💡 This paper reexamines unsupervised contrastive sentence embedding learning from a debiasing perspective, uses inverse propensity weighted sampling to eliminate biases like word frequency, sentence len