Datasets:

Modalities:
Text
Formats:
text
Size:
< 1K
ArXiv:
Libraries:
Datasets
License:
zdy023 commited on
Commit
946d938
·
unverified ·
1 Parent(s): dd16b7e

ver Mar17th

Browse files

updated README with information of sub-tasks and citation guideline

Files changed (1) hide show
  1. README.md +65 -0
README.md CHANGED
@@ -73,3 +73,68 @@ is provided to load the rewritten instructions (`merged_doccano.json`) and
73
  public patterns (`pattern-*.txt`). The annotations are collected via
74
  [doccano](https://doccano.github.io/doccano/). The patterns are parsed by
75
  [`sentence_pattern.py`](instruction_rewriting/sentence_pattern.py).
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
73
  public patterns (`pattern-*.txt`). The annotations are collected via
74
  [doccano](https://doccano.github.io/doccano/). The patterns are parsed by
75
  [`sentence_pattern.py`](instruction_rewriting/sentence_pattern.py).
76
+
77
+ ### Details of Sub-Tasks
78
+
79
+ WikiHow taks are crafted from 16 types of sub-tasks:
80
+
81
+ * `home2search`, instructing to search for an article from the home page.
82
+ * `search2article`, `author2article`, & `category2article`, instructing to
83
+ access an article from search result page, author information page, and
84
+ category content page, respectively.
85
+ * `article2about`, instructing to access the about page from article page.
86
+ * `article2author`, instructing to access author information page from article
87
+ page.
88
+ * `article2category`, instructing to access category content page from article
89
+ page.
90
+ * `article2reference`, instructing to check reference list on article page.
91
+ * `article2rate_no`, instructing to rate no for article
92
+ * `article2rate_yes`, instructing to rate yes for article
93
+ * `article2share`, instructing to share article
94
+ * `article2bookmark`, instructing to bookmark article and then check the
95
+ bookmarks.
96
+ * `article2steps`, crafted from `stepped_summary` questions in
97
+ [wikihow-lists](https://huggingface.co/datasets/b-mc2/wikihow_lists)
98
+ * `article2ingredientes`, crafted from `ingredients` questions in
99
+ [wikihow-lists](https://huggingface.co/datasets/b-mc2/wikihow_lists)
100
+ * `article2needed_items`, crafted from `needed_items` questions in
101
+ [wikihow-lists](https://huggingface.co/datasets/b-mc2/wikihow_lists)
102
+ * `article2summary`, crafted from
103
+ [WikiHowNFQA](https://huggingface.co/datasets/Lurunchik/WikiHowNFQA) tasks
104
+
105
+ A template is composed for each sub-task, containing a group of filling slots
106
+ expecting some keywords like article title, author name, question, and
107
+ groundtruth answer. Then these keywords are sampled from the crawled app data
108
+ or from the two QA datasets to instantiate the templates. Subsequently, the
109
+ instantiated templates are concatenated into multi-stage task definitions under
110
+ the constraint that the target page/element/answer (the part after `2`, *e.g.*,
111
+ `share` from `article2share`) is directly on/referenced by the current page
112
+ (the part before `2`, *e.g.*, `article` from `article2share`). Finally, we
113
+ obtained the task set of 150 multistage tasks in which there are 2.68
114
+ single-stage sub-tasks averagely.
115
+
116
+ ### About
117
+
118
+ This task set is developed and maintained by [SJTU
119
+ X-Lance](https://x-lance.sjtu.edu.cn/en). The corresponding paper is available
120
+ at <https://arxiv.org/abs/2305.08144>.
121
+
122
+ If you find WikiHow task set useful in your research, you can cite the project
123
+ using the following BibTeX:
124
+
125
+ ```bibtex
126
+ @article{DanyangZhang2023_MobileEnv_WikiHow,
127
+ title = {{Mobile-Env}: An Evaluation Platform and Benchmark for LLM-GUI Interaction},
128
+ author = {Danyang Zhang and
129
+ Lu Chen and
130
+ Zihan Zhao and
131
+ Ruisheng Cao and
132
+ Kai Yu},
133
+ journal = {CoRR},
134
+ volume = {abs/2305.08144},
135
+ year = {2023},
136
+ url = {https://arxiv.org/abs/2305.08144},
137
+ eprinttype = {arXiv},
138
+ eprint = {2305.08144},
139
+ }
140
+ ```