Frame-Based Semantic Patterns for Relation Extraction

Abstract

This paper presents novel frame-based semantic patterns, exploiting frame element and frame annotations, provided by FrameNet for relation extraction. The proposed frame-based patterns are evaluated against state-of-the-art dependency based syntactic patterns and lexico-syntactic patterns, on three independent datasets that differ in size and construction. The results show that the proposed frame-based patterns significantly improve performance, both in terms of scoring higher precision and higher recall for relation extraction, in comparison to dependency and lexico-syntactic patterns on all three datasets.