For each sentence, the system maintains a list of generated words with the original words as a reference to be used in the extraction phase. For the text retrieval, we apply co-occurrence to obtain high sensitivity, while in the relation extraction phase, we apply NLP and rule-based methods to gain high specificity. This is to make sure that learning data and test data do not overlap thus avoiding bias the final results. However, the long-term use of these drugs leads to drug resistance caused by the viral mutations that occur under drug pressure. Currently we do not take those sparse situations into account. We extracted relations from all candidate sentences of the collected abstracts and obtained 2,434 extracted relations.
The text preprocessing component then simplifies sentences, parses them using the Stanford Lexicalized Parser version 1.6  and applies grammatical relations to generate sentence components. Table 6 shows the evaluation results of the system over these two datasets. Only select a
Our task here is to determine a resistance type for each pair from these pieces of evidence, which can have the following properties: Containing false positive relations due to relation extraction method or relations are taken out of context. This method (Base_C) predicts a