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 | 0.650 | 0.540 | 0.453 | 0.395 | 0.342 | 0.302 | 0.193 | 0.320 | 0.404 | 0.469 | 0.507 | 0.537 |
Levenshtein with text | 0.580 | 0.495 | 0.407 | 0.340 | 0.298 | 0.263 | 0.172 | 0.294 | 0.362 | 0.404 | 0.442 | 0.469 |
Embedding of text | 0.800 | 0.665 | 0.560 | 0.495 | 0.436 | 0.388 | 0.237 | 0.395 | 0.499 | 0.588 | 0.647 | 0.691 |
0.5 * Image + 0.5 * Text embedding †| 0.88 | 0.755 | 0.633 | 0.568 | 0.510 | 0.460 | 0.261 | 0.448 | 0.564 | 0.674 | 0.757 | 0.819 |
Improvement* | 35% | 40% | 40% | 44% | 49% | 52% | 35% | 40% | 40% | 44% | 49% | 52% |