OpenVoiceOS/ovos-intent-bench-intents-for-eval
收藏资源简介:
该数据集名为OVOS intent bench — intents-for-eval,是一个用于意图分类基准测试的数据集,包含来自OpenVoiceOS(OVOS)插件竞技场中开放意图联盟(混合范式管道融合)参赛者对intents-for-eval数据集的每个样本预测结果。数据集按语言分割,涵盖多种语言(如加泰罗尼亚语、丹麦语、德语、英语、西班牙语、巴斯克语、法语、加利西亚语、意大利语、荷兰语、巴西葡萄牙语和欧洲葡萄牙语),每个语言下有一个JSONL文件,包含行数据遵循特定的合约格式(如数据集版本、插件版本、管道阶段、精确匹配等)。这些预测结果用于生成基准板、盲战池和基于基准的ELO阶梯,旨在评估和比较不同意图分类模型的性能。数据集由NGI0 Commons Fund通过欧盟的下一代互联网计划资助。
This dataset, named OVOS Intent Bench — intents-for-eval, is a benchmark dataset for intent classification. It contains the prediction results for each sample of the intents-for-eval dataset submitted by participants of the Open Intent Alliance (Hybrid Paradigm Pipeline Fusion) in the OpenVoiceOS (OVOS) Plugin Arena. The dataset is split by language, covering a wide range of languages including Catalan, Danish, German, English, Spanish, Basque, French, Galician, Italian, Dutch, Brazilian Portuguese, and European Portuguese. Each language corresponds to a JSONL file, where each line of data follows a specified contract format including fields such as dataset version, plugin version, pipeline stage, exact match, and so on. These prediction results are utilized to generate benchmark leaderboards, blind battle pools, and benchmark-based ELO rankings, with the goal of evaluating and comparing the performance of various intent classification models. This dataset is funded by the NGI0 Commons Fund through the European Union's Next Generation Internet initiative.




