Subtree Similarity Search Based on Structure and Text

Given a query tree, the subtree similarity search problem is finding all subtrees in a document tree that are similar to the query tree. The previous scan-based method extracts candidate subtrees based on the size difference, which only considers the structural differences and ignores the differences in the contents represented by the trees. For this reason, it suffers from the following two issues. First, for queries against a tree with a regular structure, it is difficult to differentiate subtrees in terms of structural similarity, yielding a large number of candidate results to verify. Second, the candidates are verified by computing the tree edit distance, which is cubic to the number of tree nodes. In this paper, we propose a solution for the subtree similarity search problem based on the structure and contents of the trees. We demonstrate through experiments that our proposed method outperforms the previous scan-based methods in terms of speed and is competitive with index-based methods.