Academic Relation Classification Rules Extraction with Correlation Feature Weight Selection

Abstract

For extracting classification rules of academic relations in research project applications, insufficient samples result in deviation because irrelevant features has a impact on decision tree generating. Therefore, this paper proposes a decision tree algorithm combined with correlation feature weight selection to solve this problem. The algorithm selects relevant features at first, which are assigned a prior weight when decision tree is being generated, so that relevant features can be preferentially selected. This paper states the principle of correlation feature weight selection, designing of feature extraction functions of academic relations and the extraction process of classification rules of teacher-student, co-author and co-project. The experiment results show that the proposed method is effective on extraction of academic relations.