A Multi-classification Model of Industry Based on Multi-task Learning

With the development of economy, Chinese enterprises are developing rapidly and diversifying. It is of practical significance to study the problem of multi-label classification of industry. In this paper, we propose a multi-task industry multi-classification model, MABA (Multi-task learning based on ALBERT pre-training, Bi-GRU, Attention), which uses a multi-task learning method that jointly trains industry classification tasks and industry chain classification tasks to assist in industry classification. The model was experimented on more than 790,000 data, and multi-task learning was experimentally compared with single-task learning. The results showed that multi-task learning has improved in precision, recall, and f1-score than single-task learning, which proves the effectiveness of multi-task learning.