Sprinklr/Agent_Quality_Adherence

Agent Quality Adherence数据集是为了测试业务环境中呼叫中心代理监控能力而设计的。它包括模拟的呼叫中心对话，每个对话都有不同的测试指标。这个数据集是使用Gemini-2.0-flash按照精心设计的方案合成生成的，包含了重要的品牌实体。数据集包含1987个案例/对话，每个案例平均有9个可用指标，每个LLM回合最多有4个指标分割。63%的问题答案是是，37%的问题是否。数据集适用于评估呼叫中心业务领域的代理监控系统，以及基准测试结合对话理解、段落理解和文本分类的管道。不适用于没有人工监督的高风险应用或业务领域之外的场景，也不适用于需要多模态或非英语输入的任务。

The Agent Quality Adherence dataset is designed to test the monitoring capacity of Contact Center Agents in a business context. It consists of simulated contact center conversations with different metrics to be tested per conversation. The dataset was synthetically generated using Gemini-2.0-flash following a carefully planned approach to include important Brand entities. It contains 1987 cases/dialogues, with an average of 9 available metrics per case and a maximum of 4 metric splits per LLM turn. 63% of the questions have yes as the answer, and 37% have no. The dataset is intended for evaluating Agent monitoring systems in the contact center business domain and benchmarking pipelines that combine conversation understanding, passage comprehension, and text classification. It is not suitable for high-stakes applications without human oversight, domains beyond business, or tasks requiring multi-modal or non-English inputs.