The authors describe an expert system to assist online searching through automatic query reformulation. They provide a good description of the system architecture and its components. The expert system sits between, and controls, a search and retrieval engine (which is connected to the textbase), a thesaurus, and the user interface. Portability is facilitated by incorporating the topic domain knowledge in the thesaurus, while the knowledge of search tactics and reformulation techniques is held within the expert system. The expert system will automatically reformulate (broaden or narrow) a Boolean query depending on whether too few or too many items are retrieved by the original query. The reformulation techniques used are the expansion of concepts, adjustment of textual context (such as concepts occurring in the same sentence, paragraph, and so on), and changing the Boolean operators. Retrieved items are then ranked on the basis of relative term frequency and adjacency.

The paper goes on to describe a comparative evaluation of the expert system against a traditional Boolean interface and one with access to a thesaurus. This experiment showed marginal improvements in effectiveness, and the expert system reduced the number of queries required, although retrieval took longer. Meaningful evaluations of this type in the field of information retrieval are difficult to achieve, however. The use of small numbers of inexperienced users who make judgments of relevance based on an artificial information need is fraught with problems.

The paper as a whole is clearly written and makes a useful contribution to an area of practical importance. My one remaining doubt is the apparent lack of control that the user has over the reformulation. Most users would probably prefer a more interactive facility.