Tree pattern expression for extracting information from syntactically parsed text corpora

With the public availability of a number of syntactically parsed text corpora, it has been increasingly important to efficiently extract desired information from such corpora. Many conventional works extract a desired text part by matching the parse tree of each sentence to a query that is represented as a structural form of relational predicates expressing a common structural pattern of desired text parts. However, although those works can be useful for limited types of simple queries, they are not very efficient in general because query formulations are sometimes very complicated for complex patterns of desired text parts and query matching tasks are likely to be exponentially time-consuming when considering a variety of complex sentential structures in a text corpus. In order to overcome such inadequacy, we present a novel tree pattern expression (TPE) that can represent various structural patterns intuitively and reduce pattern-matching complexity significantly. This paper first proposes TPE and its pattern-matching algorithm, and then theoretically analyzes the complexity of the proposed pattern-matching algorithm. It also illustrates a TPE-based information extraction system, which is applied to real text mining in a bio-text corpus. It finally shows some experimental results with some discussions in comparison with other systems.