应用粒子群优化-条件随机域的文本生物实体识别  

A Bio-Entity Recognition Algorithm for Literature by Conditional Random Field Model Based on Improved Particle Swarm Optimizer

作　　者：豆增发[1] 高琳[1] 

机构地区：[1]西安电子科技大学计算机学院,西安710071

出　　处：《西安交通大学学报》2010年第12期38-42,124,共6页Journal of Xi'an Jiaotong University

基　　金：国家自然科学基金重点资助项目(60933009);国家自然科学基金资助项目(60970065);高等学校博士学科点专项科研基金资助项目(200807010013)

摘　　要：针对生物医学文本中传统生物实体识别算法的精确度不高的问题,提出了一种新的基于粒子群优化-条件随机域的生物实体识别算法.新算法利用改进的粒子群优化算法训练条件随机域模型,并将训练后的条件随机域模型应用到生物实体的识别上.改进的粒子群优化算法引入粒子群聚集度来防止粒子群过早地陷入局部收敛,用迭代间对数似然相对变化率来控制算法的收敛,用线性变化的惯性因子和学习因子来控制搜索范围.实验结果表明,基于改进粒子群优化的条件随机域模型较隐马尔科夫模型、最大熵马尔科夫模型、支持向量机以及传统条件随机域模型等方法具有更高的精确率和召回率.A new bio-entity recognition algorithm is proposed to improve the precision of bio-entity recognition for biomedical literature.The new algorithm trains the conditional random field model using an improved particle swarm optimizer,and then applies the trained conditional random field model to bio-entity recognition.The aggregation degree of particle swarm is utilized to control the early local convergence of the particle swarm optimizer,the relative change ratio of log-likelihood between iterations is employed to end its iterations,and the inertia factor and learning factor are set as linear variables to control the scope of search space.Experimental results show that the proposed algorithm outperforms the models of HMM,MEMM,SVM and traditional L-BFGS CRF on precision and recall.

关 键 词：条件随机域模型 粒子群优化 粒子群聚集度 对数似然相对变化率 生物实体识别 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]