 Hello, my name is Richard Urban. I'm a doctoral student at the Graduate School of Library and Information Science at the University of Illinois. I'm going to talk today about the DLF-DCC Beta Sprint and Linked Open Data. Currently, many of the collections included in IMLS digital collections and content project are disconnected from the larger Linked Open Data community, even though many institutions share OAI records through the Open Archives Initiative Protocol for Metadata Harvesting. As part of our sprint, we looked at the feasibility of integrating IMLS-DCC and its partners into this larger world of Linked Data, in particular by looking at how we might reconcile some of our metadata with available Linked Data sources. I'm going to use the five criteria put forth by Tim Berners-Lee as a way to talk about the opportunities and challenges that we face. For example, all of this material is available on the web, but we know very little about what the licenses are for this metadata, including metadata that we create for the IMLS-DCC project itself. According to a 2006 study, only about 20% of OAI repositories included something called a set description, where a repository might specify what its license is. It's likely that an even smaller percentage of those indicated a license, an even smaller percentage used an open license. This seems to be a big sticking point in moving the community towards Linked Data. However, I believe that IMLS-DCC is well placed and to provide some leadership in this area, given our past guidance on metadata quality and other forms of aggregation. Currently, material is available as machine-readable structured data in a non-proprietary format, namely Dublin Core XML, either in a simple variety or as a more advanced IMLS-DCC collection description, which is based on the Dublin Core Collection Application Profile. We also republish about one million records as simple and qualified Dublin Core that have been harvested from participating repositories. For each of these records, there's an OAI ID that we've also used as an identifier in the URL pattern for our DPLA prototype. So if you know an OAI ID, you can construct a URL for items and collections. Identifiers for images that come from partners depend on the kinds of systems that they've implemented. So for example, if they've used Content DM, the resource identifier will depend on how Content DM has specified it. We also have another little tool called ThumbGrabber that will grab a thumbnail if an image is available. We've used the OAI ID to also construct the URL for those thumbnails. We're also exploring methods for sharing RDF serializations of this OAI data, either as a Sparkle endpoint or as RDFA that is built into the interface itself. A big challenge, of course, was linking our data to current sources. One of the things I found early on is that there's not much information about institutions themselves. Someone has kindly uploaded the public library survey to Freebase, so if I have information about a public library in our collection, it was pretty easy to reconcile that with available linked data information. However, other kinds of archives, libraries, and museums were not as well represented. One of the things we might do as a DPLA community is at least make sure that there's a linked data representation for each of these institutions. I was also able to extract controlled vocabulary terms from collection-level metadata and reconcile it with linked data sources. For example, our item type uses this RS of graphic materials terms, and I was able to reconcile that with the TALIS repository from the Library of Congress. Using Google Refine, I was able to align other terms through Freebase, for example, a mirror of the Library of Congress subject headings, or aligning it with non-library vocabularies like Freebase locations. One of the things, of course, that we've been saying all along is that quality metadata is important. It becomes particularly important in this process for well-controlled and well-defined terms, reconciling to linked open data is pretty trivial. But at the item level, we have a lot of uncontrolled terms and various formats of metadata that's going to make it much more challenging. More information about this is going to be in our full report that you can read as part of our submission.