Statistical algorithms for ontology-based annotation of scientific literature

Table 2 High level description of the algorithm.

Input • Un-Labeled Item I • Bayesian Decision Tree T Output • Label Vector in Multiple Categories L Algorithm t = Root(T) SearchList = NULL while t ~= NULL do L = L : Bt(I) for l ϵ B_t(I) do SearchList = SearchList : Child(l, t) end for t = SearchList[0] x : SearchList = SearchList end while return L

This recursive program uses the Bayesian Decision Tree defined in Definition 9, along with Bayesian Classifier of Definition 10 and the child function of Definition 11 to label an unlabeled item. Unlike a normal naive Bayes classifier that is trained on the whole training set, this algorithm steps through a decision tree whose every node contains a classifier that is trained on a narrow subset of the original training set. This subset is limited to only those items whic h are annotated with the labels of the ancestors of the current node.

ISSN: 2041-1480