药物实体和药物相互关系的联合识别

Joint extraction of drug entity and drug-drug interaction

目前在识别药物实体和药物相互关系的方法中,使用的主要是串行的方法。这类方法会带来误差传递和无法利用两步相关性的缺点。结构感知机算法能改进这两大缺点。同时对药物实体和药物相互关系进行搜索,利用实体信息和实体关系信息对结果进行打分,选取最好的结果。实验结果表明,该算法对药物实体和药物关系的识别效果对比串行算法有较大提高,针对生物信息文本设计的特征能有效提高算法性能。

The methods used to extract drug entities and drug-drug interactions are mostly pipeline approaches. These methods will bring two defects, the error producing in the first step will propagate to the next step, and they cannot make use of the rela-tions between two steps. Structured perceptron algorithm can overcome these defects. Drug entities and drug-drug interactions were searched simultaneously,and the results were rated according to entity information and entity relation information, the best result was then figured out. Experimental results suggest that the proposed algorithm outperforms pipeline approaches and the features designed for biomedical text are useful to improve the performance of algorithm.