Mining Association Rules from Semi-Structured Data.
Kohei Maruyama, Kuniaki Uehara · 2000
Despite the growing popularity of semi-structured data such as Web documents, most knowledge discovery research has focused on databases containing well structured data. In this paper, we try to find useful information from semistructured data. In our approach, we begin by representing semi-structured data in a prototype-based approach. We then detect the most typical common structure of semistructured data and re-store the data into this structure. We can consider this common structure as a structured layer. The structured layer filters out useless properties of semistructured data. Next, we apply the algorithm of mining association rules to the structured layer by using the idea of concept hierarchy. Concept hierarchy maintains relationships between concepts. The use of concept hierarchy allows us to generate extra rules in addition to originally generated rules. The extra rules contain related concepts with the concepts in the original rules. These extra rules are often more informative and useful for finding patterns. In this way, some kind of knowledge can be extracted from semi-structured data. 1.