uγ-Golomb: A new Golomb Code Variation for the Local Bernoulli Model

Abstract

Several compression codes exist today that have been developed to reduce the size of inverted file indexes used widely in information retrieval implementations targeting text databases. In the present study, we examine some of the most significant integer compression codes and propose uγ − Golomb, a variation of the most popular scheme: the Golomb code for the local Bernoulli model. As a test-bed, we utilize text databases from the TREC collection that take up almost 1GB on disk. The proposed new variation does not introduce any additional computational overhead when it is compared to the original compression code. With regard to storage utilization efficiency, experimental results reveal a considerable improvement over the original compression code in the case of non-uniform text databases.