Predicting relevance of abstracts

[damian0604]

• model with a SVM that we actually use
• a (less polished) notebook for exploring the same thing on colab with BERT, which does not seem to be worthwile to pursue (given the very good recall of the traditional models)

[damian0604]

@JBGruber added predictions