cngchis/Support-Ticket-Router-12K-Cleaned

--- license: apache-2.0 task_categories: - text-classification - text-generation language: - en tags: - synthetic - customer-support - saas - intent-classification pretty_name: Support Ticket Router (12K Cleaned) size_categories: - 10K<n<100K --- # 🔥 Support-Ticket-Router-12K-Cleaned This dataset is a cleaned and structured version of real-world-like customer support messages designed for intent classification and routing tasks in SaaS / IT support systems. It is intended for training and evaluating LLM-based or classical NLP intent classifiers for automated customer support ticket routing. --- ## 🧪 Data Source This dataset is **synthetically generated using GPT-4-class models (GPT-4 / GPT-4o-style prompting)** with additional rule-based filtering and normalization. The data was created to simulate realistic customer support scenarios in SaaS / IT service environments, including: - API integration issues - Billing and payment problems - Subscription cancellation requests - Technical bugs and system failures - General customer complaints - Plan upgrade/downgrade requests --- ### 🧠 Generation Process - Prompts were designed to mimic real customer support tickets - Multiple variations of each intent were generated for diversity - Responses were normalized into 6 fixed intent classes - Noise and ambiguous samples were filtered out - Final dataset was manually and programmatically cleaned for consistency --- ### ⚠️ Notes - This dataset is **synthetic and not collected from real users** - It is intended for **research, benchmarking, and model training only** - No real customer or personal data is included --- ## 🧠 Summary - Total records: ~12,000+ - Task: Single-label Intent Classification - Domain: Customer Support AI (SaaS / IT Services) - Labels: - api - billing - cancellation - complaint - technical - upgrade --- ## 📌 Dataset Structure Each example follows a unified format: ```json { "text": "user message", "label": "intent label", "meta": { "source": "synthetic + real-world inspired", "domain": "SaaS customer support" } } ``` Or instruction-style format (for LLM fine-tuning): ```json { "input": "Classify the customer support message into one of the following intents: api, billing, cancellation, complaint, technical, upgrade.\n\nMessage: I want to cancel my subscription.", "output": "cancellation" } ``` --- ## 🧹 Cleaning Process This dataset has been carefully processed to improve label quality and consistency: - Removed ambiguous or noisy samples - Normalized intent categories into 6 standard labels - Filtered duplicate or near-duplicate entries - Standardized user message formatting - Balanced distribution across intent classes --- ## 📊 Dataset Statistics | Label | Description | |--------------|-------------| | api | API usage, integration, endpoint issues | | billing | Payment, invoice, pricing issues | | cancellation | Stop subscription, churn intent | | complaint | Dissatisfaction without clear category | | technical | Bugs, errors, system issues | | upgrade | Plan change, feature