Cross-lingual Document Classification on MLDoc (test)
89.4Accuracy (EN->FR)MultiFit
Evaluation results
| Method | Links | |||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MultiFitMethod category=global fine-tuning style, Evaluation Scenario=zero-shot, Runs=mean value of 5 runs2021.05 | 89.4 | — | — | — | — | — | — | — | — | — | 91.6 | — | 79.1 | — | 76 | — | — | |
| XTR (Ours)Representation granularity=fixed-dimensional sentence, Model type=lightweight transformer, Evaluation Scenario=zero-shot, Runs=mean value of 5 runs2021.05 | 85.1 | — | — | — | — | — | — | — | — | 82.4 | 88.8 | 80.8 | 80.8 | 79.2 | 74.3 | 79.9 | 81.4 | |
| mBERTMethod category=global fine-tuning style, Evaluation Scenario=zero-shot, Runs=mean value of 5 runs2021.05 | 83 | — | — | — | — | — | — | — | — | — | 82.4 | — | 75 | — | 68.3 | — | — | |
| Bi-Sent2VecRepresentation granularity=fixed-dimensional word, Evaluation Scenario=zero-shot, Runs=mean value of 5 runs2021.05 | 81.6 | — | — | — | — | — | — | — | — | 82.2 | 86.5 | 79.2 | 74 | 71.5 | 75 | 72.6 | 77.8 | |
| Yu et al. (2018)Representation granularity=fixed-dimensional sentence, Evaluation Scenario=zero-shot, Runs=mean value of 5 runs2021.05 | 80.8 | — | — | — | — | — | — | — | — | 81 | 80.2 | 77.1 | 74.1 | 74.1 | 70.8 | 74.8 | 76.6 | |
| LASERRepresentation granularity=fixed-dimensional sentence, Evaluation Scenario=zero-shot, Runs=mean value of 5 runs2021.05 | 78 | — | — | — | — | — | — | — | — | 80.1 | 86.3 | 80.8 | 79.3 | 69.6 | 70.2 | 74.2 | 77.3 | |
| MultiCCA + CNNRepresentation granularity=fixed-dimensional word, Evaluation Scenario=zero-shot, Runs=mean value of 5 runs2021.05 | 72.4 | — | — | — | — | — | — | — | — | 64.8 | 81.2 | 56 | 72.5 | 74 | 69.4 | 53.7 | 68 | |
| T-LASERRepresentation granularity=fixed-dimensional sentence, Evaluation Scenario=zero-shot, Runs=mean value of 5 runs2021.05 | 70.7 | — | — | — | — | — | — | — | — | 78.2 | 86.8 | 79 | 71.4 | 74.5 | 68.7 | 76 | 75.7 | |
| BiLSTM (Europarl)transfer_mode=zero-shot, training_corpus=Europarl2018.12 | — | 88.4 | 71.83 | 66.65 | 72.83 | 60.73 | — | — | — | — | — | — | — | — | — | — | — | |
| BiLSTM (UN)transfer_mode=zero-shot, training_corpus=UN2018.12 | — | 88.83 | — | 69.5 | 74.52 | — | — | 61.42 | 71.97 | — | — | — | — | — | — | — | — | |
| LASERtransfer_mode=zero-shot, classifier=FFNN (one hidden layer, 10 units)2018.12 | — | 89.93 | 84.78 | 77.33 | 77.95 | 69.43 | 60.3 | 67.78 | 71.93 | — | — | — | — | — | — | — | — | |
| MultiCCA + CNNtransfer_mode=zero-shot2018.12 | — | 92.2 | 81.2 | 72.5 | 72.38 | 69.38 | 67.63 | 60.8 | 74.73 | — | — | — | — | — | — | — | — |