Entity Factor: A Balanced Method for Table Filling in Joint Entity and Relation Extraction

The knowledge graph is an effective tool for improving natural language processing, but manually annotating enormous amounts of expensive. Academics have conducted research on entity and relation extraction techniques, among which, the end-to-end table-filling approach a popular direction achieving joint extraction. However, once table has been populated in uniform label space, large number null labels are generated within array, causing label-imbalance problems, which could result tendency model’s encoder to predict labels; that is, model generalization performance decreases. In this paper, we propose method mitigate non-essential matrices. This utilizes score matrix calculate count non-entities percentage matrix, then projected by power constant generate entity-factor matrix. incorporated into scoring back-propagation process, gradient null-labeled cells factor layer affected shrinks, amplitude related size factor, thereby reducing feature learning labels. Experiments with two publicly available benchmark datasets show incorporation factors significantly improved performance, especially task, 1.5% both cases.