Structure-Preserving Hashing for Tree-Structured Data

Many kinds of data are tree-structured, e.g., XML documents. In this paper, a structure-preserving hashing method for rooted unordered trees is proposed, to compress a tree into compact signature while preserving its structural information, enabling efficient structural similarity estimation and search, duplication detection, etc. The proposed method exploits subpaths of fixed length in a tree. We provide theoretical analysis, showing that under moderate conditions, the signature contains enough information to reconstruct the original tree. And with the signature, similarity between trees can be estimated efficiently. Our proposed method has the advantage of a linear construction time complexity, compared to the quadratic worst-case construction time complexity of the embedded pivot method [24]. A quantitative analysis of the relation to tree edit distance is also provided. Experiments of XML document de-duplication are tested on real world data, showing the effectiveness of the proposed method.