摘要
为了识别文本中海量的地理命名信息,以CRF(条件随机场)模型识别为基础,加入制定的规则,来提高CRF模型识别的召回率,从而提高整体的地理命名实体识别效果。通过选取适合的地理命名实体识别的特征模板,验证特征的有效性以及分析CRF模型识别结果中的未识别实体样本,设计针对未识别实体的规则用以修正识别结果。实验表明,对地名和组织名结合规则进行修正后的F值达到了91.61%和85.74%,有了显著提高。
For reeognising massive geographic named entities in texts, on the basis of conditional random field （CRF） model, we add artificial rules to improve its recall rate of recognition, thereby improve the overall effect of geographic named entities recognition. By choosing proper feature template of the geographic named entity recognition, we verify the validity of CRF features and analyse the unrecognised entity samples among CRF model recognition result, and design the artificial rules targeted at the unrecognised entity to correct the recognition resuit. Experiments show that for the place names and organisation names, their F values reach 91.61% and 85.74% after to be corrected in combination with the rules, it is significantly improved.
出处
《计算机应用与软件》
CSCD
2015年第1期179-185,202,共8页
Computer Applications and Software
基金
国家自然科学基金项目(61170022)
关键词
地理命名实体识别
特征模版
条件随机场
修正规则
结合修正
Geographic named entity recognition Feature template Condition random field （CRF） Revise rule Combined correction
