I know I stored it somewhere - Contextual Information and Ranking on our Desktop

1 Motivation Future digital libraries will be distributed, and recent research has already explored some promising approaches focusing on distributed and peer-to-peer search and retrieval architectures, connecting distributed repositories efficiently and transparently. Another aspect, which has been less explored so far, is the role of the implicit personal repositories we all have on our desktops, and the efficiency we expect to gain from storing important resources in these " personal digital libraries ". Ironically, in many cases, moving important resources closer to our workspace results in less retrieval efficiency. Sophisticated web search technology usually allows us to find appropriate documents in a few seconds. Finding these documents on our desktop is surprisingly more difficult, at least if we have been storing documents for a few years or more. This is improving somewhat with the recent crop of desktop search engines, but even with these tools, searching through our (relatively small set of) personal documents with the recent beta of Google Desktop Search is inferior to searching the (rather vast set of) documents on the web with Google. The main reason for this is that one of the distinguishing features of Google-sophisticated ranking using PageRank and other features-is unavailable on our desktop. This position paper explores some of the information we have available on our desktop to extend search beyond simple full-text search, and the algorithms we can build upon this information, implementing efficient ranking of resources on our desktop. Regarding the first aspect, we propose activity-based metadata and relationships as sources of additional information in desktop search. This information in many cases represents contextual information, which is very useful for re-finding resources we already worked with. We will describe corresponding metadata for two selected scenarios. For the second aspect, we will focus on recent advances of PageRank-based ranking, extending models such as the ones describing our contextual information, and show how local and global ranking measures can be integrated in such a model. These two techniques together show considerable promise for extending efficient information access to our desktop, extending globally available information with user-centered activity-based information, and exploiting the unique information background we have available on our desktop. We are currently implementing first prototypes in the context of the open source Beagle project which aims to provide sophisticated desktop search in Linux.