Adapting a rule-based relation extraction system for BioCreative V BEL task

We tested a rule-based semantic parser in the BEL statement extraction task of BioCreative V Track4 challenge. While the system achieved an overall F-measure of 21.29% with gold standard entities, it achieved a very low performance of 13.86% with the entities extracted by ensemble of NER systems on test data set. For relation extraction, the system achieved a F-measure of 65.13% on test data set. The limitation in the rule sets to map the textual extractions to BEL function is one of the reasons for our low performance in extracting the complete BEL statement. Besides, the lack of ability to extract long distance relationships, recursive relations and the inability to make certain semantic inferences had significant influence on the overall performance of the system.