基于SAO结构的非分类关系抽取研究

Extraction of non-taxonomic relations based on SAO structure

针对非分类关系抽取中的关系识别问题,提出利用SAO结构和依存句法分析相结合的识别方法。该方法将中文专利领域的非分类关系抽取问题转化为符合SAO结构的识别问题,通过SAO结构中的动词信息可以解决关系识别的问题,并在此基础上,利用依存句法分析得到的依存关系强度结合传统的特征,分别对新特征、词特征、上下文特征、距离特征的有效性进行验证分析。实验结果表明,该方法优于传统方法,也验证了依存句法分析在非分类关系抽取中的可行性。

In order to solve the problem of relation recognition in the extraction of the non-taxonomic relation,this paper proposes a recognition method that combines the Subject-Action-Object(SAO)structure and the dependency syntax.The method transforms the extraction of the non-taxonomic relation in Chinese patent domain into the recognition problem of SAO structure.The recognition problem of relation can be solved by the verbs information in the SAO structure,and on this basis,the traditional features are combined with the dependency strength that is gotten from the dependency syntax.And then,the validity of new features,word features,context features and distance features are verified and analyzed.The experimental results not only indicate that this method is superior to traditional methods,but also verify the feasibility of the dependency syntax in the extraction of the non-taxonomic relation.