Entity and Relation Extraction with Dynamic Memory Cells

Extract entity and relation from plain text is a fundamental but important task. Generally, a sentence contains several entities, and some unrelated entity pairs produce redundant information. To reduce the noise, we propose a novel relation detection preferential framework. Unlike most existing models, our proposed method learns sentence features without any complicated feature engineering or Natural Language Processing (NLP) toolkits. The proposed model manipulates multi-relations and corresponding entity pairs in parallel. A set of dynamic memory cells are expanded to capture different relation information in relation detection module. We apply the framework on the CoNLL04 dataset. Results show that it achieves the state-of-the-art performance.