Experimental Comparison of Set Intersection Algorithms for Inverted Indexing

The set intersection problem is one of the main problems in document retrieval. Query consists of two keywords, and for each of keyword we have a sorted set of document IDs containing it. The goal is to retrieve the set of document IDs containing both keywords. We perform an experimental comparison of Galloping search and a new algorithm by Cohen and Porat (LATIN2010), which has a better theoretical time complexity. We show that the new algorithm has often worse performance than the trivial one on real data. We also propose a variant of the Cohen and Porat algorithm with a similar complexity but better empirical performance. Finally, we investigate influence of document ordering on query time.