CoolFace
Datasetpublic

nu-dialogue/multi-relational-multi-party-chat-corpus

Multi-Relational Multi-Party Chat Corpus (MRMP): Japanese text-based chats comprising first-time-meeting dialogues and family-included dialogues

sourceHugging Facecc-by-sa-4.0updated 1y agoView on Hugging Face
3likes37downloads
Dataset Card

![CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/deed.ja) ![Multi-Relational Multi-Party Chat Corpusのバージョン](https://github.com/nu-dialogue/multi-relational-multi-party-chat-corpus)

Dataset Card for Multi-Relational Multi-Party Chat Corpus

Dataset Description

Dataset Summary

Multi-Relational Multi-Party Chat Corpus は,初対面や家族といった話者間の関係性に着目した,約1,000件の日本語雑談対話からなるコーパスです.詳細は GitHub をご覧ください.

[!CAUTION] 本コーパスの使用にあたっては,次のことに十分注意してください. 本コーパスのデータから個人を特定しようとしないこと. 本コーパスを,特定の話者へのなりすましに用いないこと. * 本コーパスを話者の属性や性格特性の推定などに用いる際は,自身の情報を推定されたくない話者の権利についても留意すること (Tatman, 2020).

Dataset Usage

python
from datasets import load_dataset

dialogue_dataset = load_dataset(
        "nu-dialogue/multi-relational-multi-party-chat-corpus",
        name="dialogue",
        trust_remote_code=True
    )
interlocutor_dataset = load_dataset(
        "nu-dialogue/multi-relational-multi-party-chat-corpus",
        name="interlocutor",
        trust_remote_code=True
    )

Dataset Structure

Data Instances

本コーパスは,対話データと話者データからなります.

対話データには,対話ID,対話の種類,話者ID,発話,および,話者ごとの評価スコアが含まれます.評価スコアは,1が低いことを,5が高いことを示します.

json
{
    "dialogue_id": "B10008",
    "dialogue_type": 1,
    "interlocutors": ["うさぎ", "えのき", "てばさき"],
    "utterances": {
        "utterance_id": [0, 1, 2, 3, 4, 5, ...],
        "interlocutor_id": ["うさぎ", "えのき", "てばさき", "うさぎ", "えのき", "てばさき", ...],
        "text": ["こんにちは", "こんにちは", "こんにちは", "何かスポーツとかやられてますか?", "特にやってないです", "@うさぎ 運動は苦手ですね…", ...],
        "mention_to": [[], [], [], [], [], ["うさぎ"]]
    },
    "evaluations": {
        "interlocutor_id": ["うさぎ", "えのき", "てばさき"],
        "informativeness": [4, 4, 4],
        "comprehension": [4, 4, 4],
        "familiarity": [4, 4, 4],
        "interest": [4, 4, 4],
        "proactiveness": [4, 5, 4],
        "satisfaction": [4, 4, 5]
    }
}

話者データには,話者IDをキーとして,話者ID,ペルソナ,性格特性,属性,テキストチャットの経験が含まれます.性格特性スコアが高いほど,その性格的な傾向が強いことを示します.

json
{
    "interlocutor_id": "てばさき",
    "persona": [
        "現在文系の大学生である。",
        "好きな食べ物はジャーキー、嫌いな食べ物は生野菜だ。",
        "ゲームや漫画、小説が好きで、最近は原神というゲームにはまっている。",
        "地元のボランティアに申し込んでみたい。",
        "休日はずっと寝て過ごしたいけど課題に追われている。",
        "外出の用事があるけど寒い日はお湯を飲むことにしている。",
        "猫派だけど小動物は大体好き。",
        "ロールキャベツはなんとか作れる。",
        "埼玉生まれ埼玉育ちの埼玉県民である。",
        "最近はボカロや谷山浩子さん、Sound Horizonの曲をよく聴く。"
    ],
    "personality": {
        "BigFive_Openness": 4.083333492279053,
        "BigFive_Conscientiousness": 2.3333332538604736,
        "BigFive_Extraversion": 3.9166667461395264,
        "BigFive_Agreeableness": 4.916666507720947,
        "BigFive_Neuroticism": 4.333333492279053,
        "KiSS18_BasicSkill": 2.3333332538604736,
        "KiSS18_AdvancedSkill": 3.0,
        "KiSS18_EmotionalManagementSkill": 2.3333332538604736,
        "KiSS18_OffenceManagementSkill": 3.0,
        "KiSS18_StressManagementSkill": 3.3333332538604736,
        "KiSS18_PlanningSkill": 2.0,
        "IOS": 3,
        "ATQ_Fear": 4.142857074737549,
        "ATQ_Frustration": 4.5,
        "ATQ_Sadness": 5.285714149475098,
        "ATQ_Discomfort": 3.1666667461395264,
        "ATQ_ActivationControl": 4.142857074737549,
        "ATQ_AttentionalControl": 1.7999999523162842,
        "ATQ_InhibitoryControl": 3.857142925262451,
        "ATQ_Sociability": 4.599999904632568,
        "ATQ_HighIntensityPleasure": 5.5714287757873535,
        "ATQ_PositiveAffect": 5.599999904632568,
        "ATQ_NeutralPerceptualSensitivity": 4.800000190734863,
        "ATQ_AffectivePerceptualSensitivity": 4.199999809265137,
        "ATQ_AssociativeSensitivity": 6.800000190734863,
        "SMS_Extraversion": 2.5,
        "SMS_OtherDirectedness": 2.75,
        "SMS_Acting": 3.0
    },
    "demographic_information": {
        "gender": 1,
        "age": 0,
        "education": 0,
        "employment_status": 2,
        "region_of_residence": 10
    },
    "text_chat_experience": {
        "age_of_first_chat": 1,
        "frequency": 0,
        "chatting_partners": [0, 1, 2],
        "typical_chat_content": "業務連絡、雑談、助言、相談、感動の共有"
    },
    "pair_information": {
        "pair_flag": True,
        "relationship": [1],
        "relationship_detail": "産みと育ての母",
        "pair_interlocutor_id": "えのき"
    }
}

Data Fields

対話データの構造は次の通りです.

