Research Article

Distant Supervision with Transductive Learning for Adverse Drug Reaction Identification from Electronic Medical Records

Figure 4

Block 1 expresses the data labeling using the fact from external sources (KB seeds). The is a data set that a pair of drug and event entities can be mapped to a set of KB seeds through the distant supervision. Hence, all sentences that correspond to the same drug-event pair are assigned to the same bag and same label (labeled data ) regarding a label of such drug-event pair in a set of seeds from knowledge base. Finally, such set is used as a training data. Block 2 depicts our proposed MIL-dEM method. The label assignment for unlabeled data set (test set) can be obtained from a classifier in the previous process. Lastly, such unlabeled data is incorporated and contributed to estimating the parameters of a generative model.