec-main.538/" target="_blank">[📄 Paper]</a></li>
<li><strong>ATAIGI (NAACL 2025):</strong> A multimodal learning app leveraging generative models for low-resource Hokkien education. <a href="https://aclanthology.org/2025.naacl-demo.2/" target="_blank">[📄 Paper]</a> <a href="https://github.com/LeeLanguageLab/HokkienTranslation" target="_blank">[💻 Code]</a></li>
</ul>
</li>
</ul>

<!-- Impact Box -->
<div class="impact-box impact-blue">
<strong style="color:#0366d6;">Why it matters:</strong><br>
We build <b>methods + benchmarks + applications</b> as a complete stack for localization—going beyond simple "translation-only" pipelines to ensure cultural depth and practical usability.
</div>

<!-- Thrust 2 -->
<a id="thrust-2"></a>
<h2 class="thrust-title">Thrust 2 — Biomedical NLP (BioNLP): Data-Centric & Privacy-Aware</h2>

<div class="tag-container">
<span class="tag tag-green">🏆 BioASQ Champion (6 yrs)</span>
<span class="tag tag-gray">Briefings in Bioinformatics</span>
<span class="tag tag-gray">npj Digital Medicine</span>
<span class="tag tag-gray">Database (Oxford)</span>
</div>

<p><strong>Motivation:</strong> Biomedical data are often <strong>high-noise / low-resource</strong>, and clinical text requires strict privacy protection.</p>

<h3>🔑 Key Contributions</h3>
<ul>
<li>
<strong>BioASQ Biomedical QA (2020–2025)</strong> <a href="http://bioasq.org" target="_blank">[🌐 Project]</a><br>
Achieved <strong>six consecutive years as the top team</strong> in tasks involving evidence retrieval, semantic search, and evidence-grounded answering.
</li>
<li>
<strong>Relation Extraction: Survey & New Resource (PEDD)</strong> <a href="https://academic.oup.com/bib/article/25/3/bbae132/7644532" target="_blank">[📄 Paper]</a> <a href="https://drive.google.com/drive/folders/1BeFkvjdDMPAvY0zdBd59JECZBC-kqpb7" target="_blank">[💾 Data]</a><br>
Proposed <strong>PEDD</strong>, a high-quality document-level dataset originally developed for the <strong>AI CUP 2019</strong> competition, addressing critical inconsistencies in legacy benchmarks (<em>Briefings in Bioinformatics, 2024</em>).
</li>
<li>
<strong>Clinical De-identification & Temporal Normalization with LLMs</strong> <a href="https://www.nature.com/articles/s41746-025-01921-7" target="_blank">[📄 Paper]</a>
<ul>
<li>Discovery of <strong>inverse scaling</strong> beyond ~6B parameters without targeted adaptation (<em>npj Digital Medicine, 2025</em>).</li>
<li>Efficient solution via <strong>PEFT/LoRA</strong>, improving privacy protection while preserving clinical utility.</li>
</ul>
</li>
<li>
<strong>Data-centric Ensemble Learning</strong> <a href="https://doi.org/10.1093/database/baae127" target="_blank">[📄 Paper]</a><br>
<strong>Enhancing biomedical relation extraction through data-centric and preprocessing-robust ensemble learning approach</strong> (<em>Database, 2025</em>).
</li>
</ul>

<!-- Impact Box -->
<div class="impact-box impact-green">
<strong style="color:#27ae60;">Why it matters:</strong><br>
Many global BioNLP systems optimize model architectures on curated datasets; we emphasize <b>real-world robustness</b> and <b>privacy constraints</b>, addressing barriers to actual hospital deployment.
</div>

<!-- Thrust 3 -->
<a id="thrust-3"></a>
<h2 class="thrust-title">Thrust 3 — Digital Humanities & Historical GIS</h2>

<div class="tag-container">
<span class="tag tag-red">IJGIS 2025</span>
<span class="tag tag-red">EMNLP 2023</span>
<span class="tag tag-red">DSH 2025</span>
</div>

<p><strong>Motivation:</strong> Humanities data are frequently <strong>unstructured</strong>, <strong>archaic in language</strong>, and <strong>lack annotations</strong>, making standard supervised learning ineffective.</p>

