A New Text Semi-supervised Multi-label Learning Model Based on Using the Label-Feature Relations

Abstract

Multi-label learning has become popular and omnipresent in many real-world problems, especially in text classification applications, in which an instance could belong to different classes simultaneously. Due to these label constraints, there are some challenges occurring in building multi-label data. Semi-supervised learning is one possible approach to exploit abundantly unlabeled data for enhancing the classification performance with a small labeled dataset. In this paper, we propose a solution to select the most influential label based on using the relations among the labels and features to a semi-supervised multi-label classification algorithm on texts. Experiments on two datasets of Vietnamese reviews and English emails of Enron show the positive effects of the proposal.