Similarity function recommender service using incremental user knowledge acquisition

Abstract

Similar entity search is the task of identifying entities that most closely resemble a given entity (e.g., a person, a document, or an image). Although many techniques for estimating similarity have been proposed in the past, little work has been done on the question of which of the presented techniques are most suitable for a given similarity analysis task. Knowing the right similarity function is important as the task is highly domain- and data-dependent. In this paper, we propose a recommender service that suggests which similarity functions (e.g., edit distance or jaccard similarity) should be used for measuring the similarity between two entities. We introduce the notion of "similarity function recommendation rule" that captures user knowledge about similarity functions and their usage contexts. We also present an incremental knowledge acquisition technique for building and maintaining a set of similarity function recommendation rules. © 2011 Springer-Verlag.