Learning Implicit and Explicit Multi-task Interactions for Information Extraction

Information extraction aims at extracting entities, relations, and so on, in text to support information retrieval systems. To extract information, researchers have considered multitask learning (ML) approaches. The conventional ML approach learns shared features across tasks, with the assumption that these capture sufficient task interactions learn expressive representations for classification. However, such an is flawed different perspectives. First, representation may contain noise introduced by another task; tasks coupled complexities but this treats all equally; has a flat structure hinders of explicit interactions. This approach, however, implicit often generalization ability benefited multitasks. In article, we take advantage learned approaches while alleviating issues mentioned above developing Recurrent Interaction Network effective Early Prediction Integration (RIN-EPI) learning. Specifically, RIN-EPI two related tasks. effectively consider correlations among outputs It is, obvious are unobservable during training, leverage predictions intermediate layers (referred as early predictions) proxies well through attention mechanisms sequence models. By recurrently interactions, gradually improve individual We demonstrate effectiveness on mainstream multitasks extraction: (1) entity recognition relation classification (2) aspect opinion term co-extraction. Extensive experiments architecture, where achieve state-of-the-art results several benchmark datasets.