Team PolyU-CBSNLP at BioCreative-VII LitCovid track : ensemble learning for COVID-19 multilabel classification

Abstract

This paper briefly describes our works for the LitCovid shared task of BioCreative-VII Track 5. It is an ensemble learning-based system that utilized multiple biomedical pretrained models. In particular, we leveraged seven advanced models for initialization with homogeneous and heterogenous structures through an ensemble bagging manner. To enhance the representation abilities, we further proposed to employ additional biomedical knowledge to facilitate ensemble learning. The experimental results on the LitCovid datasets show the effectiveness of our proposed approach.