A Knowledge-driven Approach for Personalized Literature Recommendation Based on Deep Semantic Discrimination

The query and selection of scientific literatures are knowledge driven. Researchers regard public literature resources as target knowledge sources and use their own domain knowledge to explore in them. However, existing knowledge-driven methods of literature recommendation mainly focus on morphological matching and cannot effectively resolve polysemous phenomenon brought by "knowledge overload". Based on this observation, this paper presents a knowledge-driven approach for personalized literature recommendation. Domain ontology, synonyms and knowledge labels are integrated into a multidimensional domain knowledge map for modeling user knowledge requirements and literature contents based on deep semantic discrimination. The personalized recommendation is achieved by calculating knowledge distances between users and literatures. Experimental results on a real data set of PubMed show that the recommended relevance of the current method is 67%, better than other methods.