摘要：AbstractThis paper introduces a framework for semantic information retrieval based on the integration of various natural language processing (NLP) techniques, each of which annotates a base text with different kinds of information extracted from the text. Instead of running the NLP modules on the fly for individual search requests, the NLP modules are applied to the text in advance and the results are indexed in a way that enables flexible and efficient integration of them. The query language is based on a variant of the region algebra, in which we can specify a sub- structure in the annotated text that may involve different kinds of annotations. Given a query, the retrieval engine searches for the sub-structure by aggregating the different kinds of annotations through a search algorithm for the extended region algebra. We demonstrate the effectiveness and flexibility of the proposed framework through experiments with TREC Genomics Track data.
