From: DLI-IT: a deep learning approach to drug label identification through image and text embedding
 | P @k | R @k | ||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
Methods | k = 1 | k = 2 | k = 3 | k = 4 | k = 5 | k = 6 | k = 1 | k = 2 | k = 3 | k = 4 | k = 5 | k = 6 |
Image-based | 0.630 | 0.535 | 0.470 | 0.413 | 0.346 | 0.297 | 0.152 | 0.258 | 0.340 | 0.398 | 0.417 | 0.429 |
Levenshtein with text | 0.480 | 0.550 | 0.480 | 0.425 | 0.376 | 0.333 | 0.116 | 0.265 | 0.347 | 0.410 | 0.453 | 0.482 |
Embedding of text | 0.800 | 0.720 | 0.640 | 0.565 | 0.478 | 0.405 | 0.193 | 0.347 | 0.463 | 0.545 | 0.576 | 0.586 |
0.5 * Image + 0.5 * Text embedding †| 0.800 | 0.725 | 0.640 | 0.570 | 0.480 | 0.410 | 0.193 | 0.349 | 0.463 | 0.549 | 0.578 | 0.593 |
Improvement* | 27% | 32% | 33% | 34% | 28% | 23% | 27% | 32% | 33% | 34% | 28% | 23% |