How does YouTube mediate current events? An analysis of YouTube comment sections in the context of conflict

Group members: Bastian August, Deena Loman, Sophie Pattynama, Brogan Latil, Chenxuan Zhang

1. Introduction

While YouTube is primarily understood for its entertainment purposes, its political power is increasingly acknowledged (Housseinmardi et al. 2021). The comment section is one example through which YouTube reflects this political power as information undergoes interpretation and negotiation, where truth and knowledge are constantly in flux (Uba & Jansson 2020, 252). Rather than focusing on truth construction, or the fashionable consideration of ‘post-truth’, this paper aims to study the mediatory characteristics of YouTube comment sections agnostically - not as sites of truth vs. non-truth, but as sites of user participation (Jenkins 2007). By delineating specific cases of comment sections, the study builds on a corpus of literature that has previously acknowledged YouTube as a space of participatory culture (Chau 2010; Traynor et al. 2020; Trott 2022). Therefore, this report’s primary focus is to examine how YouTube comment sections reveal qualities of commenter participation, specifically surrounding instances of conflict.

To investigate the topography of comment sections, two cases are used: the Russia-Ukraine War and the Chinese White Paper Protests. Firstly, on the 24th of February 2022, BBC News (2022a) reported that “Russia initiated a large-scale military attack on Ukraine [...] on the orders of Russian President Vladimir Putin”. This was justified by the Russian president Vladimir Putin as a military operation, and argued that the invasion was a movement of self-defense (BBC, 2022a). To frame this ‘operation’, Putin put the emphasis on ‘de-Nazifying’ Ukraine. Several months later, further international outrage was triggered by the explosions of the Nord Stream-2 pipeline which supplies energy between Russia and Europe (Oltermann, Beaumont and Sabbagh, 2022). Whether this was sabotage or an accident is continuously debated, which is exemplified in comment sections on YouTube. Secondly, the White Paper Protests were initially triggered by a residential apartment fire in Urumqi, China, killing a disputed number of Uyghur citizens (BBC, 2022b). Citizens argued that the Zero-Covid restrictions prevented emergency forces from reaching the scene of the fire (BBC, 2022b). Thus, the white papers used in the ensuing protests symbolize the resentment regarding the Zero-Covid regulations and the censorship by the Chinese Communist Party (CCP) (Wong and Williams, 2022).

These two distinct case studies lead this paper to a deeper analysis of the narratives and user practices that are derived in these comment sections. Therefore, the research questions in this paper ask:

RQ 1 - As sites of participatory culture, what do commenting sections on YouTube look like in cases of conflict or contentious subject matter?

RQ 2 - What classes of participation make up these spaces?

The two distinct case studies allow for a deeper understanding surrounding YouTube ’s role in mediating current events. In this paper, the narratives and user practices present in the comment sections of the selected YouTube videos are collected and analyzed. By doing so, it will be studied how these conflict-centric videos call forth modes of participation. This research is socially relevant by emphasizing the power that platforms have in distributing and reinterpreting information and knowledge (Arthurs et al. 2017). Furthermore, YouTube specifically has an increasing ability to encourage participation and interaction between users. For instance, Ha and colleagues (2022) state that comments containing conspiracy theories on YouTube get more interactions (likes, replies, etc.) which then are pushed to the top of a comment section due to YouTube ’s algorithms.

