The Information Retrieval Intelligent Assistant (IRIA) project applies principles of memory retrieval from cognitive science to the problem of information retrieval from large heterogeneous databases. IRIA uses spreading activation over a semantic network for information retrieval, a technique which has proven effective in a variety of tasks. However, some of the very features which motivated the choice of spreading activation for information retrieval — such the use of fanout to automatically compute term weights, or the use of thresholds to automatically limit computation spent on irrelevant items — can introduce new problems as systems are scaled to larger sizes.
This paper discusses the use of semantic networks and spreading activation for information retrieval in the context of the IRIA approach, reviews some of the problems that arise as these technologies are scaled up to production systems, presents some preliminary results that illustrate these problems in practice, and discusses potential solutions.
Read the paper: