Abstract
Relation Extraction (RE) from tables is the task of identifying relations between pairs of columns. Generally, RE models for this task require labelled tables for training. Luckily, labelled tables can also be generated artificially from a Knowledge Graph (KG), which makes the cost to acquire them much lower in comparison to manual annotations. However, these tables have one drawback compared to real tables, which is that they lack associated metadata, such as column-headers, captions, etc. This is because synthetic tables are created out of KGs that do not store such metadata. Unfortunately, metadata can provide strong signals for RE from tables. To address this issue, we propose methods to artificially create some of this metadata for synthetic tables. We then experiment with a RE model that uses artificial metadata as input. Our empirical results show that this leads to an improvement of 9\%-45\% in F1 score, in absolute terms, over 2 tabular datasets.
Abstract (translated)
URL
https://arxiv.org/abs/2108.10750