In this study, Stuart Hall’s (1973) ‘encoding and decoding’ framework (Figure 1) is applied to the communicative processes evident in the YouTube comment section. In alignment with this framework, a video can be seen as a message encoded into media by its creators. Hall refers to the initial encoding of a message into media as ‘message I’. The decoding of that message by perceivers leads to another interpretation of it, which can be described as ‘message II’. In our case, this is happening in the comment section under the video which creates a cycle of new messages getting produced (encoded) and then interpreted (decoded) by the responders. In other words, the messages are refurbished through commenting practices. The degree of difference between ‘messages I’ and ‘II’ depends on the perceiver’s interpretation and understanding of the initial message. Further, reactions to the messages are agreement, negotiation, and opposition, which then shape the communication processes that ensue. The ‘meaning-structures’ of the sending and receiving persons try to make sense of the messages. This is constructed through experiences, knowledge, social institutions, and the technical structures that shape the user’s process of interpretation (Hall 1973). This process of encoding/decoding can lead to different narratives arising in the comment sections of videos on YouTube, and these comments can then lead to new narratives arising from their reply threads. In this study, we situate this process in the sphere of participatory culture, as defined by Henry Jenkins (2007).

Figure 1: Model 'Encoding and Decoding' (Hall 1973, p. 4)

2. Methodology

2.1 Data Collection

Our two cases - the Russia-Ukraine War and White Paper Protests - were selected due to their recent or ongoing presence in the news as well as their controversial qualities. To study comment sections, eight total videos were chosen based on substantial view counts, subscriber counts, and the size of their comment sections relative to related videos. As the focus of the report concerns comment ecosystems, it was necessary to choose videos with lively comment sections - however, this criteria does not include a plurality of perspectives, geolocations, and languages.

This methodology was constructed to examine how comment sections reveal participatory topographies based on the inference that a high view and comment count provides a substantive, multi-dimensional object of study. As such, the video selection includes choosing two events per case. Additionally, each event would include (1) a video from a traditional news outlet and (2) a non-traditional media outlet, meaning a vlogger, amateur, or individual’s channel. Furthermore, to examine whether the channel has a footprint outside of YouTube, or if it is exclusively YouTube based, a deeper assessment of the channel from which the video was selected was employed. Based on the preliminary research, it was discovered that there was scarcity concerning non-traditional videos regarding the White Paper Protests. A reason for this scarcity may be due to the unavailability of footage, and also the ongoing censorship of YouTube by the CCP (Wong & Williams, 2022). This asymmetry between Russia-Ukraine and White Paper Protests related content, and thus the scale of their datasets, is one limitation of our methodology. Based on this information, Figure 2 shows the data used for this research.

Figure 2: video catalogue conflicts

Ethical considerations were observed in our data collection, following the guidelines of the Association of Internet Researchers’ Ethics (2012, 2019). The publicly accessible data was collected, recorded, analyzed, and presented with care. Thus, it was chosen not to include the usernames and identities of commenters in the comment data collected and presented in this paper.

2.2. Qualitative Analysis

In order to identify the narratives within the chosen YouTube videos and their corresponding comment sections, a content analysis was conducted focussing on the context of the video itself. This implies an analysis focusing on the tone, objective, and themes of the selected videos. From these videos speech was assessed, pulling out quotes (Appendix 1) that provide additional context, essential for the analysis. The content analysis is critical to contextualize possible (counter) narratives found in the YouTube comment sections of each specific video.

A close reading and thematic analysis were used to analyze the original comments and replies found in the comment sections of the selected videos. As the focus of the report concerns analyzing comment sections, the top three most replied-to comments per video were selected. From the top three comments, thirty replies were examined (n=720); ten replies were chosen at the beginning, middle, and end of each reply thread to collect a dimension of temporality. A selective open coding scheme was developed after reading through the chosen comments (Appendix 2). This corpus provides an improved qualitative understanding of the possible narratives that are driven from the interaction between the main comment and the replies found. Thus, the division of the replies into three segments allows for a workable, inclusive sample of the narratives and behaviors in the comment section, accounting for the tendency of these threads to shift and evolve thematically. It is through this process that we identified the emerging themes and practices present in the analyzed comment sections, including ‘confrontation’, ‘conspiracy’, ‘blame’, ‘freedom’, ‘sentiment’, etc.

