Optimized partitioned binary interpolative coding

Inverted index is the core data structure in large scale information retrieval systems such as Web search engine. Index compression techniques are usually used to reduce the storage and transmission time from disk to memory. Many index compression schemes have been proposed and among them Binary Interpolative Coding (IPC) is one of the most widely used schemes due to its superior compression ratio(CR). However we find the bottleneck of CR of IPC is the existing of over-long d-gaps in the inverted lists. To avoid them and to strengthen the advantage of CR of IPC we propose two improvement schemes including an optimized one. We discuss our method in detail and also introduce some important implementation details. Experimental results on two publicly available standard corpora show that compared with normal IPC, our method reduces the amount of storage by 13% with a negligible time cost incensement of less than 1% in Boolean AND, WAND and Rank queries.