Nidhogg-zh/Multi-Querier_Dialogue

Multi-Querier Dialogue (MQDialog) Dataset

📑 Dataset Details

Dataset Description

This is the dataset of "Querier-Aware LLM: Generating Personalized Responses to the Same Query from Different Queriers".

The Multi-Querier Dialogue (MQDialog) dataset is designed to facilitate research in querier-aware personalization. It contains dialogues with various queriers for each reponder. The dataset is derived from English and Chinese scripts of popular TV shows and real-world conversations. It includes dialogues where selected leading actors act as responders, while other characters or contacts serve as queriers. The dataset contains a diverse set of 12 responders and 173 queriers. The dataset supports research on dialogue generation, response evaluation, and querier-aware personalization in multi-turn conversations.

Dataset Sources

• —English scripts: The Big Bang Theory, Friends, and Modern Family.
• —Chinese scripts: My Own Swordsman and Empresses in the Palace.
• —Real-world conversations (WeChat): Records from a single user, focusing on two-person chats. (Not public, but you can extract the data using the code we provided)

📁 Dataset Structure

• —Responders: 12 leading actors from TV scripts and a single WeChat user.
• —Queriers: 173 individuals interacting with the responders, the detailed information is listed in the Table.
• —Splits: Randomly divided into training (3761 dialogues per responder on average) and testing (917 dialogues per responder on average).

<table> <tr> <td>Language</td> <td>Data Source</td> <td># Queriers</td> <td>Querier Examples</td> <td>Responder</td> <td># train</td> <td># test</td> <td>Subset</td> </tr> <tr> <td rowspan="6">English</td> <td rowspan="2">The Big Bang Theory</td> <td>14</td> <td>Priya, Barry, Howard, Leonard, etc.</td> <td>Sheldon</td> <td>4805</td> <td>1101</td> <td>bigbangsheldon</td> </tr> <tr> <td>12</td> <td>Bernadette, Penny, Raj, Stuart, etc.</td> <td>Leonard</td> <td>4607</td> <td>1014</td> <td>bigbangleonard</td> </tr> <tr> <td rowspan="2">Friends</td> <td>12</td> <td>Amy, Chandler, Charlie, Joey, etc.</td> <td>Rachel</td> <td>3768</td> <td>870</td> <td>friendsrachel</td> </tr> <tr> <td>20</td> <td>Ben, Mike, Gary, Paul, etc.</td> <td>Ross</td> <td>3839</td> <td>960</td> <td>friendsross</td> </tr> <tr> <td rowspan="2">Modern Family</td> <td>9</td> <td>Alex, Cameron, Dylan, Gloria, etc.</td> <td>Claire</td> <td>1161</td> <td>281</td> <td>modernfamilyclaire</td> </tr> <tr> <td>8</td> <td>Haley, Jay, Luke, Mitchell, etc.</td> <td>Phil</td> <td>881</td> <td>246</td> <td>modernfamilyphil</td> </tr> <tr> <td rowspan="6">Chinese</td> <td rowspan="3">My Own Swordsman</td> <td>16</td> <td>Bai Sanniang, Guo Furong, Mo Xiaobei, etc.</td> <td>Tong Xiangyu</td> <td>3200</td> <td>831</td> <td>swordsmantongxiangyu</td> </tr> <tr> <td>16</td> <td>Bao Daren, Ji Wuming, Zhu Wushuang, etc.</td> <td>Bai Zhantang</td> <td>2995</td> <td>857</td> <td>swordsmanbaizhantang</td> </tr> <tr> <td>8</td> <td>Li Dazui, Xing Butou, Yan Xiaoliu, etc.</td> <td>Lv Xiucai</td> <td>1635</td> <td>409</td> <td>swordsmanlvxiucai</td> </tr> <tr> <td rowspan="2">Empresses in the Palace</td> <td>17</td> <td>Cao Guiren, Mei Zhuang, Liu Zhu, etc.</td> <td>Zhen Huan</td> <td>1229</td> <td>350</td> <td>palacezhenhuan</td> </tr> <tr> <td>11</td> <td>Consort Hua, Empress, Huan Bi, etc.</td> <td>Emperor</td> <td>704</td> <td>200</td> <td>palace_emperor</td> </tr> <tr> <td>WeChat Records</td> <td>30</td> <td>Author's contacts</td> <td>Author</td> <td>-</td> <td>-</td> <td>-</td> </tr> </table> *Note: The last response from the responder serves as the ground truth, while preceding dialogues constitute the dialogue history. We provide a compact version of the training set that preserves contextual relationships, because, during training, the answers within the dialogue history can be used to compute the loss, eliminating the need for the answer in the last sentence.