2.3. Quantitative Analysis

For the quantitative component of the research, a dataset was collected with the Video Information and Comments Module of the YouTube Data Tools (Rieder 2015). This provided us with eight different datasets, one per video, and a total corpus of 83,226 comments. Following, the comment sections of the datasets were imported into ‘4CAT’ and then tokenized to create a binary co-word network for each video. When conducting co-word extraction we filtered out stopwords, as their presence could impede our analyses of these networks.

Co-word datasets were fused together with ‘Gephi’ to create one combined co-word network for each case study. However, due to the high number of nodes of the fused networks, we filtered them by weighted degree of the nodes to simulate a measure of frequency. This resulted in a more concentrated number of nodes to analyze based on the most used words in those comment sections. With the use of the modularity class we ended up with certain word groups that appeared most often next to each other. This approach to conduct semantic co-word networks was previously used to “explore various prominent conversational topics in a large volume of comments, offering an alternative way to see the relationships between the words by visualization” (Kim 2021, p. 6; Drieger 2013).

Our goal with these modes of analyses is not to classify discrete comments but instead to draw out the multiplicity of their themes in order to color our understanding of comment sections. One limitation concerning our co-word analysis became clear here - even when situated in bigrams or trigrams, tokens lose their context thus obfuscating our classification scheme. This is apparent in our semantic network analyses where the word “fake” might have a high frequency with “news” indicating a ‘Media Critique’ or ‘Political Stance & Speculation’, however upon qualitative analysis we find that in some cases the bigram “fake news” is overwhelmingly associated with ‘Incident & Sentiment’ themes. Indeed, as stated in our findings, a comment, or even a single word, is capable of reflecting several dimensions.

3. Findings

This paper focuses on two distinct conflicts, in which the findings reflected similarities in commenting behavior as much as their differences. The comment sections of both conflicts similarly reflected pro/anti-Western, pro/anti-Russian, pro/anti-Ukrainian, and pro/anti-CCP sentiments, suggesting that comment spaces can act as a sort of political theater. As outlined in our methodology, we began our findings with a selective qualitative analysis of a portion of the comment sections associated with the chosen videos. We then employed a network approach to apply our qualitative findings to a larger data set.

By scouring the chosen comment sections through qualitative comment analysis, three main classes of comments were identified: ‘Political Stance & Speculation’, ‘Power, Media, & Economy Critique’, and ‘Incident & Sentiment’ (Appendix 2). Rather than separate comment types by ‘Political Stance’ and ‘Speculation’, for instance, it was chosen to collapse the two as they often accompany one another; such is the case with critiques of ‘Power, Media, & Economy’ as well as discussions of specific ‘Incidents’ and ‘Sentiment’ expression. Within each class are a variety of themes, specific narratives, and user behaviors (see definitions in Appendix 2). However, while these classes are nominally distinct from one another, we find that their themes can often co-exist, or even blur together, within one comment:

I live in Chengdu. The protest in Chengdu lasted only one night. There were about dozens of people. This protest was obviously organized. It used the fire in Urumqi to ignite emotions. The accident was caused by a problem with the community fire system, the door was not locked, But if someone wants to take advantage of the accident, the rumors will spread faster, they spread rumors that the tragedy happened because of the government’s lock-down, and they want use the fire to create a color revolution across the country. The epidemic policy has indeed made life difficult for many people, but people’s real appeal is to adjust the policy rather than subvert the government, because that It will only make our lives more chaotic, but I have no doubt that some people want China to be chaotic, I can’t stop laughing when I hear that the protesters in Chengdu are paid 500rmb once, these tactics are doomed to fail, the only effect is Give BBC NYT RFA more footage to attack China.

(White Paper Protests 2a: Democracy Now!)

As can be seen in this comment the classes ‘Political Stance & Speculation’ and ‘Power, Media, & Economy Critique’ are both found within the comment made under the YouTube channel. By mentioning both of the political stances in regards to the government and calling out to the power claims made, which can be seen in the “that the protesters in Chengdu are paid 500rmb once”.

