A Classification Approach to Boolean Query Reformulation

One of the difficulties in using current Boolean-based eration of networks and online databases makes it possible information retrieval systems is that it is hard for a user, to disseminate and retrieve information in fast and easy especially a novice, to formulate an effective Boolean ways. Information updates are often simpler and cheaper query. Query reformulation can be even more difficult in electronic form than on paper. One might think that and complex than formulation since users often have electronic access would provide all the necessary infordifficulty incorporating the new information gained from the previous search into the next query. In this article, mation from a large database in a short time with little query reformulation is viewed as a classification probeffort. However, consider retrieving relevant articles from lem, that is, classifying documents as either relevant or a collection of millions of documents. A system with low nonrelevant. A new reformulation algorithm is proposed performance can easily overload the user with unneceswhich builds a tree-structured classifier, called a query tree, at each reformulation from a set of feedback docusary information or lose potentially valuable information. ments retrieved from the previous search. The query tree What most users want is quality output, not quantity. The can easily be transformed into a Boolean query. The fact is that finding the right information gets harder as query tree is compared to two query reformulation algomore information becomes available, and the big question rithms on benchmark test sets (CACM, CISI, and Medis how the user can retrieve only the information he relars) . In most experiments, the query tree showed significant improvements in precision over the two algoquested. rithms compared in this study. We attribute this The most prevalent information retrieval (IR) model improved performance to the ability of the query tree today is the Boolean retrieval model. A query is a Boolean algorithm to select good search terms and to represent expression and, in the simplest case, a search is based on the relationships among search terms into a tree structure. exact matching.