JP SupportDesk Chat Conversations

JP SupportDesk Chat Conversations is a curated dataset of Japanese-language customer support dialogues collected from SaaS, e-commerce, and subscription-style services. Each conversation is a sequence of messages with clear speaker labels (customer vs. agent), timestamps, and conversation IDs, making it easy to reconstruct full threads or sample individual turns.

For every ticket, the data includes intent labels (e.g., 支払い, アカウント, 初期設定, 不具合報告), resolution status, and escalation flags. In addition, utterances are annotated with politeness level (ため口 / 丁寧語 / ビジネス敬語) and tone (neutral, apologetic, frustrated), capturing nuances of Japanese customer communication that are often lost in generic datasets.

The dataset is designed for training Japanese dialogue models: support bots, answer drafting assistants for agents, and intent classifiers. Because conversations range from single-turn FAQs to long multi-step investigations that include confirmation, apologies, and proposals, the corpus is suitable for both instruction tuning and long-context reasoning.

All content is either synthetic or heavily de-identified; proper nouns and sensitive details are replaced with semantic placeholders while preserving sentence flow and realistic Japanese phrasing. Data is delivered as JSONL with one conversation per record, including metadata fields such as product line, channel, and difficulty level, so teams can segment subsets for domain-specific fine-tuning or evaluation.