Abstract:
The ever-growing number of scientific publications on the Internet presents a significant
challenge for researchers and academia to efficiently discover relevant scientific papers
and stay up-to-date with the latest research in their respective fields.In response to this
challenge, recommender systems have emerged as a contemporary technology that offers
personalized recommendations tailored to users’ specific interests.
Among the various techniques for scientific paper recommendation, context-aware
citation recommendation has garnered considerable attention. This approach aims to
provide users with curated lists of high-quality candidate papers based on the analysis of
citation contexts. To address the complexities of this task, we leverage the CiteULike-a
dataset,a well-known scholarly article recommendation system, to train and evaluate our
model. Our implementation followed an information retrieval approach, comprising a retrieval model,ranking model,and post-ranking model. We utilize the BERT model based
architecture, to learn representations of both papers and citation contexts.In addition to
GRU ,to model sequential data. By incorporating citation relationships,we enhance
the distributed vector representations to capture important semantic and contextual information. To evaluate the performance of our model,we employ various evaluation
metrics, with accuracy being the primary measure.We assess the accuracy scores for
different values of top-k papers,representing the number of relevant papers considered in
the retrieval process. Our experiments revealed interesting trends,indicating that as the
number of top relevant papers increased, the accuracy score exhibited a gradual increase.
Our context-aware citation recommendation model, demonstrated promising results.By
incorporating advanced techniques and optimizing computational resources,we achieve
improved accuracy in recommending relevant scientific papers which estimated by 60
%. Our findings contribute to the advancement of knowledge in the field of information retrieval for scholarly papers, and future work could focus on exploring additional
enhancements and incorporating user feedback to further enhance the system’s performance
