Research on statistical discovery of a wide var i ety of patterns with low frequencies and its applications

点击翻译按钮获取中文摘要

The goal of this research is to develop a framework to, given large text data, discover patterns which do not appear frequently. To achieve this goal, we review our existing researches from the following two viewpoints : Mapping of letters for pattern discovery : Using an approximate pattern matching, we have proposed a pattern discovery and evaluated by experiments. In this method, we have found that mapping from several letters into one digit plays an important role. Metric space for pattern discovery : The goal of this topic is to distinguish patterns from non用atterns. Instead of a rigid metric space, we first find usual substructures, and then we find a pattern as a combination of usual substructures. We have evaluated this method by experiments on genome sequences and Web documents.