Skip to main content

Table 2 Weighted F1 scores for various models trained on single sentences. Best results for each dataset are printed in bold. For our models, training time is given (for hyperparameter settings yielding the shown score)

From: Fast and scalable neural embedding models for biomedical sentence classification

Model

PubMed 20k

PubMed 200k

Extended corpus

Logistic regression model (LR) [11]a

.831

.859 (33,006 s)

-

Forward artificial neural network (ANN) [11]a

.861

.884

-

Conditional random field (CRF) [11]a

.895

.915 (4867 s)

-

bi-ANN [1]a

.900

.916

-

fastText single-sentence (ours)

.825 (5 s)

.852 (13 s)

.852 (61 s)

fastText with sentence context and numeric sentence position (ours)

.896 (11 s)

.917 (73 s)

.919 (183 s)

  1. aResult and runtime reported by [11]; the reported runtimes given by authors include both training and testing time while we report only training time. Testing of a trained fastText model took approx. 15 s with the evaluation tool supplied by the fastText library.