A linear classifier based on entity recognition tools and a statistical approach to method extraction in the protein-protein interaction literature

Table 10 Summary of evaluation by three human annotators, over 1049 evidence sentences for PPI methods.

Label	# of sentences tagged by the Majority as Label	% of sentences tagged by the Majority as Label
Y	755	72%
M	112	11%
N	165	16%

The table shows the statistics of majority annotation labelling 1049 sentences, each by three independent annotators. For each annotation value, shown in the right column, we list how many sentences were labelled with this value by at least two of the three annotators.
The possible labels are: Y - if the sentence discusses a method which can Potentially be applied for detecting protein-protein interaction; M - if the sentence discusses a method, but the method is NOT a protein-protein interaction detection method; N - if the sentence DOES NOT discuss a method.
Note that the total number of majority-vote sentences is 1032 rather than 1049, because on 17 sentences the 3 annotators had a 3-way disagreement. (Roughly 1% of the sentences, hence the total percentage is 99%)

ISSN: 1471-2105