Abstract:
Political campaigns recently use social media networks as an important environment to gain visibility for causes such as policy disagreement. Users often show divided opinions on the enacted policy, which sparks intense political debate. In such context, we attempt to predict the advocacy of users on Twitter involved in the campaign advocacy about Citizenship Amendment Act (CAA), 2019, enacted by the Indian Government on December 12, 2019. We collect a rich network of followers on top of a pre-existing dataset consisting of 14M tweets related to CAA from December 07, 2019, to February 27, 2020, where the users on social media were divided over its non-secular roots. Our follower network dataset contains follower network for 631,919 users, with 313,068088 links. We identified the campaign advocates into 3 categories: for-the-motion advocates, against-the-motion advocates, and neutral advocates. We further try to solve the problem of predicting people’s campaign advocacy based on user activity. We then investigate transfer learning approach on top of unsupervised stance detection on the users code mixed content on platform, which provide good accuracy and provides more potential for generalizing across datasets. With the controversial nature of datasets involved, we identify possible bad actors in the dataset and analyse them with respect to content generation in the trend. We add another layer to the findings by bringing in stance to the picture.We also investigate the effect of network embeddings to see how well they can predict stance in a protest this large, and report our findings. The findings of this research can help the government and policy makers plan a policy change appropriately, keeping the constituency’s sentiments a prime focus.
