JLee0 commited on
Commit
5aafb9e
·
verified ·
1 Parent(s): a08d02b

Restyle benchmark overview, human annotation summary, and suite comparison

Browse files
Files changed (1) hide show
  1. README.md +37 -31
README.md CHANGED
@@ -23,39 +23,49 @@ configs:
23
  path: examples/test.parquet
24
  ---
25
 
26
- # Ko-MultimodalMind2Web
27
 
28
- **A Korean offline web-agent benchmark built from browser demonstrations and
29
- annotations created manually by Korean human annotators.** The release contains
30
- **549 tasks, 12,905 scored action steps, and 144 websites**.
31
- The shared Korean annotation project involved two annotators working four
32
- hours per day, five days per week, for approximately three months: **about
33
- 480 annotator-hours of combined collection and annotation effort**.
34
 
35
- The model receives a Korean task, the current screenshot, and preceding action
36
- history, then predicts the next action and its target or value.
37
 
38
- ## Human annotation and collection effort
39
 
40
- **The Korean data was built manually by two Korean human annotators.** They
41
- designed realistic tasks, operated the websites themselves, and produced browser
42
- demonstrations and annotations from those interactions. This human collection
43
- is the basis of the Korean benchmark; the published task interfaces follow the
44
- established international benchmark protocols described below.
 
 
45
 
46
- | Collection item | Reported schedule |
 
 
 
 
 
 
47
  | --- | --- |
48
- | Korean human annotators | 2 |
49
- | Work per annotator per day | 4 hours |
50
- | Working days per week | 5 |
51
- | Annotation period | Approximately 3 months |
52
- | Estimated effort per annotator | Approximately 240 hours |
53
- | Estimated combined effort | **Approximately 480 annotator-hours** |
54
-
55
- The effort estimate uses **2 × 4 hours × 5 days × approximately 12 weeks**.
56
- Three months is approximated as 12 working weeks for this estimate. This is the
57
- shared collection and annotation effort underpinning the three Korean releases:
58
- grounding, offline action prediction, and online task completion.
 
 
 
 
 
 
 
 
59
 
60
  ## How the benchmark was constructed and verified
61
 
@@ -206,10 +216,6 @@ and `point` or `bbox` where applicable. The supplied scorer uses exact matching
206
  for action values after text normalization; report any additional semantic
207
  judging policy used for free-text entry when comparing experiments.
208
 
209
- ## Related Korean benchmark releases
210
-
211
- - [Ko-Screenspot](https://huggingface.co/datasets/KORMo-VL/Ko-Screenspot): single-image GUI grounding.
212
- - [Ko-Online-Mind2Web](https://huggingface.co/datasets/KORMo-VL/Ko-Online-Mind2Web): live website task completion.
213
 
214
  ## Contact
215
 
 
23
  path: examples/test.parquet
24
  ---
25
 
26
+ # 🇰🇷 Ko-MultimodalMind2Web
27
 
28
+ An offline Korean web-agent benchmark, built manually by native Korean annotators.
 
 
 
 
 
29
 
30
+ 📋 549 tasks · 🖼️ 12,905 scored steps · 🌐 144 websites · ✍️ ~480 annotator-hours · 🧭 Offline evaluation
 
31
 
32
+ ## 📌 Overview
33
 
34
+ Ko-MultimodalMind2Web is an offline web-agent benchmark for Korean websites.
35
+ It contains 549 tasks and 12,905 scored action steps across
36
+ 144 websites, collected from human browser demonstrations. The published
37
+ task interface follows [Multimodal Mind2Web](https://huggingface.co/datasets/osunlp/Multimodal-Mind2Web):
38
+ given a Korean task, the current screenshot, and preceding action history, an
39
+ agent predicts the next operation, target, and action value. Evaluation reports
40
+ element accuracy, step success, and task success.
41
 
42
+ ## ✍️ Built by Human Annotators
43
+
44
+ All tasks and reference demonstrations were created manually by Korean human
45
+ annotators. They designed realistic requests, operated the websites themselves,
46
+ and recorded browser demonstrations of each task.
47
+
48
+ | Item | Details |
49
  | --- | --- |
50
+ | 👤 Annotators | 2 native Korean speakers |
51
+ | ⏱️ Total effort | About 480 annotator-hours |
52
+
53
+ This is the shared collection and annotation effort behind all three Korean
54
+ releases in the suite below. The estimate is based on two annotators working
55
+ four hours per day, five days per week, for approximately three months
56
+ (about twelve weeks).
57
+
58
+ ## 🧩 KORMo-VL Benchmark Suite
59
+
60
+ This dataset is part of the KORMo-VL Korean GUI-agent benchmark suite. Each
61
+ release is built from manually collected Korean data and follows the protocol
62
+ of an established international benchmark.
63
+
64
+ | Dataset | Setting | Follows |
65
+ | --- | --- | --- |
66
+ | [Ko-Screenspot](https://huggingface.co/datasets/KORMo-VL/Ko-Screenspot) | Single-image GUI grounding | [ScreenSpot](https://aclanthology.org/2024.acl-long.505/) |
67
+ | [Ko-MultimodalMind2Web](https://huggingface.co/datasets/KORMo-VL/Ko-MultimodalMind2Web) | Offline next-action prediction | [Multimodal Mind2Web](https://huggingface.co/datasets/osunlp/Multimodal-Mind2Web) |
68
+ | [Ko-Online-Mind2Web](https://huggingface.co/datasets/KORMo-VL/Ko-Online-Mind2Web) | Live task completion | [Online Mind2Web](https://github.com/OSU-NLP-Group/Online-Mind2Web) |
69
 
70
  ## How the benchmark was constructed and verified
71
 
 
216
  for action values after text normalization; report any additional semantic
217
  judging policy used for free-text entry when comparing experiments.
218
 
 
 
 
 
219
 
220
  ## Contact
221