遇见数据集

Jianshu001/arabic-conversation-batch1

收藏
Hugging Face2026-04-09 更新2026-04-12 收录
官方服务:

资源简介:

--- language: - ar task_categories: - text-generation tags: - arabic - synthetic - multi-turn size_categories: - 1K<n<10K --- # Arabic Multi-Domain Conversations — Batch 1 1,635 synthetic multi-turn Arabic conversations across 6 UAE/Middle East domains. Generated with **gpt-5.4-mini** (user + assistant), factuality checked with **o3**. ## Domains | Domain | Count | |--------|-------| | Islamic Finance | 278 | | Healthcare | 277 | | Real Estate | 277 | | Energy | 270 | | Education | 268 | | Government Services | 265 | ## Stats - 1,635 conversations, 6,554 user messages, 6,554 assistant messages - Average user message: 150 chars - Markdown headings in assistant: 0% - Factuality: 93% pass, 6% review, 0% unchecked - 3-5 turns per conversation - Runtime: 5.2 hours at 312 conversations/hour ## Generation Config - User/Assistant model: gpt-5.4-mini - Factuality check: o3 - Quality check: truncation detection - Dedup: persona_id + topic + subtopic (no repeats) ## Format JSONL: id, domain, domain_ar, topic, topic_ar, subtopic_ar, persona, conversation, metadata, factuality

提供机构:
Jianshu001
二维码
社区交流群
二维码
科研交流群
商业服务