Comprehensive Feature Information Learning for Multilabel Text Classification

Multilabel text classification (MLTC) enables the discrimination of objects by establishing associations across multiple labels through various dimensions. In the context of data mining tasks, the supervised learning approach in MLTC plays an essential role in enhancing predictive accuracy. The main focus of this research is to exploit the relationship between relevant and correlated features in multilabel text and to select the dimensionality of the vector space to be comprehensively reduced while preserving the performance of the classifiers. We proposed a novel feature selection technique called Comprehensive Measurement Feature Selection (CMFS) in multi-label text categorization to address these issues. Evaluates the significance of the feature using an intercategory and an intracategory relation. We evaluated CMFS on two reference document collections, that is, Reuters and 20Newsgroup. Our approach incorporates both correlation information and class deviation for effective feature selection. Encouraging results are obtained.