Entity Extraction in Biochemical Text using Multiobjective Optimization

Utpal Kumar Sikdar, Asif Ekbal, Sriparna Saha


In this paper we propose a multiobjectivemodified differential evolution based feature selectionand classifier ensemble approach for biochemical entityextraction. The algorithm performs in two layers. Thefirst layer concerns with determining an appropriate setof features for the task within the framework of a supervisedstatistical classifier, namely, Conditional RandomField (CRF). This produces a set of solutions, a subsetof which is used to construct an ensemble in the secondlayer. The proposed approach is evaluated for entity extractionin chemical texts, which involves identification ofIUPAC and IUPAC-like names and classification of theminto some predefined categories. Experiments that werecarried out on a benchmark dataset show the recall,precision and F-measure values of 86.15%, 91.29% and88.64%, respectively.


Multiobjective modified differential evolution (MODE), feature selection, ensemble learning, conditional random field (CRF), named entity (NE).

Full Text: PDF