On the Notion of “An Information Need” Eduard Hoenkamp University of Maastricht hoenkamp@acm.org Abstract. ‘Information need’ is a notion in IR that is ubiquitous, im- portant, and intuitively clear. So far, surprisingly, the term seems to have defied formal definition. Of course, IR can continue to prosper without a formalization of ‘information need’. Yet when a field gets more mature there comes a time that frequently used notions should be formalized to make them susceptible of scrutiny. For IR such formalization should (1) be independent of a particular query language or document model, (2) allow that users formulate a need for information that may be unavail- able or even nonexistent, and (3) allow that users try to circumscribe the very information they do not possess. To this end, the paper uses lattice theory to define a ‘formal information need’, which, we argue, coincides with the intuitive notion precisely when a user’s need for information can actually be filled. 1 Introduction When we see ourselves or others look up information on the world wide web, we can make three observations: First, people can use the same query but may be looking for different information. Second, they may be looking for the same information but use different queries. And third, they usually refine their queries to zoom in on their information need. A definition of that term, however, is quite elusive. It is something we hold in our heads and of which it is not clear that it can ever be fully satisfied. What is clear, though, is that information needs can often be discerned by the different queries that are issued, as well as by the different documents people subsequently select. That is, queries and document sets seem to act as dual descriptions of the same information need. It is difficult, however, to maintain that the two descriptions are completely isomorphic. Hence we will use a relation that is sufficient yet weaker than isomorphism, called a Galois Connection, for which we consider queries and relevant documents as a pair instead of choosing one or the other in isolation. 2 Mapping Queries to Documents and Back The formalization that follows was inspired by formal concept analysis (FCA), and so perhaps an example from that subject is helpful. FCA is an approach to meaning representation that describes concepts as a binary relation between ‘objects’ and ‘attributes’ [1]. For example, a horse can L. Azzopardi et al. (Eds.): ICTIR 2009, LNCS 5766, pp. 354–357, 2009. c  Springer-Verlag Berlin Heidelberg 2009