A New Inductive Learning Method for Multilabel Text Categorization

Abstract

In this paper, we present a new inductive learning method for multilabel text categorization. The proposed method uses a mutual information measure to select terms and constructs document descriptor vectors for each category based on these terms. These document descriptor vectors form a document descriptor matrix. It also uses the document descriptor vectors to construct a document-similarity matrix based on the "cosine similarity measure". It then constructs a term-document relevance matrix by applying the inner product of the document descriptor matrix to the document similarity matrix. The proposed method infers the degree of relevance of the selected terms to construct the category descriptor vector of each category. Then, the relevance score between each category and a testing document is calculated by applying the inner product of its category descriptor vector to the document descriptor vector of the testing document. The maximum relevance score L is then chosen. If the relevance score between a category and the testing document divided by L is not less than a predefined threshold value λ between zero and one, then the document is classified into that category. We also compare the classification accuracy of the proposed method with that of the existing learning methods (i.e., Find Similar, Naïve Bayes, Bayes Nets and Decision Trees) in terms of the break-even point of micro-averaging for categorizing the "Reuters-21578 Aptè split" data set. The proposed method gets a higher average accuracy than the existing methods.