Inferring stances of silent-participants in Twitter chatter using Label Propagation

Measuring public opinion has many interesting applications. It is a known fact that in social media networks, the majority of the users consume information without taking part in the discourse. Quantifying the opinions and stances on day to day happenings with a silent majority may not be possible with just Natural Language Processing tools. In this work we explore the potential possibility of using graph methods to propagate the found opinions of users to silent participants. For this work we use the Twitter chatter during the early pandemic times since there was a lot of polarizing ideas that was being discussed in Twitter. We also know there was some misinformation relating to important medical facts and it was important to convey proper political decisions to the user base. It also follows that understanding where users stand on certain aspects such as masks and vaccinations can be used in better informing the public about matters such as health and safety. Therefore it is important to capture a comprehensive opinion about these important matters. We report the findings of our experimental study into using Label Propagation to find user stances of silent participants. We use several implementations of Label Propagation Algorithm including some which are derivatives of the traditional method to test our hypothesis. We also present our results which perform with 0.94 with F-1 (macro) against our baselines which are at 0.49 F-1(macro).