The most general approach is to develop a service suitable for any open archive. This could be done using a variety of small open source modules and new code.
The most integrated approach would be to extend MARIAN's ability for handling ETDs with SDI capabilities, similar to those of the SIFT package from Stanford. We already have many "hooks" to make this relatively easy.
The third approach, involves extending / enhancing / customising SIFT (Stanford Information Filtering Tool) beyond a project done in fall 1998 so that the "documents" used for matching against profiles are new ETDs entering the system. These could either be just the "HTML front pages" (containing abstract and other "metadata"), or the complete text extracted from the ETD itself. SIFT has already been tailored to filter Usenet news, and so it should be possible to change it to work with ETD data. Students would get experience in applying vector and boolean retrieval models, working with data, and some CGI experience.
Finally, and probably the easiest/best approach, we may be able to use NCSTRL for this purpose. MIT and the National Library of Portugal are both using NCSTRL software for ETDs. The NCSTRL service supports profiles and a subscription service We could adapt the Dienst code used in NCSTRL for SDI.