Compressed-Domain Text Indexing and Retrieval

点击翻译按钮获取中文摘要

This project investigates the interaction between text compression and text retrieval, and aims to build a parallel text retrieval system called Codir that allows compression-domain text indexing and retrieval. Ultimately, Codir is expected to be able to efficiently respond to queries consisting of one or multiple "keywords" connected via Boolean operators, with documents in the text database that either contain those keywords, or whose "theme" matches the keywords. The most important feature of Codir is the flexibility it provides to allow space/time tradeoff and the use of this flexibility in the Dynamic Index Cache. The two target applications areas to which the technologies developed in Codir are to be applied to are Web Page Searching and Video Access using Captioning. The expected contributions of this project to the text retrieval field are: (1) the performance tradeoffs of integrating text compression and text retrieval in a unified framework; (2) the detailed characterization of the access patterns and query statistics of a production-mode use of Codir; (3) the development of new techniques to support approximate keyword search and frequent in-place updates to the text database; (4) the incorporation of advanced text classification techniques such as term weighting and theme indexing into the implementation framework of Codir. The technologies developed from this project are expected to be useful to designers of corporate information repository, internet search engines, and large-scale text-based Web servers.