A Compositional Relevance Model for Adaptive Information Retrieval

Abstract

There is a growing need for rapid and effective access to information in large electronic documentation systems. Access can be facilitated if information relevant in the current problem solving context can be automatically supplied to the user. This includes information relevant to particular user profiles, tasks being performed, and problems being solved. However most of this knowledge on contextual relevance is not found within the contents of documents, and current hypermedia tools do not provide any easy mechanism to let users add this knowledge to their documents. We propose a compositional relevance network to automatically acquire the context in which previous information was found relevant. The model records information on the relevance of references based on user feedback for specific queries and contexts. It also generalizes such information to derive relevant references for similar queries and contexts. This model lets users filter information by context of relevance, build personalized views of documents over time, and share their views with other users. It also applies to any type of multimedia information. Compared to other approaches, it is less costly and doesn't require any a priori statistical computation, nor an extended training period. It is currently being implemented into the Computer Integrated Documentation system which enables integration of various technical documents in a hypertext framework.