The ability to detect and characterize online information campaigns will rely on the ability to infer shared underlying narrative viewpoints of online content. Existing methods typically can cluster documents by topic but struggle to distinguish between different points of view. The task is challenging due to the inherent characteristics of social media texts, which are noisy, short, and often provide very little context. Yet, due to the widespread prevalence of harmful misinformation in online media, the development of viewpoint detection approaches is crucial to enable the identification of information campaigns, the characterization of their sources, and their evolution. Our work proposes a weakly supervised contrastive representation learning approach to infuse latent text representations with viewpoint information by leverPlease check and confirm if the authors given and family names have been correctly identified.aging proxy signals observed through social interaction networks. We test our solution on a Twitter dataset related to COVID-19 discussions. We demonstrate the ability of our approach to separate COVID-19 tweets by their narrative viewpoint compared to baseline pre-trained embeddings.