Applying text mining to multi-level indexing and searching for enhancing probabilistic information retrieval.

摘要

Information Retrieval (IR) refers to finding requested information from large amounts of data. Probabilistic IR model, which has been developed for decades, is one of the most successful models. With the development of information technology (IT), users today expect more accurate retrieval results from an IR system. Instead of getting many relevant articles or journals, users are preferred to obtain direct answer from the IR system. The IR system should be more intelligent to help users retrieve highly related information. In this thesis, we propose a Multi-level Indexing and Searching Framework, which is based on the probabilistic model, to achieve high accuracy retrieval performance. Different from the original IR system, the proposed framework has the capability to generate high accuracy results in both document level and passage level, according to specified retrieval requirements. To further enhance the retrieval performance, the utilization of text mining techniques in relevant process is also explored. Extensive experiments on four years Text REtrieval Conference (TREC) data sets are conducted to illustrate the effectiveness of the proposed framework and ideas.