Table 2.
Performance statistics for three image queries
| No. | Query | Search target domain | Graph | Gel | Microscopy | Diagram | List | Misc. | Total | Relevant images | Precision |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | diet AND insulin | Caption | 17 | 0 | 0 | 0 | 0 | 2 | 19 | 19 | 100% |
| Caption+Image Text (HR) | 17 | 1 | 0 | 6 | 0 | 3 | 27 | 25 | 92.59% | ||
| Δ | 0 | 1 | 0 | 6 | 0 | 1 | 8 (42.11%) | 6 (31.58%) | – | ||
| 2 | apoptosis AND p53 | Caption | 11 | 1 | 5 | 11 | 0 | 14 | 42 | 42 | 100% |
| Caption+Image Text (HR) | 12 | 1 | 5 | 18 | 4 | 15 | 55 | 54 | 98.18% | ||
| Δ | 1 | 0 | 0 | 7 | 4 | 1 | 13 (30.95%) | 12 (28.57%) | – | ||
| 3 | miR* | Caption | 1 | 1 | 0 | 0 | 1 | 2 | 5 | 4 | 80% |
| AND brain AND heart | Caption+Image Text (HR) | 1 | 3 | 0 | 0 | 6 | 3 | 13 | 11 | 84.62% | |
| Δ | 0 | 2 | 0 | 0 | 5 | 1 | 8 (160%) | 7 (175%) | – |
The second column lists the actual queries entered into YIF; the third column specifies the search target domain, where ‘Image Text (HR)’ stands for ‘Image Text (High Recall)’. The row titled ‘Δ’ indicates the number of additional images found by using the ‘Caption+Image Text (HR)’ option versus the ‘Caption’ option, i.e. the additional images found by querying against image text. The fourth to the 10th columns show the number of images retrieved broken down according to image categories, as well as the total number of images found. The 11th column lists the number of retrieved and relevant images, as judged by a human expert, and the 12th column indicates overall image search precision. In the third query, the asterisk in miR* is useful for retrieving different types and spelling variations of miRNAs, such as mir-1 and mir-22.