A Novel Convolution Kernel Model for Chinese Relation Extraction Based on Semantic Feature and Instances Partition
Huijuan Zhang, Shunwei Hou, Xin Xia · 2012
Relation extraction is an important part of the information extraction. Nowadays, researches focus on tree kernels based solutions that employ different tree structures and kernel functions. since those solutions fail to employ semantic feature effectively and have a low Recall, this paper proposes a novel convolution kernel model based on semantic feature and instances partition. This model involves synonym information as a node in a parse tree, varies partial trees as instances partition and uses the convolution tree kernel function for similarity calculation which outputs data for SVM classifier. the experimental results show that the uses of synonyms and instances partition improve the performance of relation extraction.