Co-STM text categorization method based on Supervised Topic Model

Abstract

Traditional text classification methods are only trained on homogeneous data with uniform sample distribution and complete category labeling. With the rapid development of computer-related technology, today's data has the characteristics of diverse carrier forms, lack of label data, and unbalanced categories. As a result, traditional text classification technologies can no longer meet the requirements of today's text classification. Aiming at the new characteristics of data, this paper proposes a co-STM (collaborative text classification combined with Supervised Topic Model) collaborative text classification algorithm that integrates SLDA (Supervised LDA) supervised topic model. For category imbalance data, this paper adopts a confidence calculation method based on posterior probability distance and a category-based unlabeled sample selection strategy to select credible unlabeled samples. Comparative experiment results show that the co-STM text classification algorithm can effectively improve the performance of semi-supervised text classification.