Clémentine
commited on
Commit
·
0322f30
1
Parent(s):
000658b
updated graph
Browse files
app/src/content/chapters/human-evaluation/using-human-annotators.mdx
CHANGED
|
@@ -4,6 +4,7 @@ title: "Using human annotators"
|
|
| 4 |
|
| 5 |
import bestAnnotationPractices from '../../assets/image/best_annotation_practices.png';
|
| 6 |
import Image from '../../../components/Image.astro';
|
|
|
|
| 7 |
import Note from "../../../components/Note.astro";
|
| 8 |
import Sidenote from "../../../components/Sidenote.astro";
|
| 9 |
import Accordion from "../../../components/Accordion.astro";
|
|
@@ -12,12 +13,7 @@ import Accordion from "../../../components/Accordion.astro";
|
|
| 12 |
|
| 13 |
I suggest reading Section 3 of this [review](https://aclanthology.org/2024.cl-3.1/) of good practices in data annotation quality. If you want production level quality and have the means to implement all of these methods, go ahead!
|
| 14 |
|
| 15 |
-
<
|
| 16 |
-
|
| 17 |
-
For production-level quality, read Section 3 of this [comprehensive review](https://aclanthology.org/2024.cl-3.1/) of data annotation best practices. The diagram below summarizes key principles.
|
| 18 |
-
</Note>
|
| 19 |
-
|
| 20 |
-
<Image src={bestAnnotationPractices} alt="Best annotation practices diagram" />
|
| 21 |
|
| 22 |
However, important guidelines (no matter your project size) are the following, once you defined your task and scoring guidelines.
|
| 23 |
|
|
|
|
| 4 |
|
| 5 |
import bestAnnotationPractices from '../../assets/image/best_annotation_practices.png';
|
| 6 |
import Image from '../../../components/Image.astro';
|
| 7 |
+
import HtmlEmbed from "../../../components/HtmlEmbed.astro";
|
| 8 |
import Note from "../../../components/Note.astro";
|
| 9 |
import Sidenote from "../../../components/Sidenote.astro";
|
| 10 |
import Accordion from "../../../components/Accordion.astro";
|
|
|
|
| 13 |
|
| 14 |
I suggest reading Section 3 of this [review](https://aclanthology.org/2024.cl-3.1/) of good practices in data annotation quality. If you want production level quality and have the means to implement all of these methods, go ahead!
|
| 15 |
|
| 16 |
+
<HtmlEmbed src="d3-quality-management.html" frameless />
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
However, important guidelines (no matter your project size) are the following, once you defined your task and scoring guidelines.
|
| 19 |
|
app/src/content/embeds/d3-quality-management.html
ADDED
|
@@ -0,0 +1,254 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
<div class="d3-quality-management"></div>
|
| 2 |
+
|
| 3 |
+
<style>
|
| 4 |
+
.d3-quality-management {
|
| 5 |
+
font-family: var(--default-font-family);
|
| 6 |
+
background: transparent;
|
| 7 |
+
border: none;
|
| 8 |
+
border-radius: 0;
|
| 9 |
+
padding: 0;
|
| 10 |
+
width: 100%;
|
| 11 |
+
margin: 0 auto;
|
| 12 |
+
position: relative;
|
| 13 |
+
}
|
| 14 |
+
|
| 15 |
+
.d3-quality-management .quality-grid {
|
| 16 |
+
display: grid;
|
| 17 |
+
grid-template-columns: repeat(3, 1fr);
|
| 18 |
+
gap: var(--spacing-4);
|
| 19 |
+
margin-bottom: var(--spacing-4);
|
| 20 |
+
}
|
| 21 |
+
|
| 22 |
+
.d3-quality-management .quality-card {
|
| 23 |
+
background: oklch(from var(--primary-color) calc(l + 0.4) c h / 0.35);
|
| 24 |
+
border: 1px solid oklch(from var(--primary-color) calc(l + 0.15) c h / 0.6);
|
| 25 |
+
border-radius: 16px;
|
| 26 |
+
padding: var(--spacing-4) var(--spacing-5);
|
| 27 |
+
text-align: left;
|
| 28 |
+
display: flex;
|
| 29 |
+
flex-direction: column;
|
| 30 |
+
justify-content: flex-start;
|
| 31 |
+
align-items: flex-start;
|
| 32 |
+
min-height: 160px;
|
| 33 |
+
