Weakly supervised contrastive representation learning to encode narrative viewpoint of COVID-19 tweets
Fig 5
ViewpointNN performance with k=15 of the best contrastive model (TA-FT Strong) and baselines across the different topics in the annotated set of tweets.