Figure 4.

Comparisons. The second benchmark was reproduced using two other scoring schemes (Okapi BM25 and PMRA) for comparison with MedlineRanker (Bayes). Results were produced training with only the document of interest (S, the baseline) or training with all references from the full text (T). An alternative query expansion method based on pseudo relevance feedback using the MedlineRanker scoring scheme was also compared (Bayes-PRF). ROC: receiver operating characteristic.

Ortuño et al. BMC Bioinformatics 2013 14:113   doi:10.1186/1471-2105-14-113
Download authors' original image