Combining feature-based and instance-based transfer learning approaches for cross-domain hedge detection with multiple sources

Abstract

The difference of hedge cue distributions in various domains makes the domain-specific detectors difficult to extend to other domains. To make full use of out-of-domain data to adapt to a new domain and minimize annotation costs, we propose a novel cross-domain hedge detection approach called FIMultiSource, which combines instance-based and feature-based transfer learning approaches to make full use of multiple sources. Experiments carried on BioScope, WikiWeasel, and FactBank corpora show that our approach works well for cross-domain uncertainty recognition and always improves the detection performance compared to other state-of-the-art instance-based and feature-based transfer learning approaches.