Figure 3: Narrative distribution expressed in Venn Diagram

With the classes determined, we turned to the distribution of themes and practices across our two cases. As evidenced by our Venn Diagram (Figure 3), the Russia-Ukraine War and White Paper Protests share a large number of our extracted themes. Aside from specific events, few narratives or modes of expression are unique to either case. While Covid-19 conspiracies are absent in the comments of the Russia-Ukraine War, China has been, at minimum, proximate to Covid-19 discourses since the beginning of the pandemic. As such, Covid-19 conspiracies were present in the White Paper Protests comment sections:

Don’t forget that covid virus was created by U.S. biotechnology..... They received money, the United States gave Taiwan and Hong Kong people, and Hong Kong and Taiwan people also received kickbacks in the middle. The United States gives 150-200 US dollars, and they give the locals 300-500 RMB. The Taiwanese also showed his health insurance card on the spot

(White Paper Protests 1b: zaobaosg)

Additionally, among the comments and replies across both Russia-Ukraine and White Paper Protests videos, a large number of comments critique the authenticity of the news, even if it is provided by traditional media sources. In order to add credibility, it was found that users often chose to disclose their identity on their own initiative:

I was a US Marine platoon commander in Vietnam 1968-69. I lost a lot of good Marines that year. Combat is chaos, carnage, and in many ways a lot of luck is involved in staying alive... (cont.)

(Russia-Ukraine War 1a: WarLeaks)

Consistent with the content of their videos, there was very little off-topic discussion in the comments section, with viewers expressing support or ‘shout out’ for their opinions and expressing their emotions. The reflection of this emotional support was exemplified in the comment sections of both conflicts:

I was born in Urumqi, and we moved to Beijing when I was a kid. Unfortunately my mom is stuck in Urumqi right now. She travelled back to Xinjiang to take care of my aging grandma in the summer, and she has been stuck ever since. In the past 3 month she was allowed to exit her building twice. And there is nothing I can do, I haven’t been back for 4 years already, during which my parents already missed my wedding in Canada, and two of my grandparents passed away and I missed their funeral too. F the CCP.

(White Paper Protests 1a: CNN)

R.I.P. to the Ukrainian and Russian souls who fought bravely in this battle. May peace come soon between the Ukrainian and Russian people.

(Russia-Ukraine War 1a: Military Leaks)

As can be seen in the previous paragraph, whether the video was about the White Paper Protests or the Russia-Ukraine War, the type of emotional support shared online does not differ specifically per conflict. Both share similar characteristics by sharing their empathy.

Further, a high presence of insult, mocking and hate speech was explicitly found in the comment sections from the White Paper Protests videos. This distribution is further expressed below in a matrix of the narratives (Figure 4). An example in which mockery is portrayed within the YouTube comments is reflected in the following comment, in which the user calls out the CCP:

The little pink are stupid that they even don’t know how to escape.
Lol you don’t have the negative covid test result so you can’t escape, so Chinese.
Go and fix your ears, the dog of CCP! The evidence in the video is clear that the door is locked. The highest level of lying is lie to yourself. Waiting for the next video about you screaming at home and cannot escape.

(White Paper Protests 1b: zaobaosg)

This comment not only critiques the creator of the YouTube video, but also the CCP.

To further visualize the narratives found in the comments, a matrix was developed using the top three most replied to comments per video (Figure 4). These criteria were chosen on the inference that these comments possess a semantic quality that invites substantial participation, evidenced by robust reply threads. However, this could be due to other factors such as the time of the comment, its proximity to the top of the page, or a more stochastic dimension. Through this matrix, it is demonstrated that politically charged comments have dimensional qualities; for instance, anti-West sentiments do not imply a pro-Russian or pro-CCP stance. On the contrary, it may show the similarities as much as the differences in the narratives. As each quadrant represents the six most common political stances (Russia and CCP stances are combined due to the dichotomy of ‘West versus other’ common across the comment sections), the distribution of certain beliefs and behaviors across issues become visible.

