Research Article
Learning Document Semantic Representation with Hybrid Deep Belief Network
Table 2
Document classification accuracy and document retrieval precision of six categories on 20 Newsgroups.
| Dataset | Model | L_r | Document classification | Document retrieval | Output units number | Output units number | 50 | 100 | 128 | 512 | 1000 | 50 | 100 | 128 | 512 | 1000 |
| 20 Newsgroups | RSM | 0.01 | 76.13 | 79.60 | 80.63 | 80.24 | 80.61 | 75.40 | 79.83 | 80.24 | 78.64 | 79.66 | 0.001 | 74.11 | 76.20 | 81.88 | 81.03 | 81.34 | 68.23 | 70.43 | 80.89 | 81.08 | 81.07 | 0.0001 | 57.88 | 76.89 | 76.92 | 83.34 | 82.93 | 58.66 | 71.98 | 72.64 | 80.45 | 78.93 | DBN | 0.01 | 76.52 | 79.77 | 81.03 | 82.94 | 82.11 | 77.36 | 80.21 | 80.07 | 80.99 | 79.01 | DocNADE | 0.01 | 78.82 | 81.37 | 82.70 | 83.67 | 83.76 | 78.13 | 81.01 | 81.46 | 81.81 | 81.15 | HDBN | 0.01 | 85.92 | 85.77 | 86.82 | 86.94 | 86.15 | 82.32 | 82.42 | 84.97 | 84.87 | 83.73 |
|
|