The Design And Implementation Of Biomedical Event Extraction System

In recent years, the number of biomedical literature is growing, theautomatic text extraction of these unstructured information and organize it intostructured information to facilitate analysis and management is becoming ahot topic in the field of bioinformatics. Information extraction technologydevelops ceaselessly in recent years, with the method based on machinelearning has gradually replaced the mainstream rule-based methods, theaccuracy of information extraction increased continuously and the forefront ofresearch is changing from the named entity recognition, relation extraction oftradition, to higher level of biomedical event extraction. Biomedical event isthat the correlations and the variations of between various biomedical entitiesat a molecular level. Biomedical event extraction is at a higher level comparedNER and relation extraction, therefore, the processing flow of the system isalso longer. However, in the field of domestic research there are few softwaresystems to adapted the changes of this research direction.This paper introduces the development status of biomedical eventextraction system, the processing flow and the development process, andimplemented a set for biomedical literature event extraction solution by usingsupport vector machine model, the high scalability script language of Rubyand the key-value structured non-relational database based of Redis. Thispaper adopted a classical process in handling event extraction, mainly dividedinto the following three steps: pre-processing, trigger features extraction andevent element features extraction, trigger word detection and event elementdetection. In this paper, the design and development of systems is focus onusability, especially focus on the efficiency when using the system onexperimental study.By using the metaprogramming capabilities of Ruby and the cachingfeatures of Redis, compared to the same event extraction system, the system described in this paper is easier to expand in features selection. When part ofthe features or the corpus changes, relying on rules of cache updating,processing can be performed for changing sections, without having tore-process the entire event extraction. Therefore, experiments which need tobe optimized can be performed faster, thus the system is more suitable for theresearch of biomedical event extraction.The results of the experiment show that, compared to the control, thesystem described in this article finished the extraction faster, the extractionefficiency is improved.