Table 2 Entropies of fairness attributes in the recommended clinicians.

From: Matching clinicians with clinical trials using AI

Model

CS@GT (↑)

Race entropy (↑)

Ethnicity entropy (↑)

Geographical entropy (↑)

Ground Truth

0.14 (—)

0.43 (—)

0.59 (—)

0.67 (—)

Random

0.05 (0.12)

0.42 (0.06)

0.58 (0.05)

0.66 (0.02)

Popular

0.18 (—)

0.36 (—)

0.61 (—)

0.68 (—)

LightGCN

0.30 (0.04)

0.41 (0.03)

0.61 (0.04)

0.66 (0.02)

ConvNCF

0.33 (0.03)

0.42 (0.02)

0.59 (0.03)

0.66 (0.02)

NCL

0.37 (0.05)

0.42 (0.01)

0.58 (0.03)

0.65 (0.01)

Doc2Vec

0.30 (0.02)

0.41 (0.02)

0.58 (0.03)

0.66 (0.02)

FRAMM

0.35 (0.03)

0.44 (0.02)

0.63 (0.04)

0.66 (0.02)

DocTr-regular

0.60 (0.02)

0.43 (0.03)

0.60 (0.03)

0.66 (0.02)

DocTr-genetic

0.61 (0.02)

0.47 (0.03)*

(0.02)*

0.67 (0.02)

  1. The performance metrics are reported as mean (s.d.). *The performance difference is statistically significant (P < 0.01). Bold numbers indicate best performance.