ISIS: A New Approach for Efficient Similarity Search in Sparse Databases

High-dimensional sparse data is prevalent in many real-life applications. In this paper, we propose a novel index structure for accelerating similarity search in high-dimensional sparse databases, named ISIS, which stands for Indexing Sparse databases using Inverted fileS. ISIS clusters a dataset and converts the original high-dimensional space into a new space where each dimension represents a cluster; furthermore, the key values in the new space are used by Inverted-files indexes. We also propose an extension of ISIS, named ISIS, which partitions the data space into lower dimensional subspaces and clusters the data within each subspace. Extensive experimental study demonstrates the superiority of our approaches in high-dimensional sparse databases.