Coreference Resolution in Research Papers from Multiple Domains

Coreference resolution is essential for automatic text understanding to facilitate high-level information retrieval tasks such as summarisation or question answering. Previous work indicates that the performance of state-of-the-art approaches (e.g. based on BERT) noticeably declines when applied scientific papers. In this paper, we investigate task coreference in research papers and subsequent knowledge graph population. We present following contributions: (1) annotate a corpus comprises 10 different disciplines from Science, Technology, Medicine (STM); (2) propose transfer learning papers; (3) analyse impact (KG) population; (4) release KG automatically populated 55,485 STM domains. Comprehensive experiments show usefulness proposed approach. Our approach considerably outperforms baselines our with an F1 score 61.4 (+11.0), while evaluation against gold standard shows improves quality significantly 63.5 (+21.8).