Evaluation of information retrieval systems using fuzzy set techniques

In this thesis we explore an area that has only been touched upon until now--namely, the area of non-binary evaluation of information retrieval systems. We first explore it in terms of the traditional measures of R and P. In this part of the analysis, we describe carefully how to calculate R and P in the non-binary context, something that has never been explicitly addressed before in the literature. We suggest two separate methods of calculating R and P, and explain the differences between the two methods.; We then take the analysis further into the area of composite non-binary evaluation of information systems. We use fuzzy set techniques to address this problem and develop four non-binary composite measures. The first measure is an extension of a suggestion by Voiskunskii to use a cos measure in the binary case to calculate to what degree two vectors coincide with each other. However, he shows that his suggestion is not extendible to the non-binary case. We use fuzzy set techniques and successfully make the transition to the non-binary case. The others are based on the fuzzy set technique of subsethood. We develop three measures based upon this concept. The measures are fully explored and contrasted in terms of their characteristics and advantages and disadvantages, especially in terms of the important order preservation property characteristic.