upgrade | --- ## ⚙️ Usage ```json from datasets import load_dataset dataset = load_dataset("cngchis/Support-Ticket-Router-12K-Cleaned") train = dataset["train"] test = dataset["test"] ``` --- ## 🚀 Use Cases This dataset can be used for: - Intent classification models - LLM fine-tuning (SFT / instruction tuning) - Customer support automation systems - Ticket routing systems in SaaS platforms - Benchmarking lightweight LLMs (GGUF / 4-bit models) --- ## 📚 Model Compatibility Works well with: - BERT / RoBERTa / DeBERTa - LLaMA / Mistral / Phi models - GGUF (llama.cpp) - Instruction-tuned LLMs --- ## 🧾 Citation If you use this dataset, please cite: ```json @misc{support_ticket_router_12k, title={Support-Ticket-Router-12K-Cleaned}, author={cngchis}, year={2026}, publisher={Hugging Face}, url={https://huggingface.co/datasets/cngchis/Support-Ticket-Router-12K-Cleaned} } ```--- 许可证：Apache-2.0 任务类别： - 文本分类 - 文本生成 语言： - 英语 标签： - 合成数据 - 客户支持 - 软件即服务（SaaS） - 意图分类 美观名称：清理版12K客户工单路由数据集（Support Ticket Router (12K Cleaned)） 样本量区间： - 10K<n<100K --- 🔥 清理版12K客户工单路由数据集（Support-Ticket-Router-12K-Cleaned） 本数据集为贴合真实场景的客户支持留言经清理与结构化处理后的版本，专为软件即服务（SaaS）/信息技术（IT）支持系统中的意图分类与工单路由任务设计。 本数据集可用于训练与评估基于大语言模型（Large Language Model, LLM）或经典自然语言处理（Natural Language Processing, NLP）的意图分类器，以实现自动化客户工单路由。 ## 🧪 数据源 本数据集通过GPT-4级模型（GPT-4 / GPT-4o风格提示）合成生成，并辅以基于规则的过滤与归一化处理。 本数据集旨在模拟SaaS/IT服务环境中的真实客户支持场景，涵盖以下类型： - API集成问题 - 账单与支付问题 - 订阅取消请求 - 技术漏洞与系统故障 - 一般性客户投诉 - 套餐升降级请求 ### 🧠 生成流程 - 提示词设计旨在模拟真实客户工单 - 为提升多样性，每种意图均生成多种变体 - 将生成的回复归一化为6个固定意图类别 - 过滤掉含噪声与歧义的样本 - 最终数据集通过人工与自动化手段完成清理，以保证一致性 ### ⚠️ 注意事项 - 本数据集为合成生成，并非来自真实用户 - 本数据集仅用于研究、模型基准测试与模型训练 - 数据集未包含任何真实客户信息或个人隐私数据 ## 🧠 数据集概览 - 总样本量：约12000条以上 - 任务类型：单标签意图分类 - 应用领域：客户支持人工智能（SaaS/IT服务） - 标签列表： - api - billing - cancellation - complaint - technical - upgrade ## 📌 数据集结构 每条样本均遵循统一格式： json { "text": "用户留言内容", "label": "意图标签", "meta": { "source": "合成生成 + 贴合真实场景", "domain": "SaaS客户支持" } } 或采用指令微调格式（适用于大语言模型微调）： json { "input": "请将以下客户支持留言分类至以下意图之一：api、billing、cancellation、complaint、technical、upgrade。 留言内容：我想要取消我的订阅。", "output": "cancellation" } ## 🧹 数据清理流程 本数据集经过严格处理，以提升标签质量与一致性： - 移除歧义与含噪声的样本 - 将意图类别归一化为6个标准标签 - 过滤重复或近似重复的条目 - 标准化用户留言的格式 - 平衡各意图类别的样本分布 ## 📊 数据集统计信息 | 标签 | 标签说明 | |--------------|-------------| | api | API使用、集成与接口问题 | | billing | 支付、发票与定价问题 | | cancellation | 停止订阅、客户流失相关意图 | | complaint | 无明确类别的客户不满投诉 | | technical | 软件漏洞、运行错误与系统故障 | | upgrade | 套餐变更与功能升级请求 | ## ⚙️ 使用方法 本数据集的加载示例如下： python from datasets import load_dataset dataset = load_dataset("cngchis/Support-Ticket-Router-12K-Cleaned") train = dataset["train"] test = dataset["test"] ## 🚀 应用场景 本数据集可应用于以下场景： - 意图分类模型开发 - 大语言模型微调（监督微调/指令微调） - 客户支持自动化系统搭建 - SaaS平台中的工单路由系统开发 - 轻量级大语言模型基准测试（如GGUF格式、4-bit量化模型） ## 📚 模型兼容性 本数据集兼容以下模型类型： - BERT / RoBERTa / DeBERTa - LLaMA / Mistral / Phi系列模型 - GGUF格式模型（基于llama.cpp） - 指令微调大语言模型 ## 🧾 引用规范 若您使用本数据集，请引用如下文献： bibtex @misc{support_ticket_router_12k, title={Support-Ticket-Router-12K-Cleaned}, author={cngchis}, year={2026}, publisher={Hugging Face}, url={https://huggingface.co/datasets/cngchis/Support-Ticket-Router-12K-Cleaned} }