Table 4.
The average precision, ranking loss and coverage of model BLSTM, BLSTM + ConvNet1, ConvNet2 and BLSTM + ConvNet1 + ConvNet2
| D3106 | D4802 | |||||
|---|---|---|---|---|---|---|
| RL | Cov | AP | RL | Cov | AP | |
| BLSTM | 0.0967 | 1.5895 | 0.7523 | 0.0820 | 3.3916 | 0.6901 |
| BLSTM + ConvNet1 | 0.0778 | 1.3113 | 0.7876 | 0.0603 | 2.9225 | 0.7453 |
| ConvNet2 | 0.1294 | 2.0113 | 0.6430 | 0.0673 | 3.2868 | 0.6214 |
| BLSTM + ConvNet1 + ConvNet2 | 0.0758 | 1.2848 | 0.7901 | 0.0637 | 3.0528 | 0.7414 |
On dataset D3106, the BLSTM + ConvNet1 + ConvNet2 has the best performance with lowest ranking loss, coverage and highest average precision which are 0.0758, 1.2848 and 0.7901, respectively. However, the model BLSTM + ConvNet1 + ConvNet2 is not as good as model BLSTM + ConvNet1 when tested on dataset D4802. The best values are marked out with bold text