Paraphrase Detection on MSRP
80.4AccuracyJi and Eisenstein (2013)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Ji and Eisenstein (2013)features=hand-crafted, model=SVM2016.02 | 80.4 | 86 | |
| Yin et al. (2015)sparse features=with2016.02 | 78.9 | 84.8 | |
| He et al. (2015)POS and Para. embeddings=with2016.02 | 78.6 | 84.7 | |
| Yin and Schütze (2015)pretraining=with2016.02 | 78.4 | 84.6 | |
| Lexical Decomposition and Composition Modelsparse features=without, extra annotated resources=without2016.02 | 78.4 | 84.7 | |
| He et al. (2015)POS embeddings=without2016.02 | 77.8 | — | |
| He et al. (2015)Para. embeddings=without2016.02 | 77.3 | — | |
| SDAEData Source Type=Unordered Sentences, Training Corpus=Toronto Books (70m sents, 0.9B words)2016.02 | 76.4 | 83.4 | |
| SAEData Source Type=Unordered Sentences, Training Corpus=Toronto Books (70m sents, 0.9B words)2016.02 | 74.3 | 81.7 | |
| SDAE+embs.Data Source Type=Unordered Sentences, Training Corpus=Toronto Books (70m sents, 0.9B words)2016.02 | 73.7 | 80.7 | |
| DictRep BOWData Source Type=Other structured data resource2016.02 | 73.7 | 81.6 | |
| Paragraph Vec DMData Source Type=Unordered Sentences, Training Corpus=Toronto Books (70m sents, 0.9B words)2016.02 | 73.6 | 81.9 | |
| Unigram TFIDFData Source Type=Unordered Sentences, Training Corpus=Toronto Books (70m sents, 0.9B words)2016.02 | 73.6 | 81.7 | |
| CaptionRep BOWData Source Type=Other structured data resource2016.02 | 73.6 | 81.9 | |
| DictRep RNNData Source Type=Other structured data resource2016.02 | 73.2 | 81.6 | |
| SkipThoughtData Source Type=Ordered Sentences, Training Corpus=Toronto Books2016.02 | 73 | 82 | |
| Paragraph Vec DBOWData Source Type=Unordered Sentences, Training Corpus=Toronto Books (70m sents, 0.9B words)2016.02 | 72.9 | 81.1 | |
| CaptionRep RNNData Source Type=Other structured data resource2016.02 | 72.6 | 81.1 | |
| Yin and Schütze (2015)pretraining=without2016.02 | 72.5 | 81.4 | |
| FastSentData Source Type=Ordered Sentences, Training Corpus=Toronto Books2016.02 | 72.2 | 80.3 | |
| CPHRASEData Source Type=2.8B words2016.02 | 72.2 | 79.6 | |
| SWEM-concat2018.05 | 71.5 | 81.3 | |
| FastSent+AEData Source Type=Ordered Sentences, Training Corpus=Toronto Books2016.02 | 71.2 | 79.1 | |
| SWEM-aver2018.05 | 71 | 81.1 | |
| SAE+embs.Data Source Type=Unordered Sentences, Training Corpus=Toronto Books (70m sents, 0.9B words)2016.02 | 70.6 | 77.9 | |
| LSTM2018.05 | 70.6 | 80.5 | |
| SWEM-max2018.05 | 70.6 | 80.8 | |
| CNN2018.05 | 69.9 | 80.9 | |
| SkipgramData Source Type=Unordered Sentences, Training Corpus=Toronto Books (70m sents, 0.9B words)2016.02 | 69.3 | 77.2 | |
| NMT En to FrData Source Type=Other structured data resource2016.02 | 69.1 | 77.1 | |
| DictRep BOW+embs.Data Source Type=Other structured data resource2016.02 | 68.4 | 76.8 | |
| CBOWData Source Type=Unordered Sentences, Training Corpus=Toronto Books (70m sents, 0.9B words)2016.02 | 67.6 | 76.1 | |
| DictRep RNN+embs.Data Source Type=Other structured data resource2016.02 | 66.8 | 76 | |
| NMT En to DeData Source Type=Other structured data resource2016.02 | 65.2 | 73.3 |