Extracting Relational Facts by an End-to-End Neural Model with Copy Mechanism

TLDR

Relational facts in sentences are complex, with overlapping triplets, and existing methods mainly target the normal class, missing these overlaps. The study proposes an end‑to‑end sequence‑to‑sequence model with a copy mechanism to jointly extract relational facts from sentences of any overlap class. The model classifies sentences into Normal, EntityPairOverlap, and SingleEntityOverlap, then uses either a single unified decoder or multiple separate decoders within a seq‑to‑seq framework with copy attention. On two public datasets, the proposed model significantly outperforms baseline methods.

Abstract

The relational facts in sentences are often complicated. Different relational triplets may have overlaps in a sentence. We divided the sentences into three types according to triplet overlap degree, including Normal, EntityPairOverlap and SingleEntiyOverlap. Existing methods mainly focus on Normal class and fail to extract relational triplets precisely. In this paper, we propose an end-to-end model based on sequence-to-sequence learning with copy mechanism, which can jointly extract relational facts from sentences of any of these classes. We adopt two different strategies in decoding process: employing only one united decoder or applying multiple separated decoders. We test our models in two public datasets and our model outperform the baseline method significantly.