基于群集智能的CRF与规则结合的中文地址抽取  

Address extraction in Chinese by associating CRF with rules based on swarm intelligence

作　　者：汪闯闯 姬东鸿[1] 

机构地区：[1]武汉大学计算机学院,武汉430072

出　　处：《计算机应用研究》2015年第3期727-730,共4页Application Research of Computers

基　　金：国家自然科学基金重点资助项目(61133012);国家自然科学基金面上项目(61173062)

摘　　要：地址抽取是信息采集研究的热点,但是如何保证高准确率与召回率一直是一个具有挑战性的问题。将基于CRF的方法与规则抽取方法相结合来进行地址抽取,并将优化后的结果再用来进行CRF训练;此过程不断循环迭代,直至性能不再提升,最后得到多个CRF分类器。这个不断迭代的过程就是基于群集智能的ACG算法流程,最后再通过组合多个分类器的结果形成最终结果。实验表明,该方法可以把召回率和准确率分别提高到96.44%和97.73%。Address extraction is a hot research topic in the field of information extraction. But how to ensure its high precision and recall remains a challenge. This paper presented the association between CRF and rule-based method and then used their optimized result in return for CRF training. This process iterated until reaching the best performance and resulted in several CRF classifiers. The whole procedure was just the process of ACG algorithm based on swarm intelligence. At last, it generated the final result by combining the results of all models with each other. The experiment shows that the method can promote the recall and the ~recision to 96.44% and 97.73%.

关 键 词：条件随机场 地址抽取 规则抽取方法 群集智能 ACG算法 组合 

分 类 号：TP391.43[自动化与计算机技术—计算机应用技术]