A GeoSpatial Named Entities Recognition(GSNER) method based on combination of Conditional Random Fields(CRF) and rules is proposed. This method takes extensive knowledge as trigger conditions. Triggered text fragments are put into CRF Named Entity Recognition(NER)module, and recognized NEs are deconstructed to several components, CRF and knowledge are also employed for classification of GSNE.Experimental results show that, this combination method and the NE deconstruction strategy effectively promotes the performance of GSNER: the overall Precision, Recall, Fl achieves 92.86％. 90.91％, 91.87％.%提出条件随机场(CRF)与规则相结合的地理空间命名实体识别方法.该方法以丰富的知识作为触发条件,用CRF对满足条件的片段作地名及机构名识别.识别出来的命名实体又被解构,CRF及知识用来进一步判断该命名实体是否表示事件发生地的地理空间信息.实验结果表明,统计与规则方法的结合以及解构算法有效提升了地理空间命名实体识别的性能,准确率、召回率和F1值分别达到92.86%、90.91%、91.87%.
展开▼
