通过人工智能反馈提高对比学习句子嵌入（英文版）

英文标题：Improving Contrastive Learning of Sentence Embeddings from AI Feedback中文摘要：CLAIF 方法：利用大型预训练语言模型的人工智能反馈来构建有细粒度样本相似度得分的样本对以改善对比学习，并结合人工反馈和 AI 反馈提供更好的训练信号，实验结果显示在语义文本相似性和迁移学习任务上，与其他半监督和监督对比学习方法比较，CLAIF 实现了最先进的性能。英文摘要：Contrastive learning has become a popular approach in natural languageprocessing, particularly for the learning of sentence embeddings. However, thediscrete nature of natural language makes it difficult to ensure the quality ofpositive and negative sample pairs generated through d