Positional Data Organization and Compression in Web Inverted Indexes

Abstract

To sustain the tremendous workloads they suffer on a daily basis, Web search engines employ highly compressed data structures known as inverted indexes. Previous works demonstrated that organizing the inverted lists of the index in individual blocks of postings leads to significant efficiency improvements. Moreover, the recent literature has shown that the current state-of-the-art compression strategies such as PForDelta and VSEncoding perform well when used to encode the lists docIDs. In this paper we examine their performance when used to compress the positional values. We expose their drawbacks and we introduce PFBC, a simple yet efficient encoding scheme, which encodes the positional data of an inverted list block by using a fixed number of bits. PFBC allows direct access to the required data by avoiding costly look-ups and unnecessary information decoding, achieving several times faster positions decompression than the state-of-the-art approaches.