The BAY-HIST Prediction Model for RDF Documents

In real-world RDF documents, property subject and object values are often correlated. The identification of these relationships is of significant relevance to many applications, e.g., query evaluation planning and linking analysis. In this paper we present the BAY-HIST Prediction Model, a combination of Bayesian networks and multidimensional histograms which is able to identify the probability of these dependencies. In general, Bayesian networks assume a small number of discrete values for each of the variables considered in the network. However, in the context of the Semantic Web, variables that represent the concepts in large-sized RDF documents may contain a very large number of values; thus, BAY-HIST implements multidimensional histograms in order to aggregate the data associated with each node in the network. We illustrate the benefits of applying BAY-HIST to the problem of query selectivity estimation as part of costbased query optimization. We report initial experimental results on the predictive capability of this model and the effectiveness of our optimization techniques when used together with BAY-HIST. The results suggest that the quality of the optimal evaluation plan has improved over the plan identified by existing cost models that assume independence and uniform distribution of the data values.