Dataset Card for MMChat

Table of Contents

• Dataset Card for MMChat
  • Table of Contents
  • Dataset Description
    • Dataset Summary
    • Supported Tasks and Leaderboards
    • Languages
  • Dataset Structure
    • Data Instances
    • Data Fields
    • Data Splits
  • Dataset Creation
    • Curation Rationale
    • Source Data
      • Initial Data Collection and Normalization
      • Who are the source language producers?
    • Annotations
      • Annotation process
      • Who are the annotators?
    • Personal and Sensitive Information
  • Considerations for Using the Data
    • Social Impact of Dataset
    • Discussion of Biases
    • Other Known Limitations
  • Additional Information
    • Dataset Curators
    • Licensing Information
    • Citation Information
    • Contributions

Dataset Description

• Homepage: https://www.zhengyinhe.com/datasets/
• Repository: https://github.com/silverriver/MMChat
• Paper: https://arxiv.org/abs/2108.07154

Dataset Summary

MMChat is a large-scale dialogue dataset that contains image-grounded dialogues in Chinese. Each dialogue in MMChat is associated with one or more images (maximum 9 images per dialogue). We design various strategies to ensure the quality of the dialogues in MMChat.

MMChat comes with 4 different versions:

• mmchat: The MMChat dataset used in our paper.
• mmchat_hf: Contains human annotation on 100K sessions of dialogues.
• mmchat_raw: Raw dialogues used to construct MMChat. mmchat_lccc_filtered: Raw dialogues filtered using the LCCC dataset.

If you what to use high quality multi-modal dialogues that are closed related to the given images, I suggest you to use the mmchat_hf version. If you only care about the quality of dialogue texts, I suggest you to use the mmchat_lccc_filtered version.

Supported Tasks and Leaderboards

• dialogue-generation: The dataset can be used to train a model for generating dialogue responses.
• response-retrieval: The dataset can be used to train a reranker model that can be used to implement a retrieval-based dialogue model.

Languages

MMChat is in Chinese

MMChat中的对话是中文的

Dataset Structure

Data Instances

Several versions of MMChat are available. For mmchat, mmchat_raw, mmchat_lccc_filtered, the following instance applies:

{
  "dialog": ["你只拍出了你十分之一的美", "你的头像竟然换了，奥"],
  "weibo_content": "分享图片",
  "imgs": ["https://wx4.sinaimg.cn/mw2048/d716a6e2ly1fmug2w2l9qj21o02yox6p.jpg"]
}

For mmchat_hf, the following instance applies:

{
  "dialog": ["白百合", "啊？", "有点像", "还好吧哈哈哈牙像", "有男盆友没呢", "还没", "和你说话呢。没回我"],
  "weibo_content": "补一张昨天礼仪的照片",
  "imgs": ["https://ww2.sinaimg.cn/mw2048/005Co9wdjw1eyoz7ib9n5j307w0bu3z5.jpg"],
  "labels": {
    "image_qualified": true, 
    "dialog_qualified": true, 
    "dialog_image_related": true
  }
}

Data Fields

• dialog (list of strings): List of utterances consisting of a dialogue.
• weibo_content (string): Weibo content of the dialogue.
• imgs (list of strings): List of URLs of images.
• labels (dict): Human-annotated labels of the dialogue.
• image_qualified (bool): Whether the image is of high quality.
• dialog_qualified (bool): Whether the dialogue is of high quality.
• dialog_image_related (bool): Whether the dialogue is related to the image.

Data Splits

For mmchat, we provide the following splits:

For other versions, we do not provide the offical split. More stastics are listed here:

Dataset Creation

Curation Rationale

[Needs More Information]

Source Data

Initial Data Collection and Normalization

[Needs More Information]

Who are the source language producers?

[Needs More Information]

Annotations

Annotation process

[Needs More Information]

Who are the annotators?

[Needs More Information]

Personal and Sensitive Information

[Needs More Information]

Considerations for Using the Data

Social Impact of Dataset

[Needs More Information]

Discussion of Biases

[Needs More Information]

Other Known Limitations

[Needs More Information]

Additional Information

Dataset Curators

[Needs More Information]

Licensing Information

other-weibo

This dataset is collected from Weibo. You can refer to the detailed policy required to use this dataset. Please restrict the usage of this dataset to non-commerical purposes.

Citation Information

@inproceedings{zheng2022MMChat,
  author    = {Zheng, Yinhe and Chen, Guanyi and Liu, Xin and Sun, Jian},
  title     = {MMChat: Multi-Modal Chat Dataset on Social Media},
  booktitle = {Proceedings of The 13th Language Resources and Evaluation Conference},
  year      = {2022},
  publisher = {European Language Resources Association},
}

@inproceedings{wang2020chinese,
  title={A Large-Scale Chinese Short-Text Conversation Dataset},
  author={Wang, Yida and Ke, Pei and Zheng, Yinhe and Huang, Kaili and Jiang, Yong and Zhu, Xiaoyan and Huang, Minlie},
  booktitle={NLPCC},
  year={2020},
  url={https://arxiv.org/abs/2008.03946}
}

Contributions

Thanks to Yinhe Zheng for adding this dataset.