Affiliations: [a]
School of Electronic Engineering, Beijing University of Posts and Telecommunications, Beijing, China
| [b]
Computer Network Information Center, Chinese Academy of Sciences, Beijing, China
Correspondence:
[*]
Corresponding author. Yong Li, School of Electronic Engineering, Beijing University of Posts and Telecommunications, Beijing 100876, China. E-mail: [email protected].
Abstract: The number of scientific papers has been increasing ever more rapidly. Researchers have to spend a lot of time classifying papers relevant to their study, especially into fine-grained subfields. However, almost all existing paper classification models are coarse-grained, which can not meet the needs of researchers. Observing this, we propose a lightweight fine-grained classification model for scientific paper. Dynamic weighting coefficients on feature words are incorporated into the model to improve the classification accuracy. The feature word weight is optimized by the Mean Decrease Accuracy (MDA) algorithm. Considering applicability, the lightweight processing is conducted through algorithm pruning and training sample pruning. Comparison with mainstream models shows simultaneous improvement in accuracy and time efficiency by our model.
Keywords: Artificial intelligence application, fine-grained classification, lightweight processing, machine learning, paper classification system
