A Support Vector Machine And Transformation-based Error Driven Learning Method For Biological Entity Recognition

This paper present the statistical-based machine learning methods combined with rules-based approach effective applied to the field of biological entity recognition. Using SVM, the typical of statistical learning theory, as the implementation of machine learning methods and using Transformation-based Error-Driven method improve the results of SVM. Through basic theoretical research and experiment, The combined method resolved the key issues in the field of biological entity recognition, thus combining the two methods has made better precision and recall.The method firstly extract rich feature set ,such as words features ,the context features, POS features, the word formation features, the core features and the stopword features, etc. Using JNLPBA training corpus trains SVM classifier and then using trainied SVM classifier implement biological entity recognition on JNLPBA testing corpus Through the statistics and analysis of the training Corpus and testing Corpus, we do research on the issues of machine learning method in the field of biological entity recognition, such as the learning generalization ability, feature selection, introduction of external resources, data uneven and so on.To further enhance the effectiveness of recognition, this paper use of the Transformation-based error-driven learning method to revise the SVM Tagging results in experiment.Transformation rules effectively mine the linguistic phenomenon in biomedical text and further improve the results of SVM method. Compared to other researchers, the method in this paper reaches considerable results of other mature applications.