Improving Citation Sentiment and Purpose Classification Using Hybrid Deep Neural Network Model

Abstract

Automated citation classification has received much attention in recent years from the research community. It has many benefits in the bibliometric field such as improving the methods of measuring publications’ quality and productivity of the researchers. Most of the existing approaches are based on supervised learning techniques with discrete manual features to capture linguistic cues. Though these approaches have reported good results, extracting such features are time-consuming and may fail to encode the semantic meaning of the citation sentences, which consequently limits the classification performance. In this paper, a hybrid neural model is proposed, which combines convolutional and recurrent neural networks to capture local n-gram features and long-term dependencies of the text. The proposed model extracts the features automatically and classifies the sentiments and purposes of scientific citations. We conduct experiments using two publicly available datasets and the results show that our model outperforms previously reported results in terms of precision, recall, and F-score for citation classification.