Introduction
Geoengineering has been long discussed as a way to deal with global warming. But concurrent with the growing interest in geoengineering in scientific and political circles, a marginal discourse of “chemtrail” conspiracy is also breeding on the internet (Cairns, 2016). Nowadays, anthropogenic global warming became a scientific consensus. Meanwhile, many conspiracy theorists still assert that climate change is a political manipulation in online discussions on online platforms such as Twitter, Youtube, or personal blogs.
Research Questions
To understand the alternative discourse on geoengineering on Youtube, we formulate the following research questions: (1) How do discussions in comment sections about geoengineering on Youtube differ when comparing videos of mainstream broadcasters to content that deals with the conspiracy of chemtrails? (2) Which topics are discussed mostly, what are the similarities and what are the differences? (3) Which rational and irrational fears might be expressed by commenters? An analysis of these questions can help assess future debates on the implementation of geoengineering in a real-world scenario. It might be necessary to implement technology to tackle the consequences of climate change, and there is great potential for conflict. According to Cairns (2016), there is an instability in the distinction between ‘paranoid’ and ‘normal’ views on geoengineering. Machine learning-based data analysis - critical discourse analysis combining with clustering analysis - will be applied to analyze opinions expressed in Youtube comments on a broader scale with big data.
Material and Data
For data collection, we build upon the methodological approach presented by Allgaier (2019) to alleviate potential bias that might arise from personalized video searches on Youtube. With the use of the anonymity network Tor and iterative video searches on predefined keywords containing: (1) climate engineering, (2) geoengineering, (3) chemtrails, we want to make sure that our sample of comment sections is randomized and we circumvent the problem of filter bubbles. Based on the categories developed by Allgaier (2019), we will classify the collected Youtube videos into (1) conspiracy videos, (2) professional media content, (3) science Youtubers, (4) scientific institutions, (5) other.
Technology and Methods
Subsequently to data acquisition and preprocessing, we use a quantitative approach as the basis for qualitative analysis to gain a first understanding of the data and gather new insights.
In our case, a purely quantitative approach would not be an ideal solution since the machine cannot grasp a deep semantic understanding of some aspects of language, such as irony or other communication techniques present in YouTube comments. However, we are also aware that the pre-trained word embedding model may present a discrepancy between different usage scenarios.
Cluster analysis is then applied to help find patterns within the dataset to identify the discourse structures and serve as a key component in our mixed research method.
The word embeddings model and the k-medoid clustering algorithm are, therefore, an integral part of our work pipeline, which ends with a clustering visualization tool. The result of the visualization is then taken as the basis for our discussion, which is part of an iterative approach that also involves the parameter tuning process.
Analytical Approach
We will apply critical discourse analysis as our analytical approach. Discourse analysis that recontextualizes practices in text and talk could illuminate the central role of language in politics reveal the embeddedness of language in practice, and answer ‘how’ questions (Hajer & Versteeg, 2005; ). Compared to discourse analysis, a critical discourse analysis (CDA) focuses more on power dynamics and ideological shift in a Foucaultian sense (Törnberg & Törnberg, 2016). The conventional approaches are theory-driven, operated by qualitative content analysis with a small number of samples. In this project, we aim to integrate the machine learning methods into CDA through exploratory study design to identify contested notion in the Youtube discourse on geoengineering.
Reference
Allgaier, J. (2019). Science and environmental communication on YouTube: strategically distorted communications in online videos on climate change and climate engineering. Front Commun, 4, 36.
Cairns, R. (2016). Climates of suspicion: ‘chemtrail’ conspiracy narratives and the international politics of geoengineering. The Geographical Journal, 182(1), 70-84.
Hajer, M., & Versteeg, W. (2005). A decade of discourse analysis of environmental politics: Achievements, challenges, perspectives. Journal of environmental policy & planning, 7(3), 175-184.
Törnberg, A., & Törnberg, P. (2016). Combining CDA and topic modeling: Analyzing discursive connections between Islamophobia and anti-feminism on an online forum. Discourse & Society, 27(4), 401-422.
