원문정보
초록
영어
Many machine learning methods have been applied on the biomedical named entity recognition and achieve good results on GENIA corpus. However most of those methods reply on the feature engineering which is labor-intensive. In this paper，huge potential feature information represented as word vectors are generated by neutral networks based on unlabeled biomedical text files. We propose a Biomedical Named Entity Recognition (Bio-NER) method based on deep neural network architecture which has multiple layers and each layer abstracts features based upon the features generated by lower layers. Our system achieved F-score 71.01% on GENIA regular test corpus , F-score values for 5-fold cross-validation is 71.01% and this result is closed to the state-of-the-art performance with only POS (Part-of-speech) feature and represents the deep learning can effectively performed on biomedical NER.
목차
1. Introduction
2. Architecture
2.1. Extracting Word Feature Vectors
2.2. Extracting Sentence Level Features
2.3. Label Criterion
2.4. Stochastic Gradient
3. Experiments
3.1. Task Description
3.2. Experiment Result and Analysis
4. Conclusion
References