  • —dialogue_id (string): 対話ID.アルファベットは対話の種類(例:B10008のB).数字の最初3桁は話者の組み合わせの通し番号(例:B10008の100).数字の最後2桁はその話者の組み合わせが実施した対話の通し番号(例:B10008の08).
  • —dialogue_type (string): 対話の種類.First time (A), Family (B) のいずれか.
  • —interlocutors (sequence): 話者IDのリスト.
  • —relationship (sequence): 関係性のある話者IDのリスト.
  • —utterances (sequence): 発話のリスト.
  • —utterance_id (int32): 発話ID.対話内で固有.0始まりのインデックス.
  • —interlocutor_id (string): 話者ID.
  • —text (string): 発話テキスト.
  • —mention_to (sequence): メンション先の話者IDのリスト.
  • —evaluations (sequence): 話者ごとの評価スコアのリスト.
  • —interlocutor_id (string): 話者ID.
  • —informativeness (int32): 情報量の評価スコア (1~5).
  • —comprehension (int32): 理解度の評価スコア (1~5).
  • —familiarity (int32): 親しみやすさの評価スコア (1~5).
  • —interest (int32): 興味の評価スコア (1~5).
  • —proactiveness (int32): 積極性の評価スコア (1~5).
  • —satisfaction (int32): 満足度の評価スコア (1~5).

話者データの構造は次の通りです.

  • —interlocutor_id (strint): 話者ID.
  • —persona (sequence): ペルソナ.10文からなる.
  • —personality (struct): 性格特性.
  • —BigFive_* (float32): Big Five のスコア (1~7).
  • —KiSS18_* (float32): Kikuchi's Scale of Social Skills のスコア (1~5).
  • —IOS (int32): Inclusion of Others in the Self のスコア (1~7).
  • —ATQ_* (float32): Adult Temperament Questionnaire のスコア (1~7).
  • —SMS_* (float32): Self-Monitoring Scale のスコア (1~5).
  • —demographic_information (struct): 属性.
  • —gender (string): 性別.Male, Female, Other のいずれか.
  • —age (string): 年齢.-19, 20-29, 30-39, 40-49, 50-59, 60-69 のいずれか.
  • —education (string): 教育歴.High school graduate, Two-year college, Four-year college, Postgraduate, Other のいずれか.
  • —employment_status (string): 就労状況.Employed, Homemaker, Student, Retired, Unable to work, None のいずれか.
  • —region_of_residence (string): 居住地域.日本の都道府県名.
  • —text_chat_experience (struct): テキストチャットの経験.
  • —age_of_first_chat (string): 初めてテキストチャットをした時の年齢.-9, 10-19, 20-29, 30-39, 40-49, 50-59 のいずれか.
  • —frequency (string): 普段のテキストチャットの頻度.Every day, Once every few days, Once a week, Less frequent than these のいずれか.
  • —chatting_partners (sequence): 普段のテキストチャットの相手.Family, Friend, Colleague, Other のいずれか.
  • —typical_chat_content (string): 普段のテキストチャットの内容.
  • —pair_information (struct): 関係性のある話者の情報.
  • —pair_flag (bool): 関係性のある話者がいるかどうか.
  • —relationship (sequence): 話者との関係.Frequent acquaintance, Someone known for years, Someone engaged in joint activities のいずれか.
  • —relationship_detail (string): 話者との関係の詳細.
  • —pair_interlocutor_id (string): 関係性のある話者の話者ID.

Data Splits

対話データの分割は次の通りです.

trainvalidationtest
対話数960--
発話数106,680--

話者データの分割は次の通りです.

trainvalidationtest
話者数66--

Citation Information

bibtex
@inproceedings{tsuda-etal-2025-multi-relational-multi-party-chat-corpus,
title = "Constructing a Multi-Party Conversational Corpus Focusing on Interlocutor Relationships",
    author = "Tsuda, Taro and
      Yamashita, Sanae and
      Inoue, Koji and
      Kawahara, Tatsuya
      and Higashinaka, Ryuichiro",
    booktitle="Proceedings of the 29th Workshop on the Semantics and Pragmatics of Dialogue",
    year = "2025",
    pages = "193--202"
}

@inproceedings{tsuda-etal-2025-multi-relational-multi-party-chat-corpus-ja,
    title = "{M}ulti-{R}elational {M}ulti-{P}arty {C}hat {C}orpus: 話者間の関係性に着目したマルチパーティ雑談対話コーパス",
    author = "津田 太郎 and 山下 紗苗 and 井上 昂治 and 河原 達也 and 東中 竜一郎",
    booktitle = "言語処理学会第31回年次大会発表論文集",
    year = "2025",
    pages = "4011--4016"
}

Acknowledgment

本コーパスは,JSTムーンショット型研究開発事業,JPMJMS2011 の支援を受けて構築しました.

<img src="https://github.com/nu-dialogue/multi-relational-multi-party-chat-corpus/blob/main/Moonshot_Logo.png?raw=true" alt="ムーンショットのロゴ" style="max-width: 100%; width: 300px;">

License

本コーパスは CC BY-SA 4.0 の下で提供されます.

![CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/deed.ja)