Classifier Ensemble For Bio-entity And Relation Recognition

Bio-entity recognition and relation recognition is a task in biomedical text mining. Currently, most of methods for bio-entity recognition are based on a single machine learning algorithm and it can not achieve better performance.this shows that a single algorithm can not identify the purpose of high efficiency.Through analysising various statistical learning methods,we can find between the different learning models with complementary and relevant,so the classifier fusion is a new ideas.This article has the research from the two-depth:(1)Classifier based on single and multi-classifier for Bio-entity and relation recognition.First,we research the single classifier,use Maxent and CRF algorithm.using a rich feature set.Yapex corpus to identify the protein names.System combines a rich feature set.we introduce the abbreviated word recognition module,boundary expansion module and filter module processing.Then,for the problems,we study the multi-classifier based Bio-entity.Between the use of different complementary learning model and relevance to further enhance the biomedical named entity recognition performance.it is a fusion method.that is CRF and Maxent algorithm integration of algorithms.using different learning models exist complementary and relevant of each other,it can improve the performance.(2)Classifier based on single and multi-classifier of Bio-entity recognition and relation recognition.We study the interaction between the protein identification,analyze the characteristics,including a variety of shallow linguistic features,such as interactive features,key features,context characteristics.they are effective shallow linguistic features.Taken a variety of shallow linguistic features,including interactive features of the protein entities.the first method used to identify a single classification for the single classifier approach has the one-side,taking into account the classification of different classification model of complementarity between the results and relevance,using the same fusion,identify the interaction,and achieve good recognition performance.