Table 4

Performance of CombMNZ

Components

w/ Normalization

w/ Assigned Weights

w/ Multiple

document

aspect

passage2

document

aspect

passage2

document

aspect

passage2


Best of baselines

0.2906

0.2189

0.0988

0.2906

0.2189

0.0988

0.2906

0.2189

0.0988


UniNE1+York07ga2+kyoto1

0.2671 (-8.08%)

0.1535 (-29.86%)

0.0937 (-5.13%)

0.2729 (-6.09%)

0.1854 (-15.27%)

0.0957 (-3.19%)

0.2571 (-11.53%)

0.1547 ( -29.33% )

0.0924 ( -6.49%)

UniNE1+York07ga2+UBexp1

0.2656 (-8.61%)

0.1772 (-19.03%)

0.0879 (-10.99%)

0.2591 (-10.82%)

0.1878 (-14.18%)

0.0867 (-12.30%)

0.2639 (-9.16%)

0.1753 (-19.92%)

0.0885 (-10.43%)

UniNE1+MuMshFd+kyoto1

0.2559 (-11.95%)

0.1801 (-17.70%)

0.0985 (-0.30%)

0.2503 (-13.85%)

0.1837 (-16.09%)

0.0908 (-8.06%)

0.2401 (-17.38%)

0.1599 (-26.96%)

0.0958 (-3.04%)

UniNE1+MuMshFd+UBexp1

0.2416 (-16.85%)

0.1720 (-21.43%)

0.0871 (-11.86%)

0.2466 (-15.11%)

0.1787 (-18.36%)

0.0839 (-15.09%)

0.2419 (-16.74% )

0.1716 (-21.61% )

0.0872 (-11.72%)


In order to deeply evaluate the benefits of CombMNZ, we generate CombMNZ-with-normalization, CombMNZ-with-assigned-weight and CombMNZ-with-multiple respectively. For CombMNZ-with-normalization, we employ the standard zero-one normalization method in which all the base weights are scaled between zero being the lowest value and one being the absolute highest value. For CombMNZ-with-assigned-weight, the baselines earn their weights depending on their models. Only the optimal results are presented. For CombMNZ-with-multiple, we apply the CombMNZ method for multiple times (m times, where m is set to be one of {1, 2, 3, 5}). No normalization and no additional weights has been given to the baselines. Only the optimal results are presented as well. The values in the parentheses are the relative rates of improvement over the best results of the baselines. Note that “w/” stands for “with”.

Hu et al. BMC Bioinformatics 2011 12(Suppl 5):S6   doi:10.1186/1471-2105-12-S5-S6

Open Data