Neural Inference of Political Bias in Hashtag and Entity Usage Across Platforms

In India, social media is a significant tool of political communication with parties, news organizations, and citizens taking advantage of online platforms to frame and shape narratives and opinion. Hashtags and named entities tend to be indicators of ideological stance and influence the dynamics of debate. Previous studies in the field have focused primarily on Twitter sourced English content, which fails to reflect the multiculturalism of the languages and platforms used during Indian elections. In the current paper, the author introduces Polarity-Aware Embedding Framework (PAEF), a neural framework which separates semantic meaning and ideological polarity to detect political bias on a token, post, and user level. To back up this framework, Twitter, Reddit, Koo, ShareChat, and YouTube were used to compile a multilingual and cross-platform dataset that spanned Twitter, Reddit, Koo, ShareChat, and YouTube in three electoral cycles, namely the 2019 Lok Sabha elections, the 2023 state elections, and the 2024 General Election. The experimental findings indicate that PAEF is superior to the baseline techniques, and indicate the difference in bias reinforcement across platforms and languages. The work provides a detailed yardstick by which the polarization of Indian elections can be analysed and thus, the research on the political bias in online discussion may be developed.