Protein-Protein Interaction Eztraction from Biomedical Literatures Based on Modified SVM-KNN
This paper presents a novel method to extract Protein-Protein Interaction (PPI) information from biomedical literatures based on Support Vector Machine (SVM) and K Nearest Neighbors (KNN). The two protein names, words between two proteins, words surrounding two proteins, keyword between or among the surrounding words of two protein names, ExpDistance based on word distance of two proteins, ProDistance of two proteins in a protein pair are extracted as features of the vectors. A model based on SVM is setup to extract the interaction. To improve the accuracy of SVM classifier, KNN method is introduced. Furthermore, to fit the unbalanced data distribution, a modified SVM-KNN classifier is proposed. Experiments conducted on BC-PPI corpus show that our modified SVM-KNN classifier with the two distance features is efficient at extracting Protein-Protein Interaction information. The recall, precision and F-score are 87.2%, 82.4%, 84.7% respectively which outperform most of the state-of-the-art systems.
PPI SVM KNN SVM-KNN unbalanced data distribution
Lishuang LI Linmei JING Degen HUANG
Dalian University of Technology Dalian, Liaoning, China
国际会议
大连
英文
1-7
2009-09-24（万方平台首次上网日期，不代表论文的发表时间）