Universal Sentence Encoder

29 Mar 2018Daniel CerYinfei YangSheng-yi KongNan HuaNicole LimtiacoRhomni St. JohnNoah ConstantMario Guajardo-CespedesSteve YuanChris TarYun-Hsuan SungBrian StropeRay Kurzweil

We present models for encoding sentences into embedding vectors that specifically target transfer learning to other NLP tasks. The models are efficient and result in accurate performance on diverse transfer tasks... (read more)

PDF Abstract

Evaluation results from the paper

Task Dataset Model Metric name Metric value Global rank Compare
Sentiment Analysis CR USE_T+CNN (w2v w.e.) Accuracy 87.45 # 2
Sentiment Analysis MPQA USE_T+DAN (w2v w.e.) Accuracy 88.14 # 1
Sentiment Analysis MR USE_T+CNN Accuracy 81.59 # 5
Sentiment Analysis SST-2 Binary classification USE_T+CNN (lrn w.e.) Accuracy 87.21 # 16
Semantic Textual Similarity STS Benchmark USE_T Pearson Correlation 0.782 # 1
Subjectivity Analysis SUBJ USE Accuracy 93.90 # 4
Text Classification TREC-6 USE_T+CNN Error 1.93 # 1