Srvycite: a hybrid scientific article recommendation system

<span>A recommendation system is becoming part of every work done today to reduce the effort of work done by the users in searching for items in need by recommending new items that may be useful. This theme has also been used in research article recommendation systems for recommending articles of interest to researchers from a bulk of digital research documents spread across different databases on the internet. To ease the task of this article recommendation process, we have proposed a novel approach, Srvycite, by utilizing the survey article citation network along with the original research article network. The purpose of utilizing the survey article citation network is to detect the most influential articles that are considered to be important by other researchers in the same field. The Srvycite approach utilizes the text and meta features of articles to recommend papers. To preprocess the text features utilized, we have employed Word2Vec and bidirectional encoder representations from transformers (BERT) for vectorization. Then citation graph and survey citation graphs are generated to find the most influential nodes. The weighted text similarity score is finally computed by combining the cited by values and the text similarity score from the citation and survey citation graph to list articles as recommendations for the user. This system is proven to increase the accuracy of the article recommendation by 3.8 and 2.1 in the case of the precision and recall measures for performance evaluation.</span>