Understanding Geoengineering Discourse through Machine Learning Approaches Milan Bargiel @milanbargiel , Claudio Vindimian @Cloudz333 , Xin Yu @yuxin16 , (Xixuan Zhang)
Introduction Geoengineering has been long discussed as a way to deal with global warming. But concurrent with the growing interest in geoengineering in scientific and political circles, a marginal discourse of “chemtrail” conspiracy is also breeding on the internet (Cairns, 2016). Nowadays, anthropogenic global warming became a scientific consensus. Meanwhile, many conspiracy theorists still assert that climate change is a political manipulation in online discussions on online platforms such as Twitter, Youtube, or personal blogs.
Research Questions To understand the alternative discourse on geoengineering on Youtube, we formulate the following research questions: (1) How do discussions in comment sections about geoengineering on Youtube differ when comparing videos of mainstream broadcasters to content that deals with the conspiracy of chemtrails? (2) Which topics are discussed mostly, what are the similarities and what are the differences? (3) Which rational and irrational fears might be expressed by commenters? An analysis of these questions can help assess future debates on the implementation of geoengineering in a real-world scenario. It might be necessary to implement technology to tackle the consequences of climate change, and there is great potential for conflict. According to Cairns (2016), there is an instability in the distinction between ‘paranoid’ and ‘normal’ views on geoengineering. Machine learning-based data analysis - critical discourse analysis combining with clustering analysis - will be applied to analyze opinions expressed in Youtube comments on a broader scale with big data.
Material and Data For data collection, we build upon the methodological approach presented by Allgaier (2019) to alleviate potential bias that might arise from personalized video searches on Youtube. With the use of the anonymity network Tor and iterative video searches on predefined keywords containing: (1) climate engineering, (2) geoengineering, (3) chemtrails, we want to make sure that our sample of comment sections is randomized and we circumvent the problem of filter bubbles. Based on the categories developed by Allgaier (2019), we will classify the collected Youtube videos into (1) conspiracy videos, (2) professional media content, (3) science Youtubers, (4) scientific institutions, (5) other.
Technology and Methods Subsequently to data acquisition and preprocessing, we use a quantitative approach as the basis for qualitative analysis to gain a first understanding of the data and gather new insights. In our case, a purely quantitative approach would not be an ideal solution since the machine cannot grasp a deep semantic understanding of some aspects of language, such as irony or other communication techniques present in YouTube comments. However, we are also aware that the pre-trained word embedding model may present a discrepancy between different usage scenarios. Cluster analysis is then applied to help find patterns within the dataset to identify the discourse structures and serve as a key component in our mixed research method. The word embeddings model and the k-medoid clustering algorithm are, therefore, an integral part of our work pipeline, which ends with a clustering visualization tool. The result of the visualization is then taken as the basis for our discussion, which is part of an iterative approach that also involves the parameter tuning process.
Analytical Approach We will apply critical discourse analysis as our analytical approach. Discourse analysis that recontextualizes practices in text and talk could illuminate the central role of language in politics reveal the embeddedness of language in practice, and answer ‘how’ questions (Hajer & Versteeg, 2005; ). Compared to discourse analysis, a critical discourse analysis (CDA) focuses more on power dynamics and ideological shift in a Foucaultian sense (Törnberg & Törnberg, 2016). The conventional approaches are theory-driven, operated by qualitative content analysis with a small number of samples. In this project, we aim to integrate the machine learning methods into CDA through exploratory study design to identify contested notion in the Youtube discourse on geoengineering.
Reference Allgaier, J. (2019). Science and environmental communication on YouTube: strategically distorted communications in online videos on climate change and climate engineering. Front Commun, 4, 36. Cairns, R. (2016). Climates of suspicion: ‘chemtrail’ conspiracy narratives and the international politics of geoengineering. The Geographical Journal, 182(1), 70-84. Hajer, M., & Versteeg, W. (2005). A decade of discourse analysis of environmental politics: Achievements, challenges, perspectives. Journal of environmental policy & planning, 7(3), 175-184. Törnberg, A., & Törnberg, P. (2016). Combining CDA and topic modeling: Analyzing discursive connections between Islamophobia and anti-feminism on an online forum. Discourse & Society, 27(4), 401-422.