aworld_visualwebarena
收藏魔搭社区2025-09-29 更新2025-09-06 收录
下载链接:
https://modelscope.cn/datasets/kevinakira/aworld_visualwebarena
下载链接
链接失效反馈官方服务:
资源简介:
数据集文件元信息以及数据文件,请浏览“数据集文件”页面获取。
当前数据集卡片使用的是默认模版,数据集的贡献者未提供更加详细的数据集介绍,但是您可以通过如下GIT Clone命令,或者ModelScope SDK来下载数据集
#### 下载方法
:modelscope-code[]{type="sdk"}
:modelscope-code[]{type="git"}
# Recon-Act: A Self-Evolving Multi-Agent Browser-Use System via Web Reconnaissance, Tool Generation, and Task Execution
## json format
```
{
"0": {
"info": {
"fail_error": "",
"observation_metadata": {
"text": {
"url": <currnet_url>
"obs_nodes_info": <text observation info from browser environment, align with the visualwebarena dataset>,
"tabs": <\tabs info>
},
"image": {
"obs_nodes_info": {
<bounding box coordinate (xywh) align with the text obs_nodes_info>
}
}
},
"intent": <intent>,
"input_images": <input_images>,
"ss_ori": <original screenshot relative path>,
"ss_som": <som screenshot relative path>
},
"action": { # same with visualwebarena dataset
"action_type": <action type number>,
"coords": [
0.0,
0.0
],
"element_role": 0,
"element_name": "",
"text": [],
"page_number": 0,
"url": "",
"nth": 0,
"pw_code": "",
"element_id": "",
"key_comb": "",
"direction": "",
"answer": <answer from stop action>,
"raw_prediction": <raw response of action team>,
"action_info": {
"pred_action_history": "",
"pred_action_description": "",
"pred_action": "",
"pred_action_type": "",
"pred_bbox": "",
"pred_type_value": "",
"pred_click_point": "",
"parse_error_msg": "",
"content_to_memo": ""
}
}
}
}
```
action_type see the codes below:
```
class ActionTypes(IntEnum):
"""Valid action types for browser env."""
NONE = 0
# mouse wheel and keyboard, universal across all action spaces
SCROLL = 1
KEY_PRESS = 2
# low level mouse and keyboard actions
MOUSE_CLICK = 3
KEYBOARD_TYPE = 4
MOUSE_HOVER = 5
# mid level mouse and keyboard actions
CLICK = 6
TYPE = 7
HOVER = 8
# page level actions, universal across all action spaces
PAGE_FOCUS = 9
NEW_TAB = 10
GO_BACK = 11
GO_FORWARD = 12
GOTO_URL = 13
PAGE_CLOSE = 14
# high-leval actions that playwright support
CHECK = 15
SELECT_OPTION = 16
STOP = 17
CLEAR = 18
UPLOAD = 19
WAIT = 20
DRAG = 21
```
数据集文件的元信息与数据文件,请前往「数据集文件」页面获取。
当前数据集卡片采用默认模板制作,数据集贡献者未提供更详尽的数据集介绍,但您可通过下述Git Clone命令或ModelScope软件开发工具包(SDK)下载该数据集。
#### 下载方法
:modelscope-code[]{type="sdk"}
:modelscope-code[]{type="git"}
# Recon-Act:基于网页侦察、工具生成与任务执行的自进化多智能体浏览器交互系统
## 格式说明
{
"0": {
"info": {
"fail_error": "",
"observation_metadata": {
"text": {
"url": <currnet_url>
"obs_nodes_info": <text observation info from browser environment, align with the visualwebarena dataset>,
"tabs": < abs info>
},
"image": {
"obs_nodes_info": {
<bounding box coordinate (xywh) align with the text obs_nodes_info>
}
}
},
"intent": <intent>,
"input_images": <input_images>,
"ss_ori": <original screenshot relative path>,
"ss_som": <som screenshot relative path>
},
"action": { # 与视觉网页竞技场(VisualWebArena)数据集一致
"action_type": <action type number>,
"coords": [
0.0,
0.0
],
"element_role": 0,
"element_name": "",
"text": [],
"page_number": 0,
"url": "",
"nth": 0,
"pw_code": "",
"element_id": "",
"key_comb": "",
"direction": "",
"answer": <answer from stop action>,
"raw_prediction": <raw response of action team>,
"action_info": {
"pred_action_history": "",
"pred_action_description": "",
"pred_action": "",
"pred_action_type": "",
"pred_bbox": "",
"pred_type_value": "",
"pred_click_point": "",
"parse_error_msg": "",
"content_to_memo": ""
}
}
}
}
动作类型详见下述代码:
class ActionTypes(IntEnum):
"""浏览器环境下的有效动作类型。"""
NONE = 0
# 鼠标滚轮与键盘操作,适用于所有动作空间
SCROLL = 1
KEY_PRESS = 2
# 低级鼠标与键盘操作
MOUSE_CLICK = 3
KEYBOARD_TYPE = 4
MOUSE_HOVER = 5
# 中级鼠标与键盘操作
CLICK = 6
TYPE = 7
HOVER = 8
# 页面级操作,适用于所有动作空间
PAGE_FOCUS = 9
NEW_TAB = 10
GO_BACK = 11
GO_FORWARD = 12
GOTO_URL = 13
PAGE_CLOSE = 14
# Playwright支持的高级操作
CHECK = 15
SELECT_OPTION = 16
STOP = 17
CLEAR = 18
UPLOAD = 19
WAIT = 20
DRAG = 21
提供机构:
maas
创建时间:
2025-09-03



