Reference Information Extraction and Processing Using Conditional Random Fields

Fostering both the creation and the linking of data with the scope of supporting the growth of the Linked Data Web requires us to improve the acquisition and extraction mechanisms of the underlying semantic metadata. This is particularly important for the scientific publishing domain, where currently most of the datasets are being created in an author-driven, manual manner. In addition, such datasets capture only fragments of the complete metadata, omitting usually, important elements such as the references, although they represent valuable information. In this paper we present an approach that aims at dealing with this aspect of extraction and processing of reference information. The experimental evaluation shows that, currently, our solution handles very well diverse types of reference format, thus making it usable for, or adaptable to, any area of scientific publishing.