unSeminar: "Predicting conversation length of protest-related discussions on Twitter”

Presented by Teresa Gil-Lopez and Cuihua Shen.

Key Words: text mining, sentiment analysis, record matching, database design, modeling, prediction

Abstract:

Despite the need to understand political conversation as it occurs in everyday life settings, most research has focused on political talk within formal deliberation settings rather than exploring patterns in informal political conversations. We propose to empirically examine Twitter conversation threads as an approach to the study of online public opinion dynamics. We aim to identify significant user- and message-level predictors of conversation length in the context of protest-related online discussions, such as user influence and positive or negative sentiment. We hypothesize that these factors underlie mechanisms triggering users’ responses after an initial message is made public. We seek feedback on: 1) Programmatically aligning tweets within conversations and the broader political context, 2) Implementing text mining and sentiment analysis techniques on short communications,
* 3) Developing models to test the role of influence and sentiment on conversation length and dynamics.