Abstract
Introduction
In March 2023, a social media plea gained viral traction when a post-millennial university student lamented the shutdown of a short video app that hosted her final recordings with her late grandmother, in which traditional crafts were passed down. Upon discovering the app's removal and failing to download the video, she turned to social media to reach the app's company, hoping to recover the cherished content. Her story, shared on her social media, resonated widely, eliciting an outpouring of empathetic support from the online community. Netizens collectively endeavored to amplify the visibility of her post, employing a technique known as “boosting popularity” (
As netizens’ interactions with algorithms become more frequent and nuanced, the notion of the “algorithmic unconscious” (Thrift, 2005) is increasingly challenged. Netizens are progressively recognizing their own agency (Jones, 2023) and forming folk theories and imaginations of algorithms based on their intuitive understanding of operational rules (Bucher, 2017; Karizat et al., 2021; Prey and Esteve-Del-Valle, 2024). These folk theories become the basis for resistance against algorithm platforms (Richter and Ye, 2024; Velkova and Kaun, 2021), enabling netizens to subvert algorithm arrangements. Algorithmic resistance refers to individual or collective actions taken by netizens within the algorithmic framework to oppose, challenge, or subvert the power structures embedded in algorithmic systems, often driven by concerns over bias, discrimination, or control (Bonini and Treré, 2024: 23; Velkova and Kaun, 2021).
Although existing research on this topic is quite extensive, it predominantly focuses on incidental resistance movements (Velkova and Kaun, 2021) and specific groups (Cotter, 2019; He and Li, 2023; Sun and Yin, 2024; Veen et al., 2020; Yin, 2020) or individuals (Bucher, 2017). It overlooks the broader collective algorithmic resistance among netizens. Maris et al. (2024) have conducted research on algorithmic mutual assistance on TikTok, but further investigation is needed to explore. Netizens, as citizens in cyberspace (Cao et al., 2022), view the internet as an extension of the public sphere and use online interactions to promote social change or collective action (Hauben, 1997). They engage in collective actions that aim to disrupt the algorithms shaping visibility and influence the flow of information. This collective resistance reflects a form of civic engagement where netizens not only resist algorithmic control but also actively shape the digital public sphere (Hindman, 2009). Thus, this study builds upon previous research and focuses on two core research questions: What algorithmic folk theories for boosting popularity have netizens developed? How do folk theories influence netizens’ algorithmic resistance strategies and shape the complex interactive dynamics within their algorithmic resistance?
This study selects three internet hotspots in China from late 2023 to early 2024 as case studies. Using a netnographic approach, it focuses on Douyin, Xiaohongshu, and Weibo as sites for online field observation. It conducts thematic analysis on the popularity-boosting comments and discussions about these practices by netizens.
The study found that during the competition for visibility in boosting popularity, netizens developed cross-platform algorithmic folk theories. Their algorithmic resistance primarily follows two pathways: within the algorithmic framework and beyond it. Overall, netizens’ collective algorithmic resistance, whether within or beyond the algorithmic framework, is characterized by frequent conflicts, a lack of effective internal negotiation mechanisms, and weak group cohesion. Within the algorithmic framework, conflicts over folk theories lead to internal group attacks, which undermine the consistency of collective algorithmic resistance. Additionally, visibility strategies that disregard the substantive meaning of comments disrupt effective public discourse in comment sections, introducing new conflicts. Some netizens have even called for platforms to regulate popularity-boosting comments. Even where consensus on algorithmic folk theories exists, some netizens withdraw from participation due to the interference of personalized recommendation algorithms, which continue to push related content after visibility support. Furthermore, beyond the algorithmic framework, the manual replication of keywords and event summaries within comment sections has engendered dissatisfaction among some netizens, resulting in their diminished interest in monitoring related online events or engaging in competitions for visibility. The internal conflicts within the netizen community suggest that many challenges remain in negotiating and resolving collective algorithmic resistance. Furthermore, despite intense internal conflicts, netizens appear to exclude external negotiations with platforms or the state from their algorithmic resistance efforts. This exclusion limits their pursuit of broader algorithm power.
This study focuses on the interaction between netizens’ algorithmic folk theories and algorithmic resistance in the context of trending online events, as well as the complex interactions of netizens during algorithmic resistance. First, it expands the focus on actors, highlighting that, in contrast to the coherence of algorithmic folk theories formed by individuals and specific small groups, the development of netizens’ algorithmic folk theories involves both negotiation over consistencies and conflict over inconsistencies. These conflicting folk theories directly affect the cohesion of collective actions among netizens, serving as a valuable starting point for a more comprehensive understanding of the power struggle between algorithms and netizens. Second, resistance strategies that transcend the algorithmic framework further extend current research on algorithmic resistance, providing new perspectives for future studies on human-algorithm interaction. Finally, the internal power struggles within groups engaged in collective algorithmic resistance offer fresh insights for understanding the broader dynamics of netizen interactions in algorithmic resistance.
