Blazing Red, Paper White. How does YouTube mediate China's White Paper Protest? An analysis of the bilingual news spheres

Team members: In A Hwang, Xiaolu Ji, Anna Poggi, Bingwen (Serena) Wang, Yi Zhang

Summary of Key Findings

For the Chinese query, we generated from the YouTube data tool approximately 480 Chinese- speaking videos. The findings show the most active channels are: “美国之音中文网” (a US state sponsored Chinese-speaking news channel), “新球电台” (a Japanese branch of an anti-communist organization), “BBC News 中文” (a BBC Chinese news program based in the UK), and “TVBS” (Taiwanese channels with niche reportings). The formats or genres mostly consist of serious news reporting, commentary, live streaming shorts without editing. This affirms the censorship in China since none of the channels for the Chinese query were from Mainland China but rather Chinese- speaking communities sharing the news from abroad.

For the English query, we performed the same search and generated 2082 videos. From the dataset, we discovered that most active channels are Indian news outlets with daily detailed coverage. The top 5 most viewed videos are from reputational global English-speaking news organizations, which do not align with the most active Indian channels. And the top viewed videos contain neither personal blogs nor young channels. This can be considered a demonstration of the fact that the protest in China has become a matter of importance and resonance with the rest of the world through YouTube.

1. Introduction

From the first Covid wave in 2020, China has faced some of the most draconian Covid lockdowns. Upon the re-election of Xi Jinping as the leader of Chinese Communist Party (CCP) in October 2022, tensions escalated and social unrest began to surface. Amongst the protests, a blank sheet of A4 paper assumed the symbol and an umbrella term for the nationwide dissents. Echoing its predecessor in the 2019 Hong Kong protest, the “white paper” represents what protesters wish to say but cannot due to strict censorship. It presents a dilemma to the CCP authorities – whether to repress the protesters for saying literally nothing.

YouTube is a platform known for user generated content but in the last few years several studies have been made to better understand how mis/disinformation spreads in the platform. Another relevant aspect that has only recently been looked into is how YouTube mediates events and if the spread of information is mediated. The rise of news channels and citizen journalism around events, the system with which YouTube mediates events becomes even more unclear. Thus, this research is auditing the reports of the White Paper Protest in China on YouTube and the results the platform offers.

Online protests are a result of the spread of social media and the large reach these new forms of communication have (Meek, 2011). Social media gives people the chance of sharing information and reaching a large number of users that can resonate with it. The line between online and offline, cyberspace and real life is becoming exponentially more blurred due to the strong connection they share (Meek, 2011). Meek (Meek, 2011) discusses how the online and offline are strictly connected and it is difficult to separate them. Using the example of the flash-mobs to show how something that was born online had a real effect offline, showing how mobilizing people online can lead to people taking action offline. People can make relevant connections through social media and these ties can unite people all over the world and make regional/personal issues global (Poell, 2014). At the same time, these online ties help to spread information and issues that have the potential of helping people in need. As a user generated content platform, YouTube transcends space and therefore unites people from all over the world and helps them come together over issues of shared importance both in amateur form as citizen journalism and in more professional approaches such as news companies moving to the platform (Meek, 2011).

This paper starts with the datasets we retrieved and the results we gathered from them. Using the methods of time analysis, content analysis and network analysis, we have answered our research questions that focus on how YouTube mediates current events and in particular, the White Paper Protest. In the discussion part, we compare our results with relevant literature in order to explain them and connect them to the bigger frame. Tables and figures have been added in the results for a better understanding and readability of the results.

2. Initial Data Sets

In order to capture the videos, we designed two search queries “China Covid Protest” and “封控 抗议 (Lock-down Protest)”. These two queries can be seen as equivalent, and generate more comprehensive results referring to the event compared to queries “white paper protest” or “白纸 运动 (white paper movement)”. We used the Video List module of YouTube Data Tools (Rieder, 2015) to collect all videos generated by the two search queries. By limiting the search to videos published between 2022-10-13 and 2023-01-09 (from Sitongqiao Protest to the day on which data was collected) and setting the iteration to the maximum, two datasets (2082 English videos and 480 Chinese videos in total) were collected for further analysis. We also combined the two datasets into one and generated a channel network based on the video recommendation relations, with detailed channel information, using the Video Network and Channel Info module of YouTube Data Tools (Rieder, 2015).

