README.md

metadata

dataset_info:
  features:
    - name: asset_id
      dtype: string
    - name: ad_title
      dtype: string
    - name: flg_revised
      dtype: bool
  splits:
    - name: test
      num_bytes: 45217
      num_examples: 872
  download_size: 59836
  dataset_size: 90434
configs:
  - config_name: default
    data_files:
      - split: test
        path: data/test-*
license: cc-by-sa-4.0
task_categories:
  - text-generation
language:
  - ja
tags:
  - faithfulness
  - advertisement
size_categories:
  - 1K<n<10K
pretty_name: faithcamera

Overview

In ad text generation (ATG), desirable ad text is both faithful and informative. That is, it should be faithful to the input document, while at the same time containing important information that appeals to potential customers.

The existing evaluation data, CAMERA (Mita et al.,2024), is suitable for evaluating informativeness, as it consists of reference ad texts created by ad creators. However, these references often include information unfaithful to the input, which is a notable obstacle in promoting ATG research.

Therefore, we collaborate with in-house ad creators to refine the CAMERA references and develop an alternative Japanese ATG evaluation dataset called FaithCAMERA, in which the faithfulness of references is guaranteed.

Usage

Please use this dataset together with the CAMERA dataset. Specifically, as shown in the sample code below, it is recommended that you use the asset_id to link the examples in the CAMERA dataset and this dataset.

import datasets

camera_dataset = datasets.load_dataset("cyberagent/camera", name="without-lp-images")
faithcamera_dataset = datasets.load_dataset("cyberagent/FaithCAMERA")

test_split_of_camera = camera_dataset['test']
test_split_of_faithcamera = faithcamera_dataset['test']

test_d = {r['asset_id']: r for r in test_split_of_camera}

faithful_test_instances = {}

for record in test_split_of_faithcamera:
    asset_id = record['asset_id']
    info = test_d[asset_id]
    faithful_test_instance = {
        'asset_id': asset_id,
        'keywords': info['kw'].split(" "),
        'lp_meta_description': info['lp_meta_description'],
        'ad_title': record['ad_title'],
        'lp_ocr_sentences': info['parsed_full_text_annotation']['text']
    }
    faithful_test_instances.append(faithful_test_instance)

Dataset Structure

Name	Description
asset_id	This field can be used to match each instance in this dataset with an instance in the CAMERA dataset.
ad_title	ad text (gold reference)
flg_revised	If this field is “True”, it means that the ad text has been rewritten to be faithful to the input. On the other hand, if it is “False”, it means that the ad text in the CAMERA dataset is already faithful to the input, so no rewriting is done.

Note

• Because this dataset is targeted at Japanese ad texts, an instance of asset_id: 100637, where all the text on the landing page (LP) is in English, have been excluded (in the tsv file, the reference ad text for this instance is set as an empty string).

Citation

Please cite this paper if you use our data.

@misc{kato2024faithcameraconstructionfaithfuldataset,
      title={FaithCAMERA: Construction of a Faithful Dataset for Ad Text Generation}, 
      author={Akihiko Kato and Masato Mita and Soichiro Murakami and Ukyo Honda and Sho Hoshino and Peinan Zhang},
      year={2024},
      eprint={2410.03839},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2410.03839}, 
}

References

@inproceedings{mita2024striking,
  title={Striking Gold in Advertising: Standardization and Exploration of Ad Text Generation},
  author={Mita, Masato and Murakami, Soichiro and Kato, Akihiko and Zhang, Peinan},
  booktitle={Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  pages={955--972},
  year={2024}
}

License

This project is licensed under the CC BY-NC-SA 4.0 License.