<h3>🔑 Key Contributions</h3>
<ul>
<li>
<strong>Historical Maps Without Labels (IJGIS 2025)</strong> <a href="https://www.tandfonline.com/doi/full/10.1080/13658816.2024.2425668" target="_blank">[📄 Paper]</a> <a href="https://figshare.com/articles/software/MapStyleSeg_full_code/25560276/1?file=45532515" target="_blank">[💻 Code]</a><br>
<strong>Unsupervised domain adaptation (UDA)</strong> bridging modern labeled maps and historical unlabeled maps for land-use understanding.
</li>
<li>
<strong>MingOfficial (EMNLP 2023)</strong> <a href="https://aclanthology.org/2023.emnlp-main.266/" target="_blank">[📄 Paper]</a> <a href="https://data.depositar.io/dataset/ming_official" target="_blank">[💾 Data (Depository)]</a><br>
A historical context-aware representation learning framework embedding <strong>time, space, and events</strong> into career trajectory representations.
</li>
<li>
<strong>Computational Analysis of Ming Military Power</strong> <a href="https://doi.org/10.1093/llc/fqad088" target="_blank">[📄 Paper]</a> <a href="https://www.sinica.edu.tw/cp/697" target="_blank">[📰 Media (中文)]</a><br>
A digital humanities approach to analyzing the roles of Supreme Commanders and Grand Coordinators in <em>Ming Shilu</em> (<em>Digital Scholarship in the Humanities, 2025</em>).
</li>
<li>
<strong>LLM-based Long-document Analysis</strong><br>
Designing grounded RAG workflows with context fields (metadata + text) and human-in-the-loop benchmarks to improve traceability in historical sources.
</li>
</ul>

<!-- Impact Box -->
<div class="impact-box impact-red">
<strong style="color:#c0392b;">Why it matters:</strong><br>
Instead of applying generic tools, we develop <b>algorithmic methods</b> (like UDA and Context Embedding) that specifically overcome the fundamental "no-label" barrier in historical materials.
</div>

<hr style="border:0; border-top:1px solid #eee; margin:60px 0;">

<!-- Deployment Context -->
<h2 class="section-title" style="margin-top:0;">🚀 Selected Deployment Context</h2>

<table style="width:100%; border-collapse:collapse; margin-top:20px; background:#fff; box-shadow:0 2px 8px rgba(0,0,0,0.05); border-radius:8px; overflow:hidden;">
<tr style="border-bottom:1px solid #eee;">
<td style="padding:15px 20px; width:30%; vertical-align:top; background:#f9f9f9; font-weight:bold; color:#2c3e50;">Semiconductor</td>
<td style="padding:15px 20px;">Domain LLMs for fab operation (<b>TSMC</b>, 2025)</td>
</tr>
<tr style="border-bottom:1px solid #eee;">
<td style="padding:15px 20px; vertical-align:top; background:#f9f9f9; font-weight:bold; color:#2c3e50;">National Scale</td>
<td style="padding:15px 20px;">AI semantic search for theses/dissertations (<b>National Central Library</b>, 2025)</td>
</tr>
<tr style="border-bottom:1px solid #eee;">
<td style="padding:15px 20px; vertical-align:top; background:#f9f9f9; font-weight:bold; color:#2c3e50;">Public / Education</td>
<td style="padding:15px 20px;">Teacher co-planning support with <b>NARLabs / NCHC</b> (2025)</td>
</tr>
<tr style="border-bottom:1px solid #eee;">
<td style="padding:15px 20px; vertical-align:top; background:#f9f9f9; font-weight:bold; color:#2c3e50;">Healthcare</td>
<td style="padding:15px 20px;">Clinical collaborations with <b>Cathay General</b> & <b>Landseed Int'l Hospital</b> (2021–2025)</td>
</tr>
<tr>
<td style="padding:15px 20px; vertical-align:top; background:#f9f9f9; font-weight:bold; color:#2c3e50;">International</td>
<td style="padding:15px 20px;">East Asian multilingual model merging with <b>KISTI</b> (Korea, 2025)</td>
</tr>
</table>

<div style="text-align:center;margin-top:60px;">
<a href="index.html" style="background:#2c3e50;color:#fff;padding:12px 30px;border-radius:30px;text-decoration:none;font-weight:bold;box-shadow:0 4px 10px rgba(0,0,0,0.2);">← Back to Home</a>
</div>

</div> <!-- content-wrapper end -->
</div> <!-- custom-body end -->
</body>
</html>