Figure 1.

(a) Schematic of intrinsic and extrinsic evaluation. The set of articles were randomly split 50–50 five times for extrinsic evaluation. Set A was used as the train/validation set and Set B as the test set for one instance and vice versa for a second instance. For 10 training sets and two embeddings types with 4 different dimensions (50,100,200,300) each, a total of 80 models were trained. (b) shows an expanded view of the model training pipeline for multi-label classification for a single model. Not all elements in vectors or neurons in the model are shown. Abbreviations: NN (Neural Network), WG (Wikipedia 2014 + Gigaword 5)