CXM_Arena_French
收藏资源简介:
CXM Arena French Benchmark Suite是一个专为法语设计的全面数据集,用于评估AI在客户体验管理(CXM)领域的多种能力。该数据集包括五个核心任务:1. Agent Quality Adherence:监控法语对话中客服人员的表现。2. KB Refinement:处理业务文章,识别法语知识库条目中的相似和矛盾信息。3. Intent Prediction:基于不同的发现分类法,从客服对话中识别用户意图。4. Multi-Turn RAG with Tools:在多轮法语交互中评估会话AI系统的检索、生成和工具使用能力。
CXM Arena French Benchmark Suite is a comprehensive dataset tailored specifically for the French language, aimed at evaluating a wide range of AI capabilities in the Customer Experience Management (CXM) domain. This dataset includes five core tasks: 1. Agent Quality Adherence: Monitor the performance of customer support agents during French-language conversational interactions. 2. KB Refinement: Process business articles to identify similar and contradictory information within French-language knowledge base entries. 3. Intent Prediction: Identify user intent from customer service conversations based on multiple discovery taxonomies. 4. Multi-Turn RAG with Tools: Evaluate the retrieval, generation and tool-use capabilities of conversational AI systems during multi-turn French-language interactions.
CXM Arena French Benchmark Suite 数据集概述
数据集描述
- 目的:评估客户体验管理(CXM)领域中AI能力的综合性法语基准测试套件
- 语言:法语
- 特点:基于原始CXM_Arena基准建模,全部数据为法语生成
- 生成方式:使用先进大语言模型合成生成,包含重要品牌实体和现实法语场景
核心任务
- Agent Quality Adherence:通过模拟法语对话监控联络中心座席表现
- KB Refinement:处理商业文章,识别法语知识库中的相似/矛盾信息
- Intent Prediction:基于多种分类法识别联络中心对话中的用户意图
- Multi-Turn RAG with Tools:评估多轮法语交互中的检索、生成和工具使用能力
数据集详情
- 维护方:Spinklr AI
- 许可证:CC BY-NC-4.0
- 配置信息:
- KB_Refinement:
- contradictory_pairs:422个样本/5918字节
- similarity_pairs:569个样本/7448字节
- Articles:
- KB_refinement_articles:2223个样本/4181793字节
- multi_turn_articles:2435个样本/5883604字节
- Taxonomy:
- taxonomy_1:286个样本/32328字节
- Intent_Prediction:
- train:997个样本/1520270字节
- Agent_Quality_Adherence:
- train:2914个样本/3498804字节
- Multi_Turn:
- train:4142个样本/2223182字节
- Tool_Calling:
- train:1505个样本/2553514字节
- Tools_Description:
- train:150个样本/33520字节
- KB_Refinement:
评估结果
- 评估方法:随机采样100个数据点,使用intfloat/multilingual-e5-large嵌入和gemini-2.0-flash进行LLM推理
- 结果图表:
- Agent Quality Adherence
- KB Refinement
- Intent Prediction
- Multi-Turn RAG with Tools (包含2个图表)




