Use of multiple information problem representations for information retrieval

This paper discusses the general issues of the effect and use of multiple representations of the same information problem, or topic, on information retrieval system performance. It has been known for some time that different representations of the same information problem retrieve different sets (or lists) of documents, both relevant and non-relevant. More recently, there have been a number of studies investigating the effect of combining, in various ways, such different representations, in order to try to get a single response from the information retrieval system which is better than that for any of the single representations. Here, we discuss the rationale, both empirical and theoretical, for such an approach, and survey the results of recent research projects in this area. Briefly, all of them demonstrate the same phenomenon; the more representations one can combine, the better the retrieval performance. Implications of these results for IR system design, and IR practice, are discussed.