AoIR 2019

Trust Us, Again? Twitter Campaigning Strategies in the 2019 Australian Federal Election

Axel Bruns, Tim Graham, and Dan Angus

• 5 Oct. 2019 – Association of Internet Researchers conference, Brisbane

Extended Abstract

Introduction

Australia has experienced substantial political instability at the federal level for more than a decade: since 2007, it has experienced six changes of Prime Minister, four of which were brought on not by the results of federal elections, but by personal and policy disagreements within the major parties. As a consequence of this unprecedented level of internal disunity, long-term policy-making agendas have often been sidelined by short-term factional machinations, and overall public trust in politicians from all sides, and in democracy as such, has declined to an all-time low (Evans et al. 2018). Meanwhile, a range of minor parties and independent candidates have emerged to exploit this disruption and present themselves as trustworthy alternatives to the established parties.

As Australia approaches its next federal election, likely to take place in May 2019, these minor party and independent candidates will likely target marginal electorates (where the sitting member does not hold a strong majority, and may be susceptible to challenge); conversely, the centrist Australian Labor Party and conservative Coalition of Liberal and National Parties will be working to regain the trust (and votes) lost through their internal rancour and disarray of the past twelve years. Social media will play a critical role in their campaigning: overall, Australians are comparatively early and enthusiastic adopters of social media (Sensis 2017), and more than half use social media as a key source of news (Newman et al. 2018: 127); more specifically, social media have played an important role already during previous federal election campaigns in 2013 and 2016 (Bruns 2016; Bruns & Moon 2017), and politics is a persistently prominent topic on platforms such as Twitter in Australia (Sauter & Bruns 2015).

Building on and extending an established methodological approach, this paper investigates the use of Twitter in the 2019 Australian federal election campaign. It examines the activities of political candidates, with a particular view towards their efforts to reposition themselves as trustworthy, as well as the engagement of ordinary Twitter users with these candidate accounts, in order to identify campaigning strategies and assess their effectiveness. The results of this work also extend the data gathered in previous federal election campaigns, to produce a longitudinal dataset across the 2013, 2016, and 2019 elections.

Method

Following the approach employed in Bruns (2017) and Bruns & Moon (2018), we capture the tweets posted by all officially declared federal candidates, as well as any tweets directed at their accounts (as @mentions or retweets), for the duration of the official campaign (typically 6-8 weeks to the election date). We generate a range of standard metrics as equivalent to the previous studies (including data on tweeting activity and tweets received per candidate and per party); these indicate the candidates’ level of social media effort as well as the (supportive as well as adversarial) public interest received. Further, we examine networks of interaction amongst candidate accounts (to show strategies of mutual support and coordinated antagonism within and across parties) and between candidates and the general public. We extend the methodology of previous work by undertaking hypothesis testing of tie formation between these different actors. We use an approach known as exponential random graph modelling (Robins et al. 2007) to test and validate assumptions about what structural forces might be driving network formation on Twitter during the election campaign.

Extending previous approaches, we also apply machine learning techniques to surface the key topics within the tweets by and at the candidates, and trace topical change over time. We use social semantic network analysis (Angus & Wiles 2018) to examine the degree of topical overlap between key participants and participant groups (e.g. public, politician, party member), and Structural Topic Models (STM) as our core topic modelling approach to summarise the large corpus of tweets into a small number of ‘topics’ for analysis (Roberts et al. 2013). STM extends the conventional topic model analysis of the tweet content to include document-level covariates, such as candidate party affiliation, author type (e.g. candidates, ordinary users, or media outlets), and the change of topics over time. This enables a further investigation of the underlying campaign themes and strategies employed by each party and candidate, and not least also their engagement with questions of trust: in politics and politicians in general, as well as in their ability to manage key elements of government ranging from the economy to border control.

STM also provides an opportunity to assess the extent to which the themes promoted by the candidates are addressed and adopted by ordinary users, or to which these users engage with the candidates around other topics. Our longitudinal analysis over the course of the campaign provides insights into the evolution of particular public debates; further comparisons with the 2013 and 2016 data also show whether in spite of the general turmoil in Australian politics these policy themes have remained broadly stable. Finally, we also apply sentiment analysis techniques to the tweets by and at candidates to assess the emotional tone of these debates over time and examine the use of positivity and negativity by specific candidates and their parties. For this analysis we use the SentiStrength algorithm (Thelwall et al. 2010), to assign each tweet a sentiment score ranging from extremely negative (-5) to extremely positive (+5). This approach has been used in previous studies of political tweets (Vilares et al. 2015), and Sentistrength is currently the state-of-the-art for sentiment analysis of Twitter data (Koto & Adriani 2015).

Expected Results

In keeping with the ethos of AoIR conferences, the project presented here is work in progress, as the election itself will take place in May 2019. However, our previous studies of federal and state elections in Australia have already demonstrated the feasibility and value of the approach outlined here. In particular, our reliance on a population of political candidates’ Twitter accounts has already been shown to generate a substantially more diverse dataset than the more standard approach of tracking election-related hashtags or keywords can produce: Australian politics hashtags such as #auspol or #ausvotes attract only a very narrow, self-selecting subset of those users who may discuss the federal election on Twitter, and not all political candidates will use them.

