Skip to main content
. 2017 May 25;2017:bax019. doi: 10.1093/database/bax019

Table 9.

Performance with different word embedding and different position embedding size

Word embedding Position embedding P R F1
random 0 62.08% 58.77% 60.38%
5 69.34% 55.9% 61.9%
10 70.76% 54.36% 61.48%
Wiki_bow_8w_25n 0 60.89% 54.46% 57.5%
5 59.2% 60.72% 59.95%
10 70.64% 53.54% 60.91%
Bio_skip_8w_25n 0 62.39% 57.85% 60.03%
5 67.8% 53.33% 59.7%
10 66.92% 55.18% 60.48%
Bio_skip_10w_10n 0 70.66% 49.64% 58.31%
5 61.84% 56.51% 59.06%
10 68.77% 54.87% 61.04%
Bio_bow_8w_25n 0 64.09% 54.36% 58.82%
5 69.43% 54.05% 60.78%
10 67.27% 49.95% 57.33%
Bio_bow_5w_10n 0 58.25% 59.38% 58.81%
5 60.18% 61.23% 60.7%
10 65.21% 56.72% 60.67%

The prefix Wiki (Wikipedia corpus) or Bio (BioASQ dataset) refers to the corpus used to train the word embedding model. The label bow (CBOW) or skip (skip-gram) refers to the type of architecture used to build the model. The number preceding w and n indicates the size of the context window and the negative sampling, respectively.