METHOD FOR PRODUCING FEATURE GROUP BASED ON UPPER BOUND OF CO-OCCURRENT SUBGRAPH

摘要

PURPOSE: A characteristic group generation method for considering the upper limit classification value of a co-occurrence frequency partial graph is provided to select a classification characteristic based on the small amount of labeled graph and unlabeled graph, thereby selecting the classification characteristic with the high classification ability. CONSTITUTION: A characteristic group is generated by using an individual frequency partial graph with an upper classification level based on the classification value or the upper limit value of individual frequency partial graph that is pattern-expanded in a graph database (S20). If the upper limit classification value of a co-occurrence frequency partial graph that is composed by the combination of the individual frequency partial graph is smaller than the minimum classification level of the individual frequency partial graph, the generation of a new co-occurrence frequency partial graph is stopped by adding remaining individual frequency partial graph except for the individual frequency partial graph to the co-occurrence frequency partial graph (S50). [Reference numerals] (AA) Start; (BB) No; (CC) End; (S10) Mining an occurrence frequency partial graph; (S20) Generating a characteristic group; (S30) Generating a co-occurrence frequency partial graph; (S40) Classification upper value > a minimum classification level?; (S50) Stopping generating an additional co-occurrence frequency partial graph