By contrast, our approach captures all tweets by all candidates who operate public Twitter accounts, enabling a straightforward comparison of their tweeting activities across individuals and parties. It also captures all public tweets directed at these candidates’ accounts, and while tweeting at candidates still involves a certain degree of self-selection, these tweets represent a considerably larger and more diverse subset of the Australian Twittersphere than the corresponding hashtag datasets.

In combination, then, our approach here produces substantial new evidence both on the use of Twitter in political campaigning in Australia, and on the public response to this use. By interfacing with previous studies that used equivalent methodologies, it adds to a longitudinal observation of social media campaigning strategies that stretches across three federal election cycles, and serves to complement other such longer-term work (e.g. Larsson & Moe 2016; Lilleker et al. 2016).

References

Angus, D., & Wiles, J. (2018). Social Semantic Networks: Measuring Topic Management in Discourse Using a Pyramid of Conceptual Recurrence Metrics. Chaos: An Interdisciplinary Journal of Nonlinear Science, 28(8), 085723. doi:10.1063/1.5024809

Bruns, A. (2017). Tweeting to Save the Furniture: The 2013 Australian Election Campaign on Twitter. Media International Australia, 162, 49–64. https://doi.org/10.1177/1329878X16669001

Bruns, A., & Moon, B. (2018). Social Media in Australian Federal Elections: Comparing the 2013 and 2016 Campaigns. Journalism & Mass Communication Quarterly, 95(2), 425–448. https://doi.org/10.1177/1077699018766505

Evans, M., Stoker, G., & Halupka, M. (2018). Australians’ Trust in Politicians and Democracy Hits an All-Time Low: New Research. The Conversation, 5 Dec. 2018. https://theconversation.com/australians-trust-in-politicians-and-democracy-hits-an-all-time-low-new-research-108161

Koto, F., & Adriani, M. (2015). A Comparative Study on Twitter Sentiment Analysis: Which Features Are Good? Paper presented at Natural Language Processing and Information Systems (NLDB 2015), Passau, Germany. In C. Biemann, S. Handschuh, A. Freitas, F. Meziane, & E. Métais (Eds.). Lecture Notes in Computer Science, 9103 (pp. 453-457).

Larsson, A.O., & Moe, H. (2016). From Emerging to Established? A Comparison of Twitter Use during Swedish Election Campaigns in 2010 and 2014. In A. Bruns, G. Enli, E. Skogerbø, A.O. Larsson, & C. Christensen (Eds.), The Routledge Companion to Social Media and Politics (pp. 311-324). New York: Routledge.

Lilleker, D.G., Jackson, N., & Koc-Michalska, K. (2016). Social Media in the UK Election Campaigns 2008-2014: Experimentation, Innovation, and Convergence. In A. Bruns, G. Enli, E. Skogerbø, A.O. Larsson, & C. Christensen (Eds.), The Routledge Companion to Social Media and Politics (pp. 325-337). New York: Routledge.

Robins, G., Pattison, P., Kalish, Y., & Lusher, D. (2007). An Introduction to Exponential Random Graph (p*) Models for Social Networks. Social Networks, 29(2), 173­-191.

Newman, N., Fletcher, R., Kalogeropoulos, A., Levy, D.A.L., & Nielsen, R.K. (2018). Reuters Institute Digital News Report 2018. Oxford: Reuters Institute for the Study of Journalism. http://media.digitalnewsreport.org/wp-content/uploads/2018/06/digital-news-report-2018.pdf?x89475

Roberts, M., Stewart, B., Tingley, D., & Airoldi, E. (2013). The Structural Topic Model and Applied Social Science. Paper presented at Advances in Neural Information Processing Systems (NIPS) Workshop on Topic Models: Computation, Application, and Evaluation, Lake Tahoe, Nevada. In C. Burges, L. Bottou, M. Welling, Z. Ghahramani, & K. Weinberger (Eds.). Advances in Neural Information Processing Systems 26 (NIPS 2013) (pp. 1-20).

Sauter, T., & Bruns, A. (2015). #auspol: The Hashtag as Community, Event, and Material Object for Engaging with Australian Politics. In N. Rambukkana (Ed.), Hashtag Publics: The Power and Politics of Discursive Networks (pp. 47–60). New York: Peter Lang.

Sensis. (2017). Sensis Social Media Report 2017: Chapter 1 – Australians and Social Media. Melbourne: Sensis. https://www.sensis.com.au/asset/PDFdirectory/Sensis-Social-Media-Report-2017.pdf

Thelwall, M., Buckley, K., Paltoglou, G. Cai, D., & Kappas, A. (2010). Sentiment Strength Detection in Short Informal Text. Journal of the American Society for Information Science and Technology, 61(12), 2544–2558.

Vilares, D., Thelwall, M., & Alonso, M.A. (2015). The Megaphone of the People? Spanish SentiStrength for Real-Time Analysis of Political Tweets. Journal of Information Science, 41(6), 799-813.