transition: all 0.3s ease;
|
| 34 |
+
cursor: pointer;
|
| 35 |
+
}
|
| 36 |
+
|
| 37 |
+
.d3-quality-management .quality-card:hover {
|
| 38 |
+
transform: translateY(-2px);
|
| 39 |
+
box-shadow: 0 8px 24px rgba(0, 0, 0, 0.12);
|
| 40 |
+
border-color: var(--primary-color);
|
| 41 |
+
}
|
| 42 |
+
|
| 43 |
+
/* Dark mode adjustments for better readability */
|
| 44 |
+
[data-theme="dark"] .d3-quality-management .quality-card {
|
| 45 |
+
background: oklch(from var(--primary-color) calc(l + 0.3) c h / 0.25);
|
| 46 |
+
border-color: oklch(from var(--primary-color) calc(l + 0.1) c h / 0.65);
|
| 47 |
+
}
|
| 48 |
+
|
| 49 |
+
[data-theme="dark"] .d3-quality-management .quality-card:hover {
|
| 50 |
+
box-shadow: 0 8px 24px rgba(0, 0, 0, 0.3);
|
| 51 |
+
}
|
| 52 |
+
|
| 53 |
+
.d3-quality-management .category-title {
|
| 54 |
+
font-size: 18px;
|
| 55 |
+
font-weight: 600;
|
| 56 |
+
color: var(--primary-color);
|
| 57 |
+
margin-bottom: var(--spacing-3);
|
| 58 |
+
line-height: 1.2;
|
| 59 |
+
letter-spacing: -0.01em;
|
| 60 |
+
position: relative;
|
| 61 |
+
z-index: 1;
|
| 62 |
+
}
|
| 63 |
+
|
| 64 |
+
/* Dark mode adjustments for better readability */
|
| 65 |
+
[data-theme="dark"] .d3-quality-management .category-title {
|
| 66 |
+
color: oklch(from var(--primary-color) calc(l + 0.1) calc(c * 1.2) h);
|
| 67 |
+
}
|
| 68 |
+
|
| 69 |
+
.d3-quality-management .category-items {
|
| 70 |
+
list-style: none;
|
| 71 |
+
padding: 0;
|
| 72 |
+
margin: 0;
|
| 73 |
+
font-size: 13px;
|
| 74 |
+
color: var(--text-color);
|
| 75 |
+
line-height: 1.6;
|
| 76 |
+
font-weight: 500;
|
| 77 |
+
position: relative;
|
| 78 |
+
z-index: 1;
|
| 79 |
+
}
|
| 80 |
+
|
| 81 |
+
.d3-quality-management .category-items li {
|
| 82 |
+
padding-left: 1.6em;
|
| 83 |
+
position: relative;
|
| 84 |
+
margin-bottom: var(--spacing-1);
|
| 85 |
+
}
|
| 86 |
+
|
| 87 |
+
.d3-quality-management .category-items li::before {
|
| 88 |
+
content: "•";
|
| 89 |
+
position: absolute;
|
| 90 |
+
left: 0.4em;
|
| 91 |
+
top: 0;
|
| 92 |
+
color: var(--primary-color);
|
| 93 |
+
font-weight: 600;
|
| 94 |
+
}
|
| 95 |
+
|
| 96 |
+
/* Dark mode adjustments for better readability */
|
| 97 |
+
[data-theme="dark"] .d3-quality-management .category-items {
|
| 98 |
+
color: var(--text-color);
|
| 99 |
+
}
|
| 100 |
+
|
| 101 |
+
.d3-quality-management .bottom-row {
|
| 102 |
+
display: grid;
|
| 103 |
+
grid-template-columns: repeat(2, 1fr);
|
| 104 |
+
gap: var(--spacing-4);
|
| 105 |
+
}
|
| 106 |
+
|
| 107 |
+
.d3-quality-management .caption {
|
| 108 |
+
margin-top: var(--spacing-4);
|
| 109 |
+
font-size: 13px;
|
| 110 |
+
color: var(--muted-color);
|
| 111 |
+
line-height: 1.5;
|
| 112 |
+
font-style: italic;
|
| 113 |
+
}
|
| 114 |
+
|
| 115 |
+
@media (max-width: 1024px) {
|
| 116 |
+
.d3-quality-management .quality-grid {
|
| 117 |
+
grid-template-columns: repeat(2, 1fr);
|
| 118 |
+
}
|
| 119 |
+
|
| 120 |
+
.d3-quality-management .bottom-row {
|
| 121 |
+
grid-template-columns: 1fr;
|
| 122 |
+
}
|
| 123 |
+
}
|
| 124 |
+
|
| 125 |
+
@media (max-width: 768px) {
|
| 126 |
+
.d3-quality-management {
|
| 127 |
+
padding: var(--spacing-3) 0;
|
| 128 |
+
}
|
| 129 |
+
|
| 130 |
+
.d3-quality-management .quality-grid {
|
| 131 |
+
grid-template-columns: 1fr;
|
| 132 |
+
gap: var(--spacing-3);
|
| 133 |
+
}
|
| 134 |
+
|
| 135 |
+
.d3-quality-management .bottom-row {
|
| 136 |
+
grid-template-columns: 1fr;
|
| 137 |
+
gap: var(--spacing-3);
|
| 138 |
+
}
|
| 139 |
+
|
| 140 |
+
.d3-quality-management .quality-card {
|
| 141 |
+
min-height: 140px;
|
| 142 |
+
padding: var(--spacing-3) var(--spacing-4);
|
| 143 |
+
}
|
| 144 |
