| 摘要 |
PURPOSE: A feature selection method using a similar graph structure for graph classification and a device thereof are provided to classify graphs including similar part graphs into the same division group and not to split data including the same class. CONSTITUTION: A grouping unit(220) groups frequent part graphs by using structural similarity between the frequent part graphs and an information gain of the frequent part graphs. A feature selection unit(230) selects classification performance which is the most frequent part graph as a feature in the each of frequent part graphs according to the each of frequent part graphs grouped. The grouping unit ranges the frequent part graphs according to an information gain value as a reference and groups frequent part graphs which is a similar structure about the frequent part graphs ranged. |