Data Files & Code

For each responder, dialogues with different queriers are stored in the corresponding folder, diags_two_role_{responder_name}. Intermediate results from data processing are also provided. The final datasets used for querier-aware personalization are:

• —{script_name}_diags_{responder_name}_{querier_name}_{responder_name}_response_L512_dev.json
• —{script_name}_diags_{responder_name}_{querier_name}_{responder_name}_response_L512_train.json

Additionally, dialogues with different queriers are clustered based on query similarity. The clustering results are stored in the diags_two_role_{responder_name}_clustered folder.

We have also provided the preprocessed raw data for each scripts named with {script_name}_dialgs.json. To extract dialogues for one responder by yourself, please run the python file extract_two_role_diag_{responder_name}.py under each subfolder.

Here is an example:

git clone https://huggingface.co/datasets/Nidhogg-zh/Multi-Querier_Dialogue
cd Multi-Querier_Dialogue/swordsman
python extract_two_role_diag_tongxiangyu.py

Related functions:

• —get_role_list(): get whole role name
• —extract_diag_between_two_role(): extract and only reserve diags between two roles
• —clean_diag(): remove duplicates, remove conversations with only one person, and remove empty values
• —clean_diag_with_repeated(): remove conversations with only one person, and remove empty values
• —split_train_and_dev(): split training set and validation set
• —split_diag_with_sliding_window(): construct diags with limited length through a sliding window
• —extract_diag_for_target_from_role_conv(): only reserve diags that the response is from target role

Data Instances

Below is an example from the dataset, it contains conversations between the target_role (i.e. responder) and the input_role (i.e. querier).

{
    "id": "episode_14_chunk_6_index_0_part2_piece_0",
    "conversations": [
        {
            "from": "Bernadette",
            "value": "Did you hear? Isn’t it terrible?"
        },
        {
            "from": "Leonard",
            "value": "Have you seen him?"
        },
        {
            "from": "Bernadette",
            "value": "They wouldn’t let me in. Oh my Howie."
        },
        {
            "from": "Leonard",
            "value": "It’ll be okay. It’ll be okay."
        }
    ],
    "target_role": "Leonard",
    "target_role_short": "Leonard",
    "input_role": "Bernadette",
    "input_role_short": "Bernadette",
    "role_pair_id": 8,
    "cluster_id": 2 (Only in the clustered data)
}

🚀 Quick Start

We partitioned the data from different responders into distinct subsets, each comprising a training set and a test set.

For example, to load the dataset of Sheldon from The Big Bang Theory, you can use following code:

import datasets 
dataset = datasets.load_dataset("Nidhogg-zh/Multi-Querier_Dialogue", "bigbang_sheldon")

📦 Direct Use

The dataset is suitable for:

• —Training and evaluating querier-aware multi-turn dialogue systems.
• —Studying personality-aligned response generation.
• —Benchmarking the performance of dialogue models with multi-querier setups.
• —For implementation details and further usage, please refer to the corresponding GitHub repository: https://github.com/Nidryen-zh/QuerierAwareResponder

🔧 Dataset Creation

Curation Rationale

MQDialog was created to address the need for a multilingual, multi-querier dataset that reflects querier-aware personalized response generation in diverse conversational contexts.

Data Collection and Processing

• —Scripts: Extracted dialogues between a responder (leading actor) and queriers (other characters), ensuring a clean dataset by removing errors, repeated content, and irrelevant entries.
• —Real-world records: Focused on one-on-one conversations, with new dialogue sessions defined by a time gap (e.g., 3 hours).
• —Filtering: Queriers with fewer than 20 interactions were excluded to ensure meaningful analysis.

Recommendations

• —Use the dataset in conjunction with other corpora to mitigate cultural or linguistic biases.
• —Ensure responsible use of the data, particularly when training models for real-world applications.

Citation

BibTeX

Please kindly cite our paper if helps your research:

@misc{zeng2024personalizedllmgeneratingcustomized,
      title={Personalized LLM for Generating Customized Responses to the Same Query from Different Users}, 
      author={Hang Zeng and Chaoyue Niu and Fan Wu and Chengfei Lv and Guihai Chen},
      year={2024},
      eprint={2412.11736},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2412.11736}, 
}