Non-exact matching

Abstract

We suggest a method that utilizes relations of relevance between attribute-value pairs to derive importance measures on the attribute-value pairs describing a target object. The intention with using weighted attribute-value pairs is to improve information retrieval by giving a search process an opportunity to select or reject attribute-vectors describing source targets stored in a knowledge-base. The method presupposes a matching process that considers the relative importance of attribute-value pairs when it compares two attribute vectors. Such a matching process can dynamically establish this relative importance of attributes-value pairs by letting some of a specific vector's attribute-values assign weights of importance to other attributes-value pairs in the vector. The assigned weights reflects the importance of the attribute-value pairs in relation to some desired analogy or similarity. We illustrate this weight assignment process with the help of some examples. The examples indicate that the performance of the matching process can be improved when it utilizes weighted attribute-value pairs. Especially when the similarity between attribute vectors depends on the problem solving context. The method has been tested in real life domains and was found to perform satisfactory.