The Open Archives Initiative is an organization dedicated to making digital libraries interoperable and has developed a standard way to access the data stored in a single archive. We want to exploit this standardization to develop a recommendation service that is not tied in with any particular archive.
Recommender services are when a digital library presents suggestions to a user for what that user may be interested in, based on previous behavior and assumptions the system can make about the user. Amazon is the largest online bookseller and makes extensive use of such a service to recommend books to returning shoppers.
This project is to develop a recommender service that processes a stream of user interactions with the system and can produce a list of recommendations on demand at any point in time. The stream of user interactions will be issued using an OAI-like protocol (a search engine with another OAI-like interface will be provided for requesting similar documents). Recommendations are to be requested through a third OAI interface. Using OAI-like interfaces wherever appropriate will result in significantly decreasing the learning curve for adoption of such a service.
The main research problem is to determine how to make recommendations, given the limitations due to accessing the data through only a simple well-defined interface. A sliding window over recent documents accessed ? Weighted ? How much processing should be done, when and where - and what should the system do if a user provides negative or positive feedback about a recommendation ?