cooldad7777/Bitext-customer-support-llm-chatbot-training-dataset

Bitext - 基于大型语言模型的虚拟助手客户服务标记训练数据集是一个混合合成数据集，专为微调GPT、Mistral和OpenELM等大型语言模型而设计。该数据集采用NLP/NLG技术和自动化数据标注工具生成，旨在展示如何通过两步微调方法实现客户支持领域的垂直化/领域适应。例如，企业可先使用此数据集训练微调模型，再结合少量自有数据进一步微调，以创建定制化LLM。数据集规格包括：用例为意图检测，垂直领域为客服，包含27个意图（分为10个类别）、26872个问答对（每个意图约1000对）、30个实体/槽位类型以及12种语言生成标签。类别和意图选自Bitext的20个垂直特定数据集，覆盖了这些垂直领域中的通用意图，包括汽车、零售银行、教育、医疗保健、零售/电子商务等。数据生成采用混合方法，以自然文本为源，通过NLP技术提取种子文本，再使用NLG技术扩展，整个过程由计算语言学家策划。数据集包含instruction和response等字段，总令牌数达357万，适用于训练AI对话、生成和问答模型。

Bitext - Customer Service Tagged Training Dataset for LLM-based Virtual Assistants is a hybrid synthetic dataset designed to fine-tune Large Language Models such as GPT, Mistral, and OpenELM, generated using NLP/NLG technology and automated Data Labeling tools. It demonstrates verticalization/domain adaptation for the Customer Support sector via a two-step LLM fine-tuning approach, where companies can first train a model with this dataset and further fine-tune it with their own data. The dataset specs include: Use Case - Intent Detection, Vertical - Customer Service, 27 intents assigned to 10 categories, 26872 question/answer pairs (approximately 1000 per intent), 30 entity/slot types, and 12 language generation tags. Categories and intents are selected from Bitexts collection of 20 vertical-specific datasets, covering common intents across verticals like Automotive, Retail Banking, Education, Healthcare, Retail/E-commerce, etc. The question/answer pairs are generated through a hybrid methodology using natural texts as sources, NLP to extract seeds, and NLG for expansion, all curated by computational linguists. Fields include instruction, response, category, intent, and flags, with a total token count of 3.57 million, suitable for training AI Conversational, Generative, and Q&A models.