利用D-S证据理论进行特征融合的同义实体识别

Synonymous entity recognition based on D-S evidence theory for feature fusion

针对现实中同一实体存在不同表象的问题,提出一种基于D-S证据理论特征融合的同义实体识别方法。以搜索引擎为外部知识库获取实体特征信息,利用相似函数计算特征值,由D-S证据理论融合一组特征值,经阈值判断完成同义实体的识别。特征融合识别算法在医疗机构数据集上的识别精度、召回率和F值分别达到了85.80%、81.18%、83.43%,比单纯利用实体名的算法分别提高了4.09%、4.30%和4.21%。实验表明D-S证据理论将多特征融合,对同义实体识别具有更好的识别效果。

As the same entity has different expressions in the real world,this paper proposed a synonymous entity recognition method based on the D-S evidence theory for feature fusion.First the recognition method obtained the entity features from a search engine,an external knowledge base,and calculated feature values using a similarity function.Then it identified synonymous entities through a threshold value after fusing a group of features using the D-S theory.The recognition accuracy,recall and F value of using feature fusion on the medical institution dataset were 85.80%,81.18%and 83.43%,respectively,which were 4.09%,4.30%and 4.21%higher than the method simply using the entity name.Experiments show that the D-S evidence theory has better recognition effect on synonymous entities by multi-feature fusion.