Table 11 Kruskal–Wallis H-tests on the generalization power of algorithm types “matrix factorization” (MF), “neural networks” (NN) and “graph-based” (GB) across datasets. The significance level is set to \(1\%\), and p-values are adjusted for multiple tests with the Benjamini-Hochberg method63. All tests are statistically significant.
From: Comprehensive evaluation of pure and hybrid collaborative filtering in drug repurposing
Type | GB | MF | NN |
|---|---|---|---|
H | 308.5 | 1100.2 | 21.4 |
adjusted p | 0.0 | 0.0 | \(4.10^{-6}\) |
\(\mu _{t,Rand}-\mu _\text {t,WC}\) | 0.10 | 0.15 | 0.02 |
\(N_{Rand}\) | 2500 | 4600 | 1700 |
\(N_{WC}\) | 2500 | 4600 | 1800 |
