Skip to main content

Table 8 The top 10 words with large learned attention weight in PPI, DDI, and ChemProt corpora

From: Investigation of improving the pre-training and fine-tuning of BERT model for biomedical relation extraction

Task Word stem
PPI Activ(ate), complex, associ(ate), interact, human
Protein, bind, domain, specif(y), receptor
DDI Concomitantli, combin(e), concomit(ant), increas(e), use
Concurr(ent), decreas(e), inhibit, receiv(e), administ(er)
ChemProt Phosphoryl(ate), attenu(ate), stimul(ate), deriv(e), regul(ate)
Novel, metabol(ize), reduc(e), induc(e), inhibit
  1. For the calculation of global attention weight, we use Porter’s stemmer [32] to obtain the word stem for each word since words might in different forms in the sentence. For example, the stem of “activate” is “activ”, and the words like “activation” and “activates” share the same word stem