3. Research Questions

This research investigates how YouTube mediates current events through the case of the White Paper Protest in China. It aims to map an overview of the videos related to the protests on YouTube with a focus on English and Chinese contents. More specifically, we ask: (1) How timely was the protest “reported”? (2) How is the protest presented by channels within the bilingual spheres and their corresponding political standing? (3) How are the protest-related videos mediated by co- watching history and recommendation algorithms?

4. Methodology

For the time analysis we extracted data as explained in section 2 of the paper. Once we had the .csv files, we transferred them on Excel and started to analyze them. Pivot tables on Excel were used to gather the number of posts in total per month and per day, along with all the necessary information in order to complete the histogram of how the events reflected on YouTube but also to rank the channels and videos in the order of activity and popularity. Firstly, we checked the number of posts in total and then divided them into months and later into days to see which periods were the most active and if the spikes in the datasets coincided with the timeline of the protests. Through the pivot table, it was also possible to extract most representative channels and videos for content analysis to further investigate our second research question.

After collecting the data and creating the pivot tables, we sorted the dataset based on the ranking of the most active channels along with the most viewed videos from both English and Chinese queries. From there, we selected the top five most active channels and the top 5 most viewed videos as our sample for content analysis.

We then conducted the qualitative approach by auditing the above-mentioned sample. We focused on information including the genre of the videos, reporting angle of the content, political standings, channel category, location, channel age, number of subscribers, number of comments and likes. The purpose of the analysis is to tease out potential alignment between the most active channels and the top videos, as well as the narratives revolving around the event.

As stated before, a channel network was generated based on the video recommendation relations combining the two datasets. We then used Gephi to conduct a network analysis of recommendation relations among channels. We set the node color according to channel locations and the node size according to the subscriber count, to identify the “central” channels in each cluster and the “bridging” channels between the bilingual spheres.

5. Findings

RESEARCH QUESTION 1 – How timely was the protest “reported”?

The timeline of the protest is reflected in the results of the datasets. November was the month with the majority of the videos published due to the Urumuqi fire that broke out on 24th of November 2022, which serves as a direct trigger of the protests. Most videos were published closely following the incident.

Figure 1 - English Query

Figure 2 - Chinese Query

Furthermore, an interesting point to observe is the interaction of the users with the channels. In figure 3 and 4, the users interact differently with the videos generated by the two queries. In fact, people responding to the results of the Chinese query tend to like the videos more than the English query, where the users prefer to comment.

Figure 3 - English Query Audience Interaction

Figure 4 - Chinese Query Audience Interaction

Interestingly, most of the videos were uploaded one or two days after the events, indicating that the peaks of posting occurred around key events. The Chinese query, although producing fewer results, shows that the spike in the number of videos is closer to the event than the English query in terms of temporality.

Figure 5 & 6 - Videos Published in October 2022

As shown in figure 5 and 6, the spike in the English query is a day later than in the Chinese one, whereas the numbers of videos published on the 13th of October 2022 are similar. On October 13th, the Sitongqiao Protest jumpstarted the chain of events, where people demanded the removal of “dictatorship thief” Xi Jinping. This is reflected in the spike on the same day of the protest in the Chinese query, which is also reflected in the English dataset, with the spike happening a day later. Regardless of the spike the next day, the numbers of videos published on October 13th between the two queries are similar.

Figure 7 & 8 – Videos Published in November 2022

In both sets, the month of November was the most active month given the fire incident. However, days of protests before the fire do not present more spikes except for the Guangzhou Protest on November 14th. On the days in which nothing happened, the Chinese query seems to offer a slightly bigger number of videos whereas on the days of the protests and the ones following the fire, the English query is more active in regard to video posting.

