Skip to main content

Table 2 Performance on the BioSimVerb (in ρ) using representations learned with different context configurations. (Bold: best-performing configuration and its score)

From: A neural classification method for supporting the creation of BioVerbNet

Baseline Spearman’s ρ
BOW (win=5) 0.4664
DEP-ALL 0.4323
Configurations: Verb  
POOL-ALL 0.4724
conj+obj+pcomp+prep+rel+subj 0.475
conj+obj+prep+rel+subj(Best) 0.4889
conj+obj+pcomp+prep+subj 0.4578
conj+obj+pcomp+rel+subj 0.4478
conj+obj+pcomp+prep+rel 0.4406
conj+obj+prep+subj 0.4611
conj+obj+rel+subj 0.4572
conj+obj+prep+rel 0.442
comp+obj+pcomp+prep+rel+subj 0.4376
comp+conj+obj+prep+rel+subj 0.4762
comp+conj+obj+pcomp+prep+subj 0.4655
comp+conj+obj+pcomp+rel+subj 0.4583
comp+conj+obj+pcomp+prep+rel 0.4413
comp+conj+obj+prep+subj 0.4635
comp+conj+obj+rel+subj 0.4592
comp+conj+obj+prep+rel 0.442
obj+pcomp+prep+rel+subj 0.4446
obj+prep+rel+subj 0.441
  1. BOW denotes a basic SGNS learned with bag-of-words context with context window size 5. DEP-ALL denotes a configuration where no filtering of context is used. POOL-ALL denotes a configuration where all individual context bags from the verb-related pools are used. “Best” identifies the best-performing configuration found