Figure 4: Matrix of semantic co-word analysis YouTube Comment Sections; Legend (figure 4, Narrative Matrix) 1. Blue = Russia-Ukraine (Event 1); 2. Yellow = Russia-Ukraine (Event 2); 3. Red = White Paper Protests (Event 1); 4. Green = White Paper Protests (Event 2)

The majority of our YouTube comments data set, which can be found in Appendix 3, are closely located to the y-axis on the matrix. This indicates that they do not take a polarized political stance. Corroborated by the comments themselves, this indicates that political stances are often not stated outright with great polarity but are rather embedded in proximate expressions and behaviors:

Putins the one that invaded, They are Peaceful n Free,This is Not looking Good,Please ,NATO Protect n Give them Everything to protect the Skys,Its finally time,they need protection.

(Russia-Ukraine War 1b: BBC)

This is fking ridiculous, as a Chinese Canadian, I am disgusted by the silly zero covid policy and Xi’s regime! It’s time for all Chinese people to unite and stand their ground! The desire for freedom should not be restricted!

(White Paper Protests 2a: Democracy Now!)

Both of these comments reflect an embeddedness of opinion, rather than being antithetical in their political stance. Instead both comments call for freedom.

Concerning the Russia-Ukraine War, three comments express moderate anti-Western stances, and embody practices of blame and conspiracy. In congruence, four comments are moderately pro-Russia, where three express sentiment and one more radical conspiracy. Further, three comments fall into the pro-West quadrant where users either express sentiment or call for freedom. However, there is a far greater distribution of ‘pro-’ sentiments and overall heterogeneity in the Russia-Ukraine comments than in those of the White Paper Protests.

In addition, comments around the White Paper Protests are commonly expressed with negativity towards a certain stance using ‘counter-expression’ (meaning that these comments prefer ‘anti-’ stances rather than ‘pro-’) most often directed against the CCP:

The lock down isn't about health, it's about control. The CCP is using Covid restrictions as a tool to control Chinese citizens. They have been changing people’s QR code to red to prevent them from protesting.

(White Paper Protests 2b: Warrior Poet Society)

The comments from the White Paper Protests gravitate largely around practices of blame, conspiracy, and confrontation, while few hover around expressions of sentiment and calls for freedom. While this distribution paints distinct narrative pictures of each comment sphere, it also depicts similarities in the modes of communication (sentiment, freedom, conspiracy, etc.) present in each. These modes manifest differently according to the contrasting ideologies of the commenter, but are nonetheless common commenting practices.

Taking the findings of the qualitative content analysis into account, two comment networks are used, one per case, to reveal classes of semantic meaning present in the comment sections. This data includes original comments as well as their replies. The following categories were recognized in the networks, exposing certain narratives and behavioral kinds. For the Russia-Ukraine War: ‘Sentiment & Incident’, ‘Media Critique’, and ‘Political Stance & Conspiracy’ were identified, which are then divided into pro/anti-Russia, pro/anti-West, and pro/anti-Ukraine. For the White paper protests: ‘Political Stance & Conspiracy’, ‘Sentiment & Incident’, ‘Power & Economy Critique’ and ‘Covid & Lockdown’ are identified, which are then divided into pro/anti-CCP and pro/anti-West. By using the tool ‘Gephi’, a semantic co-word network was developed for each conflict.

The semantic co-word network concerning the Russia-Ukraine War can be seen in Figure 5. Although the network is dense due to the close connection of the words and their many edges, three comment classes emerge within the network. After a close investigation of the three sub clusters we came up with those three themes which could be identified due to the appearance of certain words in the same comments: ‘Sentiment & Incident’, ‘Media Critique’, and ‘Political Stance & Conspiracy’. However those categories are still interwoven into each other as can be seen by the different coloured dots within the classes.