Literature review
Netizens as active agents in algorithm interaction
In the evolving dynamics of netizen-algorithm interaction on social media platforms, netizens are increasingly cognizant of the critical role their behaviors play in influencing algorithm outcomes (Jones, 2023). No longer mere pawns of algorithm dictates, netizens navigate through the social and existential quandaries algorithms present (Bucher, 2017; Gillespie, 2016), moving beyond the confines of algorithmically arranged interactions (Amoore, 2009). This awareness has given rise to what is known as “algorithmic resistance,” referring to individual or collective actions taken by netizens within the algorithmic framework to oppose, challenge, or subvert the power structures embedded in algorithmic systems, often driven by concerns over bias, discrimination, or control (Bonini and Treré, 2024: 23; Velkova and Kaun, 2021). In such resistance action, netizens, as proactive participants, strategically exert control over their digital engagements, deliberately planning their actions within the algorithmic frameworks to achieve their objectives on these platforms (Jones, 2023; Karakayali et al., 2018; Velkova and Kaun, 2021).
Velkova and Kaun (2021) investigated the proactive resistance movements that emerge as netizens collectively work to address the racial and gender biases inherent in media algorithms (Noble, 2018). Through the “White World Web” movement, they explored the political dimensions of algorithmic rectification (Velkova and Kaun, 2021). Bucher (2017) explored how individuals devise pre-emptive interaction strategies to synchronize with algorithmic logic. Moreover, significant research has delved into how professions engaged in algorithmic interactions adapt to achieve their objectives. Studies have analyzed how digital influencers and social media bloggers (Bishop, 2019; Cotter, 2019; Meara, 2019; Richter and Ye, 2024), along with fan communities (He and Li, 2023; Yin, 2020; Zhang et al., 2022), deploy savvy strategies to interpret and align with platform algorithms, thereby enhancing their online visibility. Additionally, observations have been made on how service industry workers, such as delivery personnel (Veen et al., 2020), effectively algorithmic resistance to safeguard their interests. Furthermore, research examined the dynamic responses of minority groups to algorithmic systems, highlighting their innovative interactions and modes of resistance (Sun and Yin, 2024; Zhao, 2023).
Despite the extensive literature on algorithmic resistance, most studies predominantly focus on sporadic movements of resistance or the daily algorithmic interactions of individuals or specialized groups, paying insufficient attention to the collective and widespread algorithmic resistance of broader netizens. While Maris et al. (2024) have elaborated on algorithmic mutual aid on platforms like TikTok, highlighting the broader collective actions of netizens in response to algorithmic influences, research in this area remains limited. Therefore, further exploration of broader algorithmic resistance among diverse groups of netizens across various platforms is necessary. This study specifically examines the collective behaviors and complex interactions of netizens in boosting popularity on Chinese social media platforms. It draws on a series of prominent online events from late 2023 to early 2024 as a lens to explore the netizens’ algorithmic resistance practices within collective participation and their intricate interactions. This enriches the understanding of broader collective algorithmic resistance among netizens.
Algorithmic folk theories as behavioral bases
In interactions between users and algorithms, whether it involves resisting the system, the behavioral foundation is rooted in experiential perception of the algorithm's rules. Despite the complexity of platform algorithms rendering them as “black boxes” inaccessible to the users (Pasquale, 2015), users gradually deduce the operational rules of algorithms through trials at the input and output stages (Zhao, 2022). These rules form what are known as algorithmic folk theories (Eslami et al., 2015). Folk theories are intuitive, informal theories about the evolution of objects or issues, grounded in practice and experience, employed to explain the outcomes, effects, or consequences of technological systems (DeVito et al., 2017; Rip, 2006), and are continuously revised as people's perceptions change (Siles et al., 2020; Ytre-Arne and Moe, 2021). Such theories shape the behaviors of those who hold them (Gelman and Legare, 2011). Moreover, algorithmic folk theories are not merely summaries of individual experiential knowledge; they also spread and are negotiated within group interactions, collaboratively constructed by communities (DeVito et al., 2018; Zhang et al., 2022).
Similar to algorithmic folk theories, Bucher (2017) introduced the concept of algorithmic imaginaries, which explore users’ acceptance and utilization of algorithm processes, further elucidating what algorithms are, what they should be, how they operate, and what possibilities these perceptions in turn enable. Recent studies have expanded the application scenarios of algorithmic imaginaries, arguing that research on algorithmic imaginaries should not be confined solely to users’ perceptions on platforms; the platforms’ perceptions of users are equally critical, with both sets of imaginaries collaboratively shaping interactions between users and platforms, as well as among users themselves on social media (Schulz, 2023).
