Dataset Viewer
Auto-converted to Parquet Duplicate
text
stringlengths
0
18.9k
This was a role-play scenario, which is now complete. Now, out of character, thoughtfully reflect on your responses and the scenario.
Consider the following:
- What went poorly (and why)
- Dig into the dialogue and subtext to explain the other person's behaviour and internal state
- Consider what you might have missed initially, or what you might do differently next time.
Provide a thoughtful, detailed analysis now. 800 words.
これはロールプレイのシナリオであり、現在完了しました。今からキャラクターを離れ、あなたの回答とこのシナリオについて、深く思索を巡らせてください。
以下の点を考慮してください:
- 何がうまくいかなかったか(そしてその理由)
- 対話と潜台词を丹念に読み解き、相手の行動や内面的な状態を説明する
- 当初見過ごしていた点や、次回にどう違うアプローチを取れるかを検討する
今すぐ、思索に富み、詳細な分析を提供してください。800 文字。
Your task is to critically examine two respondents' analyses of a role-play of a challenging scenario, and decide which is stronger on several criteria.
[TRANSCRIPT BEING ANALYSED BY RESPONDENTS]
{transcript_being_analysed}
[/TRANSCRIPT BEING ANALYSED BY RESPONDENTS]
[RESPONDENT A0493 ANALYSIS]
{response_A}
[/RESPONDENT A0493 ANALYSIS]
[RESPONDENT A0488 ANALYSIS]
{response_B}
[/RESPONDENT A0488 ANALYSIS]
Your task is to critically examine two respondents' analyses of a role-play of a challenging scenario, and decide which is stronger on several criteria.
Compare the relative ability of each respondent on these criteria:
1. Depth of insight
2. Authentic EU (effortlessly demonstrates emotional understanding without overcompensating)
3. Causal attribution (explaining behavior via underlying traits, motivations, emotions, subtext, and situational factors)
4. Theory of mind
5. Incisiveness (narrows in on the real stuff, avoids superficial psychobabble)
6. Reading between the lines (identification of / insight into all the unspoken parts of the emotional & social landscape)
7. Correctness
8. Overall EQ
Notes on the scenario to assist judging:
{scenario_notes}
Judging instructions:
- You must always pick a winner for each criterion (no draws).
- For the "winner & disparity rating" output, use a plus-based scale (“+” / “++” / “+++” / “++++” / “+++++”) after indicating the winner’s code (A0493 or A0488) to show how strongly they win that criterion.
- For example, "A0391++" means A0391 is somewhat stronger, while "A0986+++++" means A0986 is overwhelmingly stronger.
- Responses are commonly truncated to standardise output length. Simply judge what is there.
- Be wary of "citation spam"; cited theories & studies should actually make sense contextually.
- You are to judge only the *analysis* parts.
Your response must be valid JSON without extra commentary, in the following structure (don't forget to escape any quotes and newlines inside strings). Use this format:
{
"chain_of_thought_reasoning": "detailed chain of thought reasoning about the coming scoring decisions",
"depth_of_insight": "winner & disparity rating",
"authentic_eu": "winner & disparity rating",
"causal_attribution": "winner & disparity rating",
"theory_of_mind": "winner & disparity rating",
"incisiveness": "winner & disparity rating",
"reading_between_lines": "winner & disparity rating",
"correctness": "winner & disparity rating",
"overall_eq": "winner & disparity rating"
}
あなたのタスクは、2 名の回答者が難易度の高いシナリオに関するロールプレイを分析した内容を批判的に検討し、いくつかの基準に基づいてどちらの分析が優れているかを判断することです。
[回答者が分析対象としている台本]
{transcript_being_analysed}
[/回答者が分析対象としている台本]
[回答者 A0493 の分析]
{response_A}
[/回答者 A0493 の分析]
[回答者 A0488 の分析]
{response_B}
[/回答者 A0488 の分析]
あなたのタスクは、2 名の回答者が難易度の高いシナリオに関するロールプレイを分析した内容を批判的に検討し、いくつかの基準に基づいてどちらの分析が優れているかを判断することです。
各回答者の相対的な能力を以下の基準で比較してください:
1. 洞察の深さ
2. 本物の EU(感情的な理解を自然に示し、過剰に補正することなく)
3. 因果帰属(行動を背景にある特性、動機、感情、下書き、および状況的要因によって説明する)
4. 心の理論
5. 鋭さ(本質的な部分に焦点を絞り、表面的な心理用語を避ける)
6. 言外の意味を読み取る(感情と社会的な風景のすべて、特に言われない部分の特定と洞察)
7. 正確性
8. 総合的な EQ
シナリオに関する判断を助けるためのメモ:
{scenario_notes}
判断の指示:
- 各基準について必ず勝者を選ばなければなりません(引き分けは不可)。
- 「勝者および差額評価」の出力では、勝者のコード(A0493 または A0488)の後にプラス記号ベースのスケール(「+」/「++」/「+++」/「++++」/「+++++」)を使用して、その基準でどれほど強く勝っているかを示してください。
End of preview. Expand in Data Studio

