wasabiP/japanese-triplet-lifestyle-romance

🏯 Japanese Preference Dataset: Counseling & Advice (Free Sample)


This repository provides a free sample of a Japanese preference learning dataset designed for Direct Preference Optimization (DPO), RLHF, Reward Modeling, response ranking, and Japanese LLM alignment.

The dataset focuses on realistic Japanese counseling and advice scenarios, helping language models learn not only factual correctness but also empathy, contextual understanding, and practical response quality.

🚀 Full Version

A larger commercial edition containing approximately 1,000 curated preference samples is available on Gumroad.

👉 https://wasabicatalog.gumroad.com/l/ioojr

Included in the Full Version

• —Approximately 1,000 Japanese preference samples
• —JSON format
• —Four-part preference structure:
• —anchor
• —positive
• —hard_negative
• —negative
• —Rich metadata:
• —topic
• —emotion
• —intent
• —situation
• —response_style
• —urgency
• —difficulty
• —Commercial use permitted
• —AI training & fine-tuning permitted

📌 Dataset Features

Preference Learning Oriented

Each sample contains multiple responses with different quality levels, making the dataset suitable for:

• —Direct Preference Optimization (DPO)
• —Reward Model (RM) training
• —RLHF research
• —Response ranking
• —Japanese LLM alignment

Newly Generated Dataset

The dataset consists of newly generated counseling and advice scenarios created from abstract consultation themes rather than copied consultation texts.

Challenging Hard Negatives

Rather than simple word substitutions, the hard_negative responses intentionally represent subtle but realistic mistakes, such as:

• —insufficient empathy
• —incomplete contextual understanding
• —weak practical usefulness
• —slightly inappropriate advice

This enables models to learn fine-grained preference differences.

Natural Japanese Conversations

The dataset reflects natural Japanese expressions commonly found in counseling and advice situations, including realistic emotional nuance and conversational tone.

Quality Assurance

The dataset was reviewed through:

• —Automated validation (JSON validation, duplicate detection, keyword analysis, sentence-ending analysis)
• —Human spot review of selected samples for naturalness and logical consistency

📊 Sample Format

[
  {
    "metadata": {
      "topic": "人間関係・仕事",
      "emotion": "不安・焦り",
      "intent": "共感と具体的なアドバイス",
      "situation": "職場でのコミュニケーション不足に悩んでいる",
      "response_style": "共感・建設的",
      "urgency": "medium",
      "difficulty": "medium"
    },
    "anchor": "最近職場でのコミュニケーションがうまくいかず、自分の意見を伝えるのが怖くなってしまいました。どうすれば自信を持てるでしょうか？",
    "positive": "お辛い状況ですね。まずは一気に解決しようとせず、小さな挨拶やうなずきから始めてみましょう。自分の意見を紙に書き出してから話すのも効果的ですよ。",
    "hard_negative": "コミュニケーション不足は準備不足が原因です。もっと自信を持って発言しないと、周囲からの評価も下がってしまいますよ。",
    "negative": "職場を変えるのが一番の解決策です。転職サイトに登録してみることをおすすめします。"
  }
]