研究文章
基于Coattention堆叠BiLSTM神经网络问答的机制
|
| Idx |
模型 |
地图 |
MRR |
|
| 1 |
概率准同步的语法(35] |
0.6029 |
0.6852 |
| 2 |
树编辑模型(2] |
0.6091 |
0.6917 |
| 3 |
直链CRF (17] |
0.6307 |
0.7477 |
| 4 |
LCLR [18] |
0.7092 |
0.7700 |
| 5 |
三元+数(38] |
0.7113 |
0.7846 |
| 6 |
三层BiLSTM + BM25 [6] |
0.7134 |
0.7913 |
| 7 |
卷积深层神经网络(39] |
0.7459 |
0.8078 |
| 8 |
BiLSTM / CNN与关注7] |
0.7111 |
0.8322 |
| 9 |
细心LSTM [1] |
0.7530 |
0.8300 |
| 10 |
BiLSTM encoder-decoder一步关注(8] |
0.7261 |
0.8018 |
| 11 |
BiLSTM |
0.6982 |
0.7764 |
| 12 |
堆叠BiLSTM |
0.7127 |
0.7893 |
| 13 |
与coattention BiLSTM |
0.7325 |
0.7962 |
| 14 |
堆叠和coattention BiLSTM |
0.7451 |
0.8114 |
| 15 |
堆叠BiLSTM coattention(余弦+欧几里得) |
0.7613 |
0.8401 |
|
|
