“…3.2.2 Baselines. For comparison, several representative models were selected as baselines on the dataset used in this paper, including W2V-BiLSTM-CRF, Glove-BiLSTM-CRF, Text-CNN, SCI-BERT, and Rule-based method by Wang et al (2022). Among them, W2V-BiLSTM-CRF and Glove-BiLSTM-CRF use LSTM as the feature extractor, biphasic recurrent neural networks to combine pre and post-text related information, and CRF as the output layer, which is a CRF commonly used in NLP tasks to solve sequence annotation problems.…”