Academic Social Network-Based Recommendation Approach for Knowledge Sharing

Academic information overload has brought researchers great difficulty due to the rapid growth of scientific articles. Methods have been proposed to help professional readers find relevant articles on the basis of their publications. Although effectively sharing publications is essential to spreading knowledge and ideas, few studies have focused on knowledge sharing from an author perspective. This study leverages the online academic social network to propose a recommendation approach for knowledge sharing. In our approach, we integrate researcher-level and document-level analyses in the same model. Our model works in two stages: 1) researcher-level analysis and 2) document-level analysis. The former combines research topic relevance, social relations, and research quality dimension, and the latter uses the machine learning method to learn the vector representation for each word. Online social behavior information is also leveraged to enhance readers' short-term interests. Our approach is deployed in ScholarMate, a prevalent academic social network. Compared with other baseline methods (CB, LDA, and part of the proposed approach), our approach significantly improves the accuracy of recommendations. Moreover, our method can disseminate papers efficiently to readers who have no publications.