Double distance-calculation-pruning for similarity search
MetadataShow full item record
Many modern applications deal with complex data, where retrieval by similarity plays an important role. Complex data main comparison mechanisms are based on similarity predicates. They are usually immersed in metric spaces where distance functions are employed to express the similarity and a lower bound property is usually employed to prevent distance calculations. Retrieval by similarity is implemented by unary and binary operators. Most of the studies aimed at improving the efficiency of unary operators, either by using metric access methods or mathematical properties to prune parts of the search space during query answering. Studies on binary operators to solve similarity joins aim to improve efficiency and most of them use only the metric lower bound property for pruning. However, they are dependent on the query parameters, such as the range radius. In this paper, we propose a generic concept that uses both lower and upper bound properties based on the Metric Spaces Theory to increase the avoidance of element comparisons. The concept can be applied on any existing similarity retrieval method. We analyzed the prunability power increase and show an example of its application on classical join nested loops algorithms. Practical evaluation over both synthetic and real data sets shows that our method reduced the number of distance evaluations on similarity joins.