Application of Capsule Network in Text Classification Problem

Abstract. The rise of the Internet and the big data era has led to a significant increase in text data, making text classification an essential task in natural language processing. Conventional deep learning models have advanced considerably in understanding text context dependencies; however, they still struggle with long-range dependencies and multi-scale features. In recent years, capsule networks (CapsNets) have demonstrated outstanding results in the area of image recognition due to their unique structure and dynamic routing mechanism, which has inspired researchers to explore their application in text classification tasks. This article examines the advancements in software and research related to CapsNets in the context of text classification challenges. Through its capsule units and dynamic routing algorithms, capsule networks can better express the part-whole hierarchical structure of text and effectively capture the complex relationships in text. Compared with traditional neural networks, capsule networks have shown advantages in dealing with both tasks involving text classification with either a single label or multiple labels. In addition, this paper also examines ways to optimize and enhance CapsNets for text classification, such as incorporating attention mechanisms to improve the model's capacity to recognize important information, as well as refining routing algorithms.