Figure 9 & 10 Videos Published in December 2022

The wave of videos in December stays active for the English query but slowly decreases. At the same time, both queries diminish exponentially in the number of videos published during this month. The data retrieved for the Chinese query ends in December whereas there are still videos published in January for the English query (figure 11).

Figure 11 – Videos Published in January 2022

RESEARCH QUESTION 2 – How is the protest presented by channels within the bilingual spheres and their corresponding political standing?

Once we gathered the histogram, we conducted content analysis per query as illustrated in the methodology section.

Figure 12 – Most Active Channels in the Chinese query

The most active Chinese-speaking channels are based out of the US, Japan, UK, and Taiwan, rather than Mainland China as a result of the Great Firewall. Furthermore, the number of posts of the channels do not necessarily align with the views or the number of subscribers. Most of the videos are created in the format of serious news reporting, commentary, or live streaming shorts without editing. Interestingly, the most viewed video ranked the fifth in the Chinese dataset is from a personal blog (“國強啊國強”) with only 3.69k subscribers. The channel creator uploaded a 30- second short that contained a first-person viewpoint of the protest event and it went viral.

Figure 13 – Most Watched in the English Query

From the English dataset, the most noticeable aspect is how the most active channels are predominantly Indian media outlets, which share detailed daily coverage. Mainstream news institutes such as CNN and Yahoo Finance are also among the most active channels, despite that they are ranked much lower. Similar to the Chinese results, the top five most viewed videos do not align with the most active channels and unlike the results in the Chinese dataset, they contain neither personal blogs nor young channels.

RESEARCH QUESTION 3 – How are the protest-related videos mediated by co-watching history and recommendation algorithms?

Figure 14. Channel network based video recommendation relations

The network visualization shows how different channels distribute in bilingual clusters connected by recommendation relations. The smaller, upper cluster is mostly made of channels producing Chinese content, while the bigger cluster below is composed of English channels. Among Chinese- speaking channels, the central and dominant channels are Taiwanese media outlets, surrounded by channels located in Hong Kong and the United States. In the English cluster, channels from India, USA, and the UK take the central position, with the number of channels located in India ranking the highest among all areas. Although most channels are established news organizations including big names such as CNN, BBC News, TVBS, India Today, it is found that the “bridging” channels connecting the two main clusters are mostly highly personal accounts. These channels are originally not made for news production but personal usage. They have much fewer subscribers compared to organizational channels, but one or several videos containing primary material went viral during the event.

6. Discussion

Thomas Poell analyzes the history of online protests in restrictive countries such as Iran, Egypt, Tunisia, and China, focusing on how the control over who can access what on the internet can influence the way in which people express their dissent (Poell, 2014). His line of thought coincides with our findings since all the Chinese speaking videos were not from Mainland China but Chinese-speaking communities abroad that commented on the events. Poell (2014) explains how China, Belarus and Iran use strong bans on internet content in order to control the sharing of information. The author argues that social media are never neutral but embedded in a bigger frame in which the dictatorial regimes control the internet and who can have access to what. Poell (2014) analyzes that just giving more freedom to activists online is not enough because it is situated within a frame of “communicative capitalism” and thus lifting the ban is inadequate. Sharing videos and information online is far from effective in an era that is as digital as ours. In fact, Meek (2011) discusses how the online and offline are strictly connected and the importance of gathering people around an issue in order to resonate and gain momentum. The most viewed videos and most active channels offer information about the events and their relevance, resonating with people all around the world. The blank paper became the symbol of a political movement, extending to cover aspects beyond a mere Covid protest. The White Paper Protest represents the need to share what is happening and take action in a situation that is unbearable.

