Semantic User Profiles: Learning Scholars' Competences by Analyzing Their Publications

摘要

Semantic publishing generally targets the enhancement of scientific artifacts, such as articles and datasets, with semantic metadata. However, smarter scholarly applications also require a better model of their users, in order to understand their interests, tasks, and competences. These are generally captured in so-called user profiles. We investigate a number of existing linked open data (LOD) vocabularies and propose a description of scientists' competences in LOD format. To avoid the cold start problem, we suggest to automatically populate these profiles based on the publications (co-)authored by users, which we hypothesize reflect their research competences. Towards this end, we developed the first complete, automated workflow for generating semantic user profiles by analyzing full-text research articles through natural language processing. We evaluated our system with a user study on ten researchers from two different groups, resulting in mean average precision (MAP) of up to 92%. We also analyze the impact of semantic zoning of research articles on the accuracy of the resulting profiles. Finally, we demonstrate how these semantic user profiles can be applied in a number of use cases, including article ranking for personalized search and finding scientists competent in a topic - e.g., to find reviewers for a paper.