Probabilistic similarity join on uncertain data

Abstract

Abstract. An important database primitive for commonly used feature databases is the similarity join. It combines two datasets based on some similarity predicate into one set such that the new set contains pairs of objects of the two original sets. In many different application areas, e.g. sensor databases, location based services or face recognition systems, distances between objects have to be computed based on vague and uncertain data. In this paper, we propose to express the similarity between two uncertain objects by probability density functions which assign a probability value to each possible distance value. By integrating these probabilistic distance functions directly into the join algorithms the full information provided by these functions is exploited. The resulting probabilistic similarity join assigns to each object pair a probability value indicating the likelihood that the object pair belongs to the result set. As the computation of these probability values is very expensive, we introduce an efficient join processing strategy exemplarily for the distance-range join. In a detailed experimental evaluation, we demonstrate the benefits of our probabilistic similarity join. The experiments show that we can achieve high quality join results with rather low computational cost. 1