sprinklr-huggingface/CXM_Arena

CXM Arena Benchmark Suite 是一个用于评估人工智能在客户体验管理 (CXM) 领域能力的综合数据集。它涵盖了五个核心任务：代理质量遵守、KB 精炼、文章搜索和理解、意图预测以及多轮 RAG 与工具。该数据集是使用高级语言模型（如 Gemini-2.0-flash 和 GPT-4o）合成的，旨在模拟现实业务场景。数据集包含多个配置，每个配置对应一个任务或相关资源。数据集的生成过程注重了重要品牌实体的选择和整合。对于涉及知识库的任务，数据集提供了并行“干净”和“嘈杂”的内容，以测试系统的鲁棒性。多轮 RAG 组件的工具调用子集基于对 456 个生成对话的分析。数据集还包括了各种任务的评估结果和基准，以及如何使用数据集的说明。

The CXM Arena Benchmark Suite is a comprehensive dataset designed to evaluate AI capabilities within the Customer Experience Management (CXM) domain. It includes five core tasks: Agent Quality Adherence, KB Refinement, Articles Search and Comprehension, Intent Prediction, and Multi-Turn RAG with Tools. The dataset is synthetically generated using advanced language models, primarily Gemini-2.0-flash and GPT-4o, to simulate realistic business scenarios. It comprises multiple configurations, each corresponding to a task or associated resource. The datasets generation process focuses on the careful selection and incorporation of important brand entities. For tasks involving knowledge bases, it provides parallel clean and noisy content to test system robustness. The tool calling subset of the Multi-Turn RAG component is based on an analysis of 456 generated conversations where tool selection was relevant. The dataset also includes evaluation results and benchmarks for each task, along with instructions on how to use the dataset.