Improving On-line Scientific Resource Profiling by Exploiting Resource Citation Information in the Literature

We study the task of on-line scientific resource profiling, which aims at better understanding and summarizing on-line scientific resources to promote resource search and recommendation systems. To this end we propose to exploit the resource citation information in scientific literature by extracting the fine-grained relations between the cited on-line resources and other resource-related scientific terms. In this paper we create a dataset (SciResTR) and develop a framework (SciResTR-IE) which jointly extracts all the related scientific terms and the resource-term relations. Extensive experiments demonstrate that our framework outperforms other baselines significantly, by around 5% in scientific information extraction tasks absolutely. We further show that our proposed system can automatically construct several on-line-resource-centered networks from a large corpus of scientific articles, which is a first step towards utilizing resource citation information in the literature to improve on-line scientific resource profiling.