cklxx commited on
Commit
80dc71c
·
verified ·
1 Parent(s): f9ef837

Initial release

Browse files
Files changed (4) hide show
  1. .gitattributes +2 -0
  2. README.md +72 -0
  3. data/test.jsonl +3 -0
  4. data/train.jsonl +3 -0
.gitattributes CHANGED
@@ -58,3 +58,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
 
 
 
58
  # Video files - compressed
59
  *.mp4 filter=lfs diff=lfs merge=lfs -text
60
  *.webm filter=lfs diff=lfs merge=lfs -text
61
+ data/test.jsonl filter=lfs diff=lfs merge=lfs -text
62
+ data/train.jsonl filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,72 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-4.0
3
+ language:
4
+ - en
5
+ task_categories:
6
+ - text-classification
7
+ tags:
8
+ - web-agent
9
+ - browser-agent
10
+ - web-navigation
11
+ - element-grounding
12
+ - mind2web
13
+ pretty_name: WebChain Elements
14
+ size_categories:
15
+ - 10K<n<100K
16
+ configs:
17
+ - config_name: default
18
+ data_files:
19
+ - split: train
20
+ path: data/train.jsonl
21
+ - split: test
22
+ path: data/test.jsonl
23
+ ---
24
+
25
+ # WebChain Elements — human browser trajectories as element-level decision steps
26
+
27
+ **10,930 steps** (2,555 tasks, 243 real websites) from [WebChain](https://huggingface.co/datasets/webagentlab/webchain)
28
+ (webagentlab, CC-BY-4.0) turned into one-step decision examples in the text format browser-agent decision heads use:
29
+ the task goal, the page (URL, title, visible text), the list of interactive elements, the actions taken so far, and the
30
+ **gold operation + gold element** the human chose.
31
+
32
+ | split | steps | tasks | websites | CLICK / TYPE_TEXT / PRESS_ENTER |
33
+ |---|---|---|---|---|
34
+ | train | 8,435 | 1,935 | 199 | 7,273 / 1,053 / 109 |
35
+ | test | 2,495 | 620 | 44 | 2,109 / 364 / 22 |
36
+
37
+ The test split holds out whole **websites** (none of its 44 sites appears in train). Tasks are mostly multi-constraint
38
+ (filters, sorts, dates) on shopping, travel, real-estate and job sites.
39
+
40
+ ## Fields
41
+
42
+ ```
43
+ goal the task, e.g. "Look for the lowest-priced women's double-breasted trench coats."
44
+ website the site's primary host; domain / intent: WebChain's labels
45
+ url, title the page the step was taken on; page_text: its visible text (<= 6,000 chars)
46
+ candidates interactive elements (≤ 60, page order): {id, kind: click|fill, label, role, value?, checked?, ...}
47
+ history earlier steps of the same trajectory: {action, kind, text, page_changed}
48
+ gold_op CLICK | TYPE_TEXT | PRESS_ENTER
49
+ gold_id the chosen candidate's id (click:N / fill:N; "press_enter" for PRESS_ENTER)
50
+ gold_label its label; gold_text: the typed text for TYPE_TEXT
51
+ ```
52
+
53
+ ## How it was built
54
+
55
+ WebChain records, per step, a CSS selector plus links to the page's accessibility tree (AX) and DOM snapshot. For 3,000
56
+ randomly chosen trajectories (seed 0), every step's AX tree was fetched and flattened; the **gold element was recovered
57
+ exactly** by resolving the step's CSS selector in the DOM snapshot and following its `data-imean-axt-id` to the AX node
58
+ (text matching only as a fallback — it disagreed with the selector on ~13 % of a sample, so it is never the default).
59
+ Candidates are the page's interactive AX nodes plus some non-semantic text nodes (the clicked target is often a
60
+ `<span>`/`<div>`), capped at 60 in page order. Steps whose gold element has **no label** or whose label is **duplicated**
61
+ among the candidates were dropped (they cannot be learned from text), as were hover / drag / copy / paste steps.
62
+ WebChain has no scroll and no explicit "task done" steps, so neither appears here.
63
+
64
+ Converter: [`code/finetune/convert_webchain.py`](https://huggingface.co/cklxx/laya-browser/blob/main/code/finetune/convert_webchain.py)
65
+ in [cklxx/laya-browser](https://huggingface.co/cklxx/laya-browser), whose model v19s was trained on the train split
66
+ (held-out test: click operation 0.94, click target top-1 0.48).
67
+
68
+ ## License and attribution
69
+
70
+ CC-BY-4.0, derived from WebChain by webagentlab
71
+ ([paper](https://arxiv.org/abs/2603.05295), [dataset](https://huggingface.co/datasets/webagentlab/webchain)).
72
+ Page texts come from public websites as recorded in WebChain.
data/test.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ced414347ba5fe4acd57a16a96d570705a9300b09120154508c3acf824d7e616
3
+ size 32699955
data/train.jsonl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b49589999e9fd24f834e94b852b4a4db50e56ae15cb56258208c6b257dfd9e0b
3
+ size 110322521