A Wavelet Transform Based Structural Similarity Model for Semi-structured Texts

摘要

The semi-structured texts including Xml and Html texts are a basic information format in the Internet and World Wide Web. The text content values and the tree-organized structure are two aspects of a semi-structured text. Usually, the same text contents with different structures imply different objects. So the structural similarity of semi-structured texts is an essential key point to search, index, retrieve, query, or compare information in web pages. We presents a Wavelet Transform Based Structural Similarity Model (WTBSSM) in order to fast measure the structural similarity of semi-structured texts and compress the structural information into a short vector so as to develop an efficient semi-structured text index system. This paper introduces the Binary Encoding Method to convert a semi-structured text into a {-1, 1} sequence. Then the text structure signals are decomposed by means of Discrete Wavelet Transform to get the approximation coefficients, which is only a half length of the original signals. Finally, the structure similarity is measured by the Euclidean distance of approximation coefficients. The experimental results show that the WTBSSM can keep almost the same distance distribution to the direct distance of the original signals with a half or a quarter of information. The comparisons with a method of shorten DWT coefficients suggests that WTBSSM is better than it.