AndroidHowTo dataset

🚨 Attention 🚨 The code in this repo was designed to facilitate the download and generation of the AndroidHowTo dataset. All credit goes to the authors of the Seq2Act paper, I am only reproducing their results. If you use the dataset, please cite their work:

@inproceedings{seq2act,
  title = {Mapping Natural Language Instructions to Mobile UI Action Sequences},
  author = {Yang Li and Jiacong He and Xin Zhou and Yuan Zhang and Jason Baldridge},
  booktitle = {Annual Conference of the Association for Computational Linguistics (ACL 2020)},
  year = {2020},
  url = {https://www.aclweb.org/anthology/2020.acl-main.729.pdf},
}

Dataset files

The generated AndroidHowTo dataset is available in the androidhowto_dataset folder. The downloaded files before generation are available in prep. If you want to generate the dataset yourself, follow the steps below.

Running the code to generate AndroidHowTo dataset:

Download all the files in used_warc.paths

python download_and_extract.py

This will take a LONG time, come back after two days ⏳

For each file in used_warc.paths a output file will be generated inside prep. Make sure the prep folder is empty, or else you might have problems 🐛

After you download and parsed all the files (3,414 in total), you can merge all the files into one, by running:

python merge_files.py

The output file is crawled_output.json.

Then, generate the TFrecords by running:

python -m seq2act.data_generation.create_commoncrawl_dataset \
--input_instruction_json_file="crawled_output.json" \
--input_csv_file="common_crawl_annotation.csv" \
--vocab_file="commoncrawl_rico_vocab_subtoken_44462" \
--output_dir="androidhowto_dataset/"