Learning the Language of BlackTwitter

ASA Symposium on Data Science and Statistics

Invited Talk: Linguistic Diversity in NLP Session, Slides

Abstract --- Free expression on social media has awaken many often-silenced voices, especially those from marginalized groups. One such example of organized quiet voices has formed into BlackTwitter, a sub-culture within Twitter. It is a community that expresses the sentiments on current issues and topics that impact the Black community. A defining hallmark of BlackTwitter is the blacktag, a hash tag that includes content related to the African-American/Black community. Various racialized hashtags span the cultural experiences of African Americans. Recent examples of blacktags include #BlackLivesMatter, #ICantBreathe #HandsUpDontShoot, #BlackGirlMagic and #OscarsSoWhite. Blacktags have been instrumental in mobilizing and producing social change and awareness. BlackTwitter has been the driving force behind administering initiatives such as the BlackLivesMatter Movement and My Brother’s Keeper.

BlackTwitter remains a nebulous, yet impactful, ecosystem of users and conversations. Although BlackTwitter has played a significant role in amplifying declarations of the African-American community toward social change, the slate of text and sentiment analysis algorithms seemingly lack the robustness to capture the layers of Black cultural nuances. The limited and diverse language structure of Twitter’s 280 characters further challenges the accuracy and generalizability of these algorithms. A systematic integration of keywords and topic choice can boost existing text and sentiment analysis processes.

In this talk, I briefly survey the evolution of BlackTwitter. I share an approach to identify and evaluate blacktags. I also discuss potential approaches of enhancing sentiment analysis algorithms, particularly with respect to cultural undertones.

This work is based on:

Marshall, B., Blunt, T., Thompson, T. (2018). The Impact of Live Tweeting on Social Movements. Proceedings of the IEEE Information Reuse and Integration, pp. 209-216.