Finding superior skyline points for multidimensional recommendation applications

Abstract

In a typical Web recommendation system, objects are often described by many attributes. It also needs to serve many users with a diversified range of preferences. In other words, it must be capable to efficiently support high dimensional preference queries that allow the user to explore the data space effectively without imposing specific preference weightings for each dimension. The skyline query, which can produce a set of objects guaranteed to contain all top ranked objects for any linear attribute preference combination, has been proposed to support this type of recommendation applications. However, it suffers from the problem known as 'dimensionality curse' as the size of skyline query result set can grow exponentially 34 World Wide Web (2012) 15:33-60 with the number of dimensions. Therefore, when the dimensionality is high, a large percentage of objects can become skyline points. This problem makes such a recommendation system less usable for users. In this paper, we propose a stronger type of skyline query, called core skyline query, that adopts a new quality measure called vertical dominance to return only an interesting subset of the traditional skyline points. An efficient query processing method is proposed to find core skyline points using a novel indexing structure called Linked Multiple B'-trees (LMB). Our approach can find such superior skyline points progressively without the need of computing the entire set of skyline points first.

Landslide Susceptibility Assessment using Skyline Operator and Majority Voting

“…Skyline operator is concerned with retrieving objects, called skyline points, from a set of objects such that the retrieved objects are not dominated by any other object in the set. The skyline operator is implemented in domains such as recommendation [25,26], scientometrics [27,28].…”

Progressive ranking based on a dominance list

“…The main idea consists in introducing more comparability by defining other, mostly weaker, dominance relations. To name a few : -dominance [5], K-dominance [2], dominanceback [6] and quasi-dominance [3]. The relevance of the different dominance relations is obviously disputable and depends on the context and/or the user.…”