Classifying protein-protein interaction articles using word and syntactic features

BMC Bioinformatics

Table 2 The feature combinations used for submitted runs on the article classification task

	BC3 Dev Set	Multi-word			MeSH Term	Stemmed GRs	Feature Cut	Higher Order
		UNI	BI	TRI
Run 1		X	X		X
Run 2	X	X	X		X
Run 3		X	X	X	X	X	X
Run 4	X	X	X	X	X	X	X
Run 5		X	X	X	X	X	X	X

The training data used in official submissions includes all examples of previous BioCreative PPI article tasks. However, the BioCreative III development set was selectively added for training in different runs. Unigrams (UNI), bigrams (BI), and trigrams (TRI) were used as multi-word features. MeSH feature is unigrams and bigrams from MeSH terms. For grammar relations (GRs), stemming was performed on Run 3 through Run 5. Feature cut was performed based on the frequency threshold four.

ISSN: 1471-2105