<p>Due to the explosion in scientific publications in recent years, researchers face a lot of difficulties in locating suitable and pertinent publications for citation in their studying topics. The abundance of millions of scientific papers in digital libraries worsens this issue. This leads to a gap in research innovation due to the inability to retrieve comprehensive literature. The automated citation recommendation models are ideally suited to address this issue by suggesting a list of valuable articles for citing when the user provides a citation context. However, current citation recommendation models still have the defeat of low precision and overfitting (model demonstrates strong performance on the training dataset but fails to generalize effectively to unseen or validation data). In this paper, we propose a hybrid citation recommendation model named as SPECTER-BS which incorporates the SPECTER and author networks to improve citation recommendation quality. Our proposed SPECTER is an innovative method for generating document-level embeddings of scientific papers. It achieves this by pretraining a transformer language model with the robust signal of document-level relatedness found within the citation graph. Author networks take the author’s names and citation count of as input and generate a single bibliographic score for each candidate paper. The combination of a semantic analysis model and a scoring method based on the number of citations of the article promises to improve the quality of citation recommendation models. The experimental results demonstrate that our proposed SPECTER-BS model achieves a 6–10% improvement in standard evaluation metrics—mean reciprocal rank (MRR) and Recall@10—across three widely used benchmark datasets, which are: ACL-200, ACL-600, and RefSeer.</p>