Multi-label Text Classification with Deep Neural Networks

Abstract

Text classification is a foundational task in natural language processing (NLP). Traditional methods rely heavily on human-designed features, while deep learning models based on neural networks can automatically capture contextual information. We explore and introduce various neural network architectures to extract information and key components in texts. An extensive set of experiments and comparisons on accuracy, speed, memory-consumption are conducted. Methods based on the proposed models won the first place in the Zhihu Machine Learning Challenge 2017. The code has been made publicly available ¹ .