基于卷积树核的事件论元角色抽取方法

Argument Extraction Algorithm Based on Convolution Tree Kernel

事件论元角色抽取是事件抽取的关键环节,句法分析信息对事件论元角色抽取具有重要作用.传统基于机器学习的方法通常将句法分析信息转化为平面特征,并不能全面利用句法分析信息.为此,提出基于卷积树核的事件论元角色抽取方法.首先,构造基本树结构,将句法分析信息转化为结构特征;其次,针对句法结构树包含较多冗余信息的问题,设计相应裁剪算法,优化树结构、减少卷积树核计算的时间复杂度;最后,构造复合核将平面特征与结构特征相结合,并训练支持向量机分类器完成事件论元角色抽取.实验证明,本文方法使事件论元角色抽取效果有了明显提升.

Argument extraction is the key step of event extraction,syntactic information has an important role in argument extraction.Traditional argument extraction algorithms based on machine learning usually changed the syntactic information into plane features and didn＇t make full use of syntactic information.To solve this problem,this paper proposed an extraction method based on convolution tree kernel.First,this paper constructed the basic structure of the tree to change the syntactic information into structured features;Secondly,since the syntactic trees contained much redundant information,this paper proposed an clipping algorithm to optimize the tree structure and reduce the computing time of convolution tree kernel;Finally,by constructing a compound kernel which combined the plane features and the structured features and training a SVM classifier,this paper completed the event argument extraction.Experimental results showthat this method makes the effect of event argument extraction improving significantly.