Multi-label Classification for Marine Science and Technology Literature

We propose a deep-learning based multi-label text classification model F-CNN for marine science and technology documents. We have set up a data set that covers about 300 thousand marine related fields for multi-label text classification. In order to solve the problem of disequilibrium of class distribution of data sets, we propose two ways of data enhancement which improves the efficiency of the model. Comparing with traditional text classification methods, the new F-CNN model optimizes the structure of traditional Text-CNN models according to the characteristics of the paper data, and has good classification effect and high execution efficiency in multi-label classification of document data.