Label-aware learning to enhance unsupervised cross-domain rumor detection

Abstract

Recently, massive research has achieved significant development in improving the performance of rumor detection. However, identifying rumors in an invisible domain is still an elusive challenge. To address this issue, we propose an unsupervised cross-domain rumor detection model that enhances contrastive learning and cross-attention by label-aware learning to alleviate the domain shift. The model performs cross-domain feature alignment and enforces target samples to align with the corresponding prototypes of a given source domain. Moreover, we use a cross-attention mechanism on a pair of source data and target data with the same labels to learn domain-invariant representations. Because the samples in a domain pair tend to express similar semantic patterns, especially on the people’s attitudes (e.g., supporting or denying) towards the same category of rumors. In addition, we add a label-aware learning module as an enhancement component to learn the correlations between labels and instances during training and generate a better label distribution to replace the original one-hot label vector to guide the model training. At the same time, we use the label representation learned by the label learning module to guide the production of pseudo-label for the target samples. We conduct experiments on four groups of cross-domain datasets and show that our proposed model achieves state-of-the-art performance.