Discovery of Characteristic subgraph Patterns Using Relative Indexing and the Cascade Model

: Relational representation of objects using graphs reveals much information that cannot be obtained by attribute value representations alone. There are already many databases that incorporate graph expressions. We focus on syntactic trees in language sentences, and we attempt to mine characteristic subgraph patterns. The mining process employs two methods: relative indexing of graph vertices and the cascade model. The former extracts many linear subgraphs from the database. An instance is then represented by a set of items, each of which indicates whether a specific linear subgraph is contained within the graph of the instance. The cascade model is a rule induction method that uses levelwise expansion of a lattice. The basic assumption of this mining process is that characteristic subgraphs may be well represented by the concurrent appearance of linear subgraphs. The resulting rules are shown to be a good guide for obtaining valuable knowledge in linguistics.

• Type: Conference
• Year: 2000
• Where: PKDD
• Authors: Takashi Okada, Mayumi Oyama
• Keywords: CASCADE Model, Characteristic Subgraphs, Data Mining, Linear Subgraphs, PKDD 2000