基于混合核方法的上下位语义抽取

摘要

上下位语义抽取对于知识库构建、信息检索、智能语音以及其他语义应用都具有重要意义。如何有效地描述语义对象的上下文相似度是语义抽取的关键。文本核方法能在更高的维度上比较文本的语义相似性，显示出良好的应用前景。但是，目前常用的文本的语法解析树核以及文本序列核对子句长度较为敏感。提出一种新的混合文本核方法，在利用文本中词法和语法信息改进现有的解析树核和文本串核的基础上，对于句子长度具有自适应性。实验显示与已有核方法相比较，该方法取得了较好的效果，显著地提高了上下文语义抽取的准确率和召回率。%Hypernymy extraction is of paramount importance in building semantic knowledge base,information retrieval,intelligent voice and other semantic applications.The key of the semantic extraction lies on how to effectively depict the context similarity of semantic objects. Text kernel method shows good applied prospects due to its performance of comparing the semantics similarity of the text with much higher dimension.However the kernels of the parsing tree and the text sequence,which are all commonly used currently,are sensitive to the length of clauses.In this paper,we propose a novel hybrid text kernel method,based on improving existing parsing tree kernel and text sequence kernel by using morphology and syntax information in the text,the method has the adaptability on the length of sentence.Experiment demonstrates that the this method achieves pretty good effect in comparison with existing kernel methods in improving the precision and recall ratios of context semantic extraction remarkably.