The histograms, which show the timeline of the events and the integration of YouTube videos uploaded on the crucial days, demonstrate the relation between the offline and online phenomena. On October 13th, the protest started with the Beijing Sitongqiao protest (四通桥抗 议). As shown by the histograms both queries offered results on that day but unlike the Chinese query, the English had its peak the next day, showing a delay in news release (figure 5 & 6) . Similarly, in figure 7 and 8, the table of the query in Chinese shows an increase in posts on the 25th and 26th of November followed by the peak on the 27th and 28th. However, the table presenting the English query shows a crucial increase in the number of posts on the 27th and 28th of November and then it reaches its peak on the 29th, continuing a delay observed above.

RQ 1 According to these images, it is possible to observe the difference in the reaction time between the English and Chinese channels. An important factor is the amount of content posted by the two queries: the English channels exceeded the number of posts uploaded by Chinese channels due to the Great Firewall implemented by China. As mentioned by Xu (2014), the Chinese authorities have been governing the information flow through multiple measures, preventing most users in Mainland China from accessing Western mainstream websites and platforms. Therefore, Chinese channels show a limited amount of content uploaded on YouTube compared to English query channels. However, Chinese channels show a faster reaction time to the events than English channels, which is noticeable by looking at when the content was uploaded online.

RQ 2 Another significant finding is that for both languages, the video view counts do not necessarily align with the channel’s number of posts on the event or its number of subscribers. First-hand information from the ground tends to be reused and relayed across channels while integrated into news reporting, commentaries, and expert analyses. Live streaming shorts received broad exposure despite the channel’s small number of subscribers. It can be observed that primary materials, especially the live scenes, generate more visual impact and give rise to a greater range of resonance amongst the audience. In addition, Indian channels are most active among the English channels, yet the view counts do not adhere to the channel subscribers and video counts: despite the activeness of Indian channels, none of their content reaches high view counts. One of the reasons might be the niche audience they are targeting, and the language (sometimes reporting in Hindi) they are using.

RQ 3 The channels network also shows how content is mediated by platform co-watching history and recommendation algorithms, part of the YouTube platform infrastructure. The discrepancy between clusters of two languages indicates how the recommendation system classifies videos by different languages. While the most recommended channels in each cluster are still established news organizations with substantial subscribers, it is surprising to find the bridging channels between two clusters to be highly personal accounts with little fan base but viral videos. This can be attributed to the co-watching history generated by users who consume both Chinese and English videos.

7. Conclusion

In conclusion, it is possible to notice how YouTube mediates to some extent what is happening during the White Paper Protest. The fact that the most viewed videos do not coincide with the most active channels shows how some channels have a bigger reach than others. Furthermore, the amount of views in the shorts shows the necessity to share news fast and the rising of citizen journalism. On the other hand, from what we observed, YouTube allows users to publish and talk about the issue in a way that allows more and more people to resonate with and respond to it. The histograms show how the events reported on YouTube reflect the events in real life and how the two are strictly connected.

The limitations of this research are connected with both the wording of the queries and the number of reiterations. For future research more queries could be made to see if the results differ. Furthermore, the mediation of the results could be compared with other online protests or events which YouTube could have mediated. Comparing how different protests have been or are moderated through YouTube might help get a better understanding of how the platform mediates and what kind of videos it recommends when looking into current events. Furthermore, this research does not take into consideration the overlap of multilingual content, for example, the English query could generate video content in Chinese and different languages. In this case, a more comprehensive review of the datasets would contribute to future research.

References

Meek, D. (2011). YouTube and Social Movements: A phenomenological analysis of participation, events and Cyberplace. Antipode, 44(4), 1429–1448. https://doi.org/10.1111/j.1467-8330.2011.00942.x

Poell, T. (2014). Social Media Activism and State Censorship. In Social media, politics and the state (pp. 189-206). Routledge.

Rieder, Bernhard. 2015. “!YouTube Data Tools (Version 1.24).” https://tools.digitalmethods.net/netvizz/!YouTube/.

Xu, B. (2014). Media Censorship in China. Council on Foreign Relations, 1–6.

-- BernRieder - 30 Jan 2023