Traditional supervised learning makes the closed-world assumption that theclasses appeared in the test data must have appeared in training. This alsoapplies to text learning or text classification. As learning is usedincreasingly in dynamic open environments where some new/test documents may notbelong to any of the training classes, identifying these novel documents duringclassification presents an important problem. This problem is called open-worldclassification or open classification. This paper proposes a novel deeplearning based approach. It outperforms existing state-of-the-art techniquesdramatically.
展开▼