asystematicapproach.PDFVIP

Processing complex similarity queries: a systematic approach? Anna Yarygina1, Boris Novikov1, and Natalia Vassilieva1,2 1 Saint-Petersburg University anya safonova@mail.ru, borisnov@ 2 HPLabs nvassilieva@ Abstract. In this paper we introduce a set of operations for complex manipulation with approximate queries represented by scoring functions. We show correspondence of these operations to the user expectations and applications needs. We evaluate the e?ectiveness of our techniques in the context of content-based image retrieval task and compare it to the known fusion methods. 1 Introduction Both amount and complexity of stored data rapidly grow. The information is presented in the form of complex objects with explicit or implicit structure. Ex- amples may include complex hypermedia objects containing text, images, sound, and video, and dynamic pages typically generated from structured data with at- tributes of di?erent types. Neither navigational link chasing nor simple keyword queries cannot be su?cient for advanced search or complex object retrieval. The reasons to evaluate complex structured queries are: – A need to combine search criteria for di?erent types of information; – A query re?nement, e.g. based on user pro?le or feedback; – Advanced users may need query structuring. Database query languages, e.g. relational, are nearly perfect when an ex- act result is expected. However, in the world of similarity-based queries with approximate results the behavior of logical operators does not always meet hu- man expectations. For example, logical operators cannot capture the expected increase in con?dence due to multiple sources of support. Approximate similarity-based search results are usually represented by scores of objects. A concept of data fusion is suitable to combine r