Table 2.

Features used in two individual sequence labeling modules: CRFs and SSVMs.

Feature	Description
Bag-of-words	Unigrams: w0, w−1, w1, w−2, w0;
	Bigrams: w−2w−1, w−1,w0, w0w1, w1w2;
	Trigrams: w−2w−1w0, w−1w0w1, w0w1w2
	wi is a token at position relative the current token.
Part-of-speech (POS) tags	Unigrams: p0, p−1, p1, p−2, p2
	Bigrams: p−2p−1, p−1p0, p0p1, p1p2;
	Trigrams: p−2p−1,p0, p−1,p0,p1, p0,p1,p2
	pi is a POS tag at position i relative the current token.
Combinations of tokens and POS tags	w−1p−2, w1p−1, w−1p0, w2p−1, w0p0, w0p1, w1p0, w1p1, w1p2,
Sentence information	Length of the current sentence; whether there is any bracket unmatched in the current sentence?
Affixes	Prefixes and suffixes of the length from 1 to 5.
Orthographical features	Whether the current word is an upper Caps word? Contains a digit or not? Has uppercase characters inside? Etc.
Word shapes	Any or consecutive uppercase character(s), lowercase character(s), digit (s) and other character(s) in the current word is/are replaced by ‘A’, ‘a’, ‘#’ and ‘-’ respectively.
Section information	Which section the current word belongs to, title or abstract?
Word representation features [5]	Brown clustering (https://github.com/percyliang/brown-cluster);Word2vec (https://code.google.com/p/word2vec/).
Dictionary features	Chemical dictionary: CTD, DrugBank, MeSH, Pharmacogenetics Knowledge Base (PharmGKB) (26), UMLS, and Wikipedia;
	Disease dictionary: CTD, MeSH, UMLS, disease ontology (27), National Drug File Reference Terminology (NDF-RT) (28) and Wikipedia.
Frequency features	Whether the frequency of the current word is higher than a given value (4 in our system) and the inverse document frequency of it is less than another given value (0.1 in our system)?
Character N-grams	Character N-grams (N = 1, 2, …, 4) within the current word.