FREE-TEXT RETRIEVAL-SYSTEMS - RESEARCH-AND-DEVELOPMENT IN INFORMATION-RETRIEVAL

A majority of text retrieval systems which exist today are based upon the Boolean logic: AND, OR, and NOT. Despite its commercial popularity Boolean retrieval has several obvious limitations. The alternative NN searching method overcomes some of these problems, for example, the d ifficulty of query formulation, but it also has some disadvantages. Ac cordingly, there have been several attempts to merge the two approache s into hybrid retrieval models. However, these methods are theoretical ly unsound and researchers are now increasingly considering the use of Multi-Strategy retrieval systems in which a range of techniques are a vailable. Thus, if one strategy failed to give an optimum performance for a particular query the alternative strategies could be used. Exper iments were conducted in an operational environment with the aim of de signing a Multi-Strategy retrieval system. Unlike the other studies, h ere the two chosen search strategies were based on entirely different concepts in that one produces ranked output and the other produces unr anked output. The series of tests carried out to establish the differe nce in retrieval effectiveness of these two different search systems, and to see whether it is statistically significant, yielded no signifi cant results. In general, the statistical criteria considered produced no reliable mark to reveal the true nature of relationship, if there was any, of query characteristics on the retrieval performance. Althou gh it was not possible to derive a more continuous measure of relative effectiveness in operational systems, because of the non-availability of information on the total number of relevant items in the collectio n, it could be concluded that largely the findings were in agreement w ith the previous conclusions. Unlike the retrieval mechanisms consider ed here, most of the techniques retrieve different sets of documents o n varied document representations despite the comparable performance. So, it might be possible to choose automatically any two retrieval mec hanisms and provide the user retrieval output of both together, exclud ing the common items, but the cost factor would be increased. It is un likely to satisfy the users otherwise without sacrificing the cost fac tor, while considering different levels of search satisfaction, and vi ce versa.