Research On Cross Domain Hedge Detection

Hedege detection is also named as uncertainty detection. Hedge is an important phenomenon in nature language text. Hedge information often appears in the situation that no truth value can be attributed or the speaker intentionally omits some information from the statement, making it vague, ambiguous, or misleading.For many Nature Language Processing application, it is a necessary preprocess to identify certain and uncertain information. For example, in information extraction and retrieval, it is an essential step to identify certain and uncertain information. Actually, what we need is certain information. We could acquire the information by ignore the hedge since the uncertain information always be expressed by lexical cues.However, annotated data of hedge we could found is just in several special domains.The linguistic expressions in different domains make the hedge distribution different. Traditional machine learning works well under the assumption that the training data and test data are in the same distribution. The different hedge distribution of domains and the high cost for labeling the sufficient training data in particular domains makes it hard to extend to other real-world application.In this paper, we present a novel framework, which combines transfer learning and semi-supervised learning, to solve the problem of data scarcity. Our cross domain framework works better than other state-of-the-art transfer learning approaches in several public data sets. In cross domain hedge detection, a new method of candidate phrases classification was used. Our experiments in three domains show that our framework improves the detection performance compared to existing cross domain detection method.