Approximate Similarity Search from another ”Perspective” (EXTENDED ABSTRACT) ⋆ Giuseppe Amato and Pasquale Savino ISTI-CNR, Pisa, Italy firstname.lastname@isti.cnr.it 1 Introduction We propose a new approach to perform approximate similarity search in metric spaces [8]. The idea at the basis of this technique is that when two objects are very close one to each other they ’see’ the world around them in the same way. Accordingly, we can use a measure of dissimilarity between the view of the world, from the perspective of the two objects, in place of the distance function of the underlying metric space. To exploit this idea we represent each object of a dataset by the ordering of a number of reference objects of the metric space according to their distance from the object itself. In order to compare two objects of the dataset we compare the two corresponding orderings of the reference objects. We show that efficient and effective approximate similarity searching can be obtained by using inverted files, relying on this idea. We also show that the proposed approach performs better than other approaches proposed in literature. 2 Perspective based space transformation Let D be a domain of objects and d : D×D → R be a metric distance function between objects of D. Let RO ⊂D, be a set of reference objects chosen from D. Given an object o ∈D, we represent it as the ordering of the reference objects RO according to their distance d from o. More formally, an object o ∈D is represented with ¯ o = O RO d,o , where O RO d,o is the ordered list containing all objects of RO, ordered according to their distance d from o. We denote the position in O RO d,o of a reference object ro i ∈ RO as O RO d,o (ro i ). For example, if O RO d,o (ro i )=3,ro i is the 3rd nearest object to o among those in RO. We call ¯ D the domain of the transformed objects. ∀o ∈D, ¯ o ∈ ¯ D. Figure 1 exemplifies the transformation process. Figure 1a) sketches a number of reference objects (black points), data objects (white points), and a query object (gray point). Figure 1b) shows the encoding of the data objects in the transformed space. We will use this as a running example throughout the reminder of the paper. As we anticipated before, we assume that if two objects are very close one to each other, they have a similar view of the space. This means that also the orderings of the ⋆ This work was partially supported by the DELOS NoE and the Multimatch project, funded by the European Commission under FP6 (Sixth Framework Programme).