Citation Polarity Identification in Scientific Research Articles Using Deep Learning Methods

The way in which scientific research articles are cited reflects how previous works are utilized by other researchers and stakeholders and can indicate the impact of that work on subsequent experiments. Citations can be perceived as positive, negative, or neutral. While current citation indexing systems provide information on the author and publication name of the cited article, as well as the citation count, they do not indicate the polarity of the citation. This study aims to identify the polarity of citations in scientific research articles. The method uses pre-trained language models, BERT, Bio-BERT, RoBERTa, Bio-RoBERTa, ELECTRA, ALBERT, and SPECTER, as the word embeddings in a deep-learning classifier. Most citations have a neutral polarity, resulting in imbalanced datasets for training deep-learning models. To address this issue, a class balancing technique is proposed and applied to all datasets to improve consistency and results. Ensemble techniques are utilized to combine all of the model predictions to produce the highest F1-scores for all three labels.