Research and Implementation of a Multi-label Learning Algorithm for Chinese Text Classification

Multi-label learning has received significant attention in the research community over the past few years. Traditional supervised learning techniques do not fit it well, as real-world objects might be complicated and have multiple semantic meanings simultaneously. In our work, we set our goals to mine the involved product attributes in comment data from JD.com. This task is fundamental and significant to businesses for studying the online market feedbacks from consumers. In this paper, we formally define the three types of text categorization problems and analyze the relations among them. Then, we assign some single-label multiclass classifiers to the new training datasets which are created by our constructing algorithms. Thus, multilabel learning is transformed into a series of single-label multi-class binary classification problems: whether an unseen instance belongs to a certain class or not. Finally, we assemble the outputs of all single-label multi-class classifiers to obtain the multiple labels. In the end of this paper, we conducted comprehensive experiments to evaluate the performance of our proposed algorithms.