dependency tree kernels for relation extraction（依赖树关系提取的内核）.pdf

Dependency Tree Kernels for Relation Extraction Aron Culotta Jeffrey Sorensen University of Massachusetts IBM T.J. Watson Research Center Amherst, MA 01002 Yorktown Heights, NY 10598 USA USA culotta@ sorenj@ Abstract Entity Type Location Apple Organization Cupertino, CA We extend previous work on tree kernels to estimate Microsoft Organization Redmond, WA the similarity between the dependency trees of sen- tences. Using this kernel within a Support Vector Table 1: An example of extracted fields Machine, we detect and classify relations between entities in the Automatic Content Extraction (ACE) corpus of news articles. We examine the utility of sula et al., 2002; McCallum and Wellner, 2003). different features such as Wordnet hypernyms, parts We describe a relation extraction technique based of speech, and entity types, and find that the depen- on kernel methods. Kernel methods are non- dency tree kernel achieves a 20% F1 improvement parametric density estimation techniques that com- over a “bag-of-words” kernel. pute a kernel function between data instances, 1 Introduction