Index compression for information retrieval systems
Roi Blanco González · 2008
Given the increasing amount of information that is available today, there is a clear need for Information Retrieval (IR) systems that can process this information in an efficient and effective way. Efficient processing means minimising the amount of time and space required to process data, whereas effective processing means identifying accurately which information is relevant to the user and which is not. Traditionally, efficiency and effectiveness are at opposite ends (what is beneficial to efficiency is usually harmful to effectiveness, and vice versa), so the challenge of IR systems is to find a compromise between efficient and effective data processing. This thesis investigates the efficiency of IR systems. It suggests several novel strategies that can render IR systems more efficient by reducing the index size of IR systems, referred to as index compression. The index is the data structure that stores the information handled in the retrieval process. Two different approaches are proposed for index compression, namely document reordering and static index pruning. Both of these approaches exploit document collection characteristics in order to reduce the size of indexes, either by reassigning the document identifiers in the collection in the index, or by selectively discarding information that is less relevant to the retrieval process by pruning the index.