A comparison of metadata extraction techniques for crowdsourced bibliographic metadata management

Abstract

Social research networks such as Mendeley and CiteULike offer various services for collaboratively managing bibliographic metadata and uploading textual artifacts. One core problem thereby is the extraction of bibliographic metadata from the textual artifacts. Our work investiages the use of Conditional Random Fields and Support Vector Machines, implemented in two state-of-the-art real-world systems, namely ParsCit and the Mendeley Desktop, for automatically extracting bibliographic metadata. We compare the systems' accuracy on two newly created real-world data sets gathered from Mendeley and Linked-Open-Data repositories. Our analysis shows that two-stage SVMs provide reasonable performance in solving the challenge of metadata extraction from user-provided textual artifacts. © 2012 Authors.