An Architecture for Automatic Reference Linking

Along with the explosive growth of the Web has come a great increase in on-line scholarly literature. More and more researchers are turning to their computers for the latest results in their field. Not only is Web retrieval usually faster than a walk to the library, but the information obtained from the Web is generally more current than what appears in printed publications. The increasing proportion of on-line scholarly literature makes it possible to implement functionality desirable to all researchers – the ability to access cited documents immediately from the citing paper. Implementing this direct access is called “reference linking”. While many authors insert explicit hyperlinks into their papers to support reference linking, it is by no means a universal practice. The approach taken by the Digital Library Research Group at Cornell employs value-added surrogates as a generalizable mechanism for providing reference-linking behavior in Web documents. Given the URL of an on-line document, a surrogate object is constructed for that paper, which then processes document’s content to extract reference linking data. The surrogate then exposes the reference linking data through a well-defined API, permitting the construction of reference linking services by external clients. We present two examples of the many possible reference linking applications buildable on this API. We also describe a metric that measures the API’s performance; currently we are (automatically) extracting reference linking information from HTML papers with more than 80% accuracy.