10 datasets found

Universal sentence encoder

NLSC uses two pre-trained models, one uses 19.7B words from 700 dimensions trained on English tweets, and the other uses 1.7B words from 700 dimensions trained on Wikipedia...

Fine-grained analysis of sentence embeddings using auxiliary prediction tasks.

The dataset used in the paper is the Stanford Natural Language Inference (SNLI) dataset, which consists of 549,367 premise-hypothesis pairs for train/dev/test sets and target...

SICK is a dataset for recognizing textual entailment (RTE), containing 4.5K/0.5K/5.0K train/dev/test examples. Each example consists of a hypothesis and a premise, and the goal...

The SentEval dataset is a library for evaluating the quality of sentence embeddings.

The STS12, STS13, STS14, STS15, STS16, STSb, SICK-R datasets contain sentence pairs from various sources.

The SICK-Relatedness dataset contains 1,000 sentence pairs from the categories of captions, news, and forums.

The STS benchmark dataset contains 8,628 sentence pairs from the categories of captions, news, and forums.

PAUSE is a generic and end-to-end sentence embedding approach that exploits the labels and explores the unlabeled sentence pairs simultaneously.