A Kernel for Measuring Structural Semantic Similarities

Semantic similarity is nowadays one of the widely discussed topics in data mining, natural language processing and some related research fields. Semantic similarity between two entities usually comes into one's sight when tackling such issues. In this paper, however, we adopt such a standpoint that semantic similarities also exhibit within the document structures besides within linguistic hierarchies the entities embed. We discuss the measurements of such structural semantic similarity in the paper. We define semantic content to depict the semantic capacity of a structure and present a kernel for measuring semantic similarities between tree-structured data. After the recursive generation of all matched subtrees, the semantic similarity between two structures is calculated.