Explanatory search

This post falls into the "using your blog as a bookmark menu" category. Via Gene Golovchinsky's post, "Search is not magic", comes a reference to Jeremy Picken's concept of explanatory search (as a complement to exploratory search). The idea, as elaborated by Gene, is that it is not enough for the search engine to be a black-box, magical device which produces answers for reasons and using methods that are inscrutable to the user; the user needs to have a model (exact or approximate) of how these results were obtained. Without such a model, it is difficult for the user to know how to develop their search and improve their results, and similarly difficult to estimate how comprehensive the search might be and what other material might be out there. As Gene points out, this is one of the advantages of Boolean queries: how the query works is easily understood by the user, and so what effect changes to the query might have is readily predictable.

I've mentioned in previous posts (and it is not an idea original to me) that the predominant model in information retrieval research has been a black-box one. Put another way, traditional information retrieval has been a descriptive science rather than an explanatory or predictive one; that is, we have developed very sophisiticated and specialised methods of measuring how well an IR system performs, but quite weak models of why it produces the results it does. The effect on the field has been a lot of heuristic tinkering to produce optimised (by some metric) results; the extreme outcome of this process is the Learning to Rank movement. I'd always thought of the desire to move beyond merely descriptive science to an explanatory and predictive science as being primarily a desire for scientific maturity; but Jeremy and Gene suggest that it has a strong practical analogue in the need to make the retrieval process understandable, predictable, and manipulable by the user.

Leave a Reply