+
}
|
| 145 |
+
</style>
|
| 146 |
+
|
| 147 |
+
<script>
|
| 148 |
+
(() => {
|
| 149 |
+
const bootstrap = () => {
|
| 150 |
+
const scriptEl = document.currentScript;
|
| 151 |
+
let container = scriptEl ? scriptEl.previousElementSibling : null;
|
| 152 |
+
if (!(container && container.classList && container.classList.contains('d3-quality-management'))) {
|
| 153 |
+
const candidates = Array.from(document.querySelectorAll('.d3-quality-management'))
|
| 154 |
+
.filter((el) => !(el.dataset && el.dataset.mounted === 'true'));
|
| 155 |
+
container = candidates[candidates.length - 1] || null;
|
| 156 |
+
}
|
| 157 |
+
|
| 158 |
+
if (!container) return;
|
| 159 |
+
|
| 160 |
+
if (container.dataset) {
|
| 161 |
+
if (container.dataset.mounted === 'true') return;
|
| 162 |
+
container.dataset.mounted = 'true';
|
| 163 |
+
}
|
| 164 |
+
|
| 165 |
+
// Data structure for quality management categories
|
| 166 |
+
const categories = [
|
| 167 |
+
{
|
| 168 |
+
title: 'Annotation Process',
|
| 169 |
+
items: [
|
| 170 |
+
'Iterative Annotation',
|
| 171 |
+
'Careful Data Selection',
|
| 172 |
+
'Annotation Scheme',
|
| 173 |
+
'Guideline Design',
|
| 174 |
+
'Pilot Study',
|
| 175 |
+
'Validation Step'
|
| 176 |
+
]
|
| 177 |
+
},
|
| 178 |
+
{
|
| 179 |
+
title: 'Annotators',
|
| 180 |
+
items: [
|
| 181 |
+
'Workforce Selection',
|
| 182 |
+
'Qualification Test',
|
| 183 |
+
'Annotator Training',
|
| 184 |
+
'Annotator Debriefing',
|
| 185 |
+
'Monetary Incentive'
|
| 186 |
+
]
|
| 187 |
+
},
|
| 188 |
+
{
|
| 189 |
+
title: 'Quality Estimation',
|
| 190 |
+
items: [
|
| 191 |
+
'Error Rate',
|
| 192 |
+
'Control Questions',
|
| 193 |
+
'Agreement'
|
| 194 |
+
]
|
| 195 |
+
},
|
| 196 |
+
{
|
| 197 |
+
title: 'Quality Improvement',
|
| 198 |
+
items: [
|
| 199 |
+
'Correction',
|
| 200 |
+
'Updated Guidelines',
|
| 201 |
+
'Filtering',
|
| 202 |
+
'Annotator Feedback',
|
| 203 |
+
'Annotator Deboarding'
|
| 204 |
+
]
|
| 205 |
+
},
|
| 206 |
+
{
|
| 207 |
+
title: 'Adjudication',
|
| 208 |
+
items: [
|
| 209 |
+
'Manual Curation',
|
| 210 |
+
'Majority Voting',
|
| 211 |
+
'Probabilistic Aggregation'
|
| 212 |
+
]
|
| 213 |
+
}
|
| 214 |
+
];
|
| 215 |
+
|
| 216 |
+
// Create the quality management structure
|
| 217 |
+
const topRowHtml = categories.slice(0, 3).map(cat => `
|
| 218 |
+
<div class="quality-card">
|
| 219 |
+
<div class="category-title">${cat.title}</div>
|
| 220 |
+
<ul class="category-items">
|
| 221 |
+
${cat.items.map(item => `<li>${item}</li>`).join('')}
|
| 222 |
+
</ul>
|
| 223 |
+
</div>
|
| 224 |
+
`).join('');
|
| 225 |
+
|
| 226 |
+
const bottomRowHtml = categories.slice(3).map(cat => `
|
| 227 |
+
<div class="quality-card">
|
| 228 |
+
<div class="category-title">${cat.title}</div>
|
| 229 |
+
<ul class="category-items">
|
| 230 |
+
${cat.items.map(item => `<li>${item}</li>`).join('')}
|
| 231 |
+
</ul>
|
| 232 |
+
</div>
|
| 233 |
+
`).join('');
|
| 234 |
+
|
| 235 |
+
container.innerHTML = `
|
| 236 |
+
<div class="quality-grid">
|
| 237 |
+
${topRowHtml}
|
| 238 |
+
</div>
|
| 239 |
+
<div class="bottom-row">
|
| 240 |
+
${bottomRowHtml}
|
| 241 |
+
</div>
|
| 242 |
+
<div class="caption">
|
| 243 |
+
<strong>Figure 1:</strong> Data annotation best practices recommended by the comprehensive review at aclanthology.org/2024.cl-3.1 - check out their Section 3.
|
| 244 |
+
</div>
|
| 245 |
+
`;
|
| 246 |
+
};
|
| 247 |
+
|
| 248 |
+
if (document.readyState === 'loading') {
|
| 249 |
+
document.addEventListener('DOMContentLoaded', bootstrap, { once: true });
|
| 250 |
+
} else {
|
| 251 |
+
bootstrap();
|
| 252 |
+
}
|
| 253 |
+
})();
|
| 254 |
+
</script>
|