Research on tree kernel-based personal relation extraction

Abstract

In this paper, a kernel-based personal relation extraction method is presented. First, a personal relation corpus is built through filtering and expansion from the ACE2005 Chinese corpus. Then, the structured information, which is appropriate for personal relation extraction, is constructed by applying pruning rules on the basis of the shortest path-enclosed tree. After that,TongYiCi CiLin semantic information is embedded into the structured information. Finally, re-sampling techniques are employed to alleviate the data imbalance problem inherent in the corpus distribution. Experimental results show that, the pruning rules, the embedding of semantic information and the application of re-sampling techniques can improve the F1 score by 3.5, 3.0 and approximate 3.0 units respectively compared with the baseline system. It suggests that the method we propose is effective for personal relation extraction. © 2012 Springer-Verlag.