Detecting tension in online communities with computational Twitter analysis

Policing cyber-neighbourhoods: tension monitoring and social media networks

“…'Tension' per precision performance is not as accurate at 0.53. However, this result compares favourably with other automated methods of tension detection online (see Burnap et al 2013). Next, we turned to examining the temporal dimension and visualised spikes and troughs in tension.…”

“…Adapting language based tools advocated by Sacks (1992) and Housley and Fitzgerald (2002) via the process of Collaborative Algorithm Design, social scientists in collaboration with computer scientists have developed an automated social media tension-monitoring system that demonstrates a high level of agreement with human police coders (especially in the case of high tension). When compared to sentiment analysis tools the tension engine produced more efficacious results (see Burnap et al 2013), particularly in relation to identifying spikes in tension. While this paper reports the first scientific evidence on identifying tensions in social media networks, there are however several shortcomings that need highlighting.…”

“…On the question of whether sentiment analysis tools yields results that are commensurate with our tension-monitoring engine we have published a paper which demonstrates the superior efficacy of the latter (Burnap et al 2013). The paper shows that while the harmonised mean of precision and recall -the F-measure -is similar for both tools for the 'Some Tension' class, the tension engine performs better at the higher end of the tension scale, which makes it more useful for detecting raised spikes in tension that are so important in predicting the likelihood of public disorder.…”

“…It is a model of human category and associated linguistic work based on the systematic observation of everyday human practice and empirical materials (Housley 2002, Housley and Fitzgerald 2002). As a consequence, it provides a means of developing some bespoke rules that can inform the analysis of large scale social media data (in this case Twitter data) that can identify relevant 'topic events' and provide a way of analysing content in terms of category configuration and different forms of attribution and activity (See Burnap et al 2013 for a technical description).…”

Big and broad social data and the sociological imagination: A collaborative response

“…In this paper, we focus on Twitter data as it arguably provides the most open and voluminous social media data source and has thus become established as a key data source for public opinion and behaviour mining. Twitter data has been used to measure public mood (Bollen et al, 2009), opinion (Pak and Paroubek, 2010;Thelwall et al, 2011), tension and cohesion (Burnap et al, 2013a;Williams et al, 2013) and to explore communication patterns (Bruns and Stieglitz, 2012). The COSMOS platform currently provides nine modes of analysis, some of which operate at the individual tweet level and others at a corpus level (i.e.…”