Although I have been using citeseer as a valueable bibliographic resource (probably even the most valuable) for a couple of years now. I just recently stumbled upon the citeseer directory, a web directory for computer science literature. The major clou is that the topical clusters (e.g. Information Retrieval, Machine Learning) can be sorted according to their HITS scores (Hub, Authority). The directory is especially useful for finding good surveys about certain topics. The major drawback however is the fact that the HITS algorithm favors mature items (because those are more likely to have been extensivly cited). Although a discounted authority score (in respect to the publishing date) is offered, the directory is not that valuable in finding recent authoritive items.
The directory can be found at http://citeseer.ist.psu.edu/directory.html
In the course of my thesis - dealing with use of Relevance Feedback (RF)in Information Retrieval Systems - I'm going to maintain my reference list with citeulike. All papers in the context of RF will be tagged with relevancy and feedback, resp.