Method and system for extracting information from unstructured text using symbolic machine learning

Abstract

A method (and structure) of extracting information from text, includes parsing an input sample of text to form a parse tree and using user inputs to define a machine-labeled learning pattern from the parse tree.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS [0001] The present Application is related to U.S. Provisional Patent Application No. 60 / 586,877, filed on Jul. 12, 2004, to Johnson et al., entitled “System and Method for Extracting Information from Unstructured Text Using Symbolic Machine Learning”, having IBM Docket YOR920040239US1, assigned to the present assignee, and incorporated herein by reference.BACKGROUND OF THE INVENTION [0002] 1. Field of the Invention [0003] The present invention generally relates to extracting information from text. More specifically, in a relational learning system, a pattern learner module receives a small number of learning samples defined by user interactions in relational pattern templates format wherein elements are defined in a precedence relation and in an inclusion relation, and calculates a minimal most specific generalization (MMSG) for these samples so that information matching the generalized template can then be extracted from unseen text. [0004] 2...

AI Technical Summary

Benefits of technology

[0024] It is another exemplary feature of the present invention to provide a method that allows a user with no special knowledge of linguistics to dynamically define patterns on the basis of a small number of example sentences or pseudo-examples in which the user has marked those named entity mentions that are involved in a relation instance. The defined patterns can then be used to identify relation instances in hitherto unseen sentences with high precision.

[0030] Thus, the present invention provides an improved method for relational learning in which a non-specialist can intuitively use the tool that embodies this method to develop a PI pattern template to be used for comparison with unseen text.

Problems solved by technology

Extracting relational information from text is an important and unsolved problem in the area of Unstructured Information Management.

Manual approaches are very costly to develop, since they require experts in computational linguistics or related disciplines to develop formal grammars or special purpose programs.

Non-specialists cannot customize manual systems for new domains, tasks or languages.

Statistical methods are quite popular, but they suffer from the problem of labeling sufficient data accurately for training a model.

This is a major problem for such approaches.

There are currently no adequate solutions to the problem of trainable relation extraction systems, especially no adequate systems that can be used by non-specialists.

Images