A Weakly-Supervised Rule-Based Approach for Relation Extraction

Rule-based approaches for information extraction usually achieve good precision values, even if they often need a lot of manual e ort to be implemented. In this paper, we present a novel rule-based strategy for semantic relation extraction that takes advantage of partial syntactic parsing in order to simplify the linguistic structures containing instances of semantic relations. We also propose a distant supervision strategy that automatically extracts generic lexico-syntactic patterns by means of semi-structured resources such as Wikipedia infoboxes. These generic patterns are then transformed into extraction rules that are used to update a partial dependency grammar. Several evaluations of this method show that it improves the recall while maintaining high-precision values. Experiments were performed on Spanish texts.

keywords: information extraction, relation extraction, ontologies, lexico-syntactic patterns, text compression