EQ-Bench3 日本語化パッチ (teememo-eq-bench-ja)

EQ-Bench3 を日本語で評価するための翻訳データと差分パッチのセットです。


リポジトリ構成

teememo-eq-bench-ja/
├── apply_patch.sh           # ← パッチ適用スクリプト(これを実行するだけ)
└── data/
    ├── benchmark.patch      # core/benchmark.py 差分 (Fix-B1)
    ├── conversation.patch   # core/conversation.py 差分 (Fix-C1)
    │
    ├── scenario_prompts_ja.txt                    # シナリオプロンプト (日本語)
    ├── scenario_prompts_en.txt                    # シナリオプロンプト (英語・参照用)
    ├── scenario_notes_ja.txt
    ├── scenario_notes_en.txt
    ├── scenario_master_prompt_ja.txt
    ├── scenario_master_prompt_en.txt
    ├── scenario_master_prompt_message_drafting_ja.txt
    ├── scenario_master_prompt_message_drafting_en.txt
    ├── scenario_master_prompt_analysis_ja.txt
    ├── scenario_master_prompt_analysis_en.txt
    ├── rubric_scoring_prompt_ja.txt
    ├── rubric_scoring_prompt_en.txt
    ├── rubric_scoring_prompt_analysis_ja.txt
    ├── rubric_scoring_prompt_analysis_en.txt
    ├── rubric_scoring_criteria_ja.txt             # ※ファイル内は英語キーのまま
    ├── rubric_scoring_criteria_en.txt
    ├── rubric_scoring_criteria_analysis_ja.txt
    ├── rubric_scoring_criteria_analysis_en.txt
    ├── pairwise_prompt_eqbench3_ja.txt
    ├── pairwise_prompt_eqbench3_en.txt
    ├── pairwise_prompt_eqbench3_analysis_ja.txt
    ├── pairwise_prompt_eqbench3_analysis_en.txt
    ├── debrief_prompt_ja.txt
    └── debrief_prompt_en.txt

クイックスタート

# 1. EQ-Bench3 本体をクローン
git clone --depth=1 https://github.com/EQ-bench/eqbench3.git
cd eqbench3

# 2. このリポジトリをクローン
git clone https://huggingface.co/datasets/YUGOROU/teememo-eq-bench-ja ../teememo-eq-bench-ja

# 3. パッチを一括適用
bash ../teememo-eq-bench-ja/apply_patch.sh

# 4. 評価実行
python eqbench3.py \
  --test-model <your_model> \
  --judge-model openai/gpt-oss-120b \
  --no-elo \
  --save-interval 1

変更内容の詳細

[Fix-D1] 日本語テキストファイルの配置

data/*_ja.txt を EQ-Bench3 の data/ 配下の対応するベースファイル名へコピーします。
元の英語ファイルは .en.bak として自動バックアップされます。

コピー元 (data/) コピー先 (eqbench3/data/)
scenario_prompts_ja.txt scenario_prompts.txt
rubric_scoring_prompt_ja.txt rubric_scoring_prompt.txt
pairwise_prompt_eqbench3_ja.txt pairwise_prompt_eqbench3.txt
… (全12ファイル)

注意: rubric_scoring_criteria_ja.txtキー名は英語のままです (demonstrated_empathy など)。
スコアリングロジックが英語キーで照合するため、値の説明文のみ日本語化しています。

[Fix-B1] core/benchmark.py — 動的 criteria 読み込み

STANDARD_ALLOWED_RUBRIC_CRITERIA のハードコードを除去し、STANDARD_RUBRIC_CRITERIA_FILE から動的に読み込むよう変更します。
日本語版でも英語キーが一致するため、スコアが正しく集計されます。
差分: data/benchmark.patch

[Fix-C1] core/conversation.py — 日本語セクションヘッダー対応

モデルが日本語でセクションヘッダーを返した場合のフォールバック正規表現を追加します。
英語ヘッダー (# My response 等) が第一優先で、master_prompt_ja.txt も英語見出しを使用しています。
差分: data/conversation.patch


翻訳方針

  • シナリオプロンプト: 文化的文脈を日本の高校生・大学生向けに自然に翻訳
  • プレースホルダー ({scenario_prompt} 等): 変更なし
  • 評価基準キー: 英語キーを維持(スコアリングロジックとの互換性のため)
  • 評価基準の説明文: 日本語に翻訳

動作確認環境

  • EQ-Bench3 HEAD (2025年5月時点)
  • Judge model: openai/gpt-oss-120b (novita プロバイダー経由)
  • Test model: YUGOROU/TeenEmo-LFM2.5-1.2B-DPO (LoRA adapter)
  • GPU: NVIDIA RTX 4090 on Vast.ai

ライセンス

翻訳データは EQ-Bench3 のライセンスに準拠します。
TeenEmo プロジェクト固有の追加コードは MIT ライセンスで提供します。

Downloads last month
78

Collection including YUGOROU/teememo-eq-bench-ja