A Study of Four Index Structuresfor Set - Valued

We review and study the performance of four diierent index structures for indexing set-valued attributes designed to speed up set equality, subset and superset queries. All index structures are based on traditional techniques, namely signatures and inverted les. More speciically, we consider sequential signature les, signature trees, extendible signature hashing, and a B-tree based implementation of inverted lists. The latter is reened by a compression scheme in order to keep space requirements within acceptable bounds. The performance study is based on real implementations subjected to a benchmark accounting for diierent set sizes, domain sizes, and data distributions (uniform and skewed).