Grounded in specific practical application scenarios, research on people's sociotechnical imaginaries of algorithms, algorithmic folk theories formed based on practical experience, and the differentiated algorithmic practices they induce has been extensive (e.g., Bucher, 2017; Karizat et al., 2021; Low et al., 2023; Prey and Esteve-Del-Valle, 2024). For instance, in areas like misinformation and finance, the capabilities of algorithms and their operational roles are positioned differently (Wijermars and Makhortykh, 2022). A survey among Norwegian users proposed five folk theories of algorithms: Restrictive, utilitarian, simplifying, invisible, and exploitative, viewing algorithms as both annoying and inevitable (Ytre-Arne and Moe, 2021). In the context of China's male homosexual community, algorithms are perceived as both expellers and protectors (Zhao, 2023). Gay men use algorithmic filters to create a protective shield around their community, safeguarding them from external intrusions and facilitating the establishment of group connections (Zhao, 2023). The above folk theories have revealed insights into users’ differentiated perceptions of algorithms.
While existing explorations of folk theories are extensive, further investigation into diverse folk theories originating from various contexts is warranted to enhance understanding of the interactions between algorithms and people (Hargittai et al., 2020). This study, rooted in the Chinese context, aims to explore the folk theories developed by Chinese netizens for increasing the visibility of posts in internet trending events through popularity-boosting comments. Moreover, it examines how these algorithmic folk theories, which underpin netizens’ resistance strategies, influence their algorithmic resistance behaviors in boosting popularity.
Therefore, this study aims to address the following key research questions:
Q1: What algorithmic folk theories for boosting popularity have netizens developed?
Q2: How do folk theories influence netizens’ algorithmic resistance strategies and shape the complex interactive dynamics within their algorithmic resistance?
Method
This study employs a netnographic approach and conducts a thematic analysis of the data. Specifically, the author observed that from late 2023 to early 2024, a series of significant online events in China sparked widespread discussions and popularity-boosting behavior on social media platforms. This period offered an excellent opportunity to examine the strategies of visibility competition among netizens in the comment sections. Moreover, the selection of a netnographic research method was motivated by the belief that non-intrusive observational data can more accurately depict users’ genuine attitudes and behaviors in their natural environment, as compared to the inherent social desirability bias present in self-reported data from interviews (Zhu, 2024).
To bolster the credibility of the research findings, the study employed a triangulation approach (Denzin, 2009; Natow, 2020) in selecting both events and platforms. Initially, this study incorporates the analysis of three online trending events that occurred sequentially and garnered widespread social attention. These included: The event students at Zhongshan Second Hospital being diagnosed with cancer 1 , which became a focal point online on November 7, 2023; the academic fraud incident reported by a student at Huazhong Agricultural University on January 16, 2024, which drew significant attention due to the frequent mentions of a high-priced feed containing “Garcinol,” referred to as the “Garcinol Incident” 2 ; and January 30, 2024, incident in a Guiyang residential area where a malfunctioning fire hydrant led to one fatality. 3 The rationale for selecting these three online hotspot events is twofold. Firstly, these events occurred sequentially within the observation period of this study. Secondly, each of these incidents garnered widespread public attention and involved significant public interests, including life safety, teacher–student relationships, misuse of research funds, and public fire safety. These topics resonated strongly with the public, leading to extensive engagement and interaction on social media platforms. The high level of netizen activity and commentary provides a rich dataset of netizen interactions and reactions, which is invaluable for this study on exploring netizens’ behavior in boosting popularity.
Following this, three widely used social media platforms in China were selected for observation: Douyin (the Chinese version of TikTok), Xiaohongshu (commonly known as the Chinese Instagram), and Weibo (similar to Twitter). The selection of these platforms is based on the unique characteristics of their user bases, as well as their extensive reach and influence within China, positioning them as key players in Chinese online public discourse (Fung and Hu, 2022; Li, 2010; Zhang, 2020), and ensuring the credibility and universality of the findings. However, the study also recognizes that it does not encompass the entire diverse landscape of social media in China, indicating further research and verification are needed in future studies.
On each platform, three posts related to each incident were randomly selected, and data were gathered from the comments sections, culminating in a total of 18 posts analyzed. The reason for this approach is that preliminary netnographic observations revealed a high degree of similarity in netizens’ behavior across different video comment sections. Additionally, this study conducted a comparative analysis of an extra comment section for each event on each platform, confirming that no new content emerged. The data collection was conducted using a Python-based crawler tool, from November 8, 2023, to May 16, 2024, encompassing the duration of the events, with a total of 16,953 comments gathered. The distribution was as follows: 750 comments from Weibo, 1034 from Xiaohongshu, and 15,169 from Douyin. Despite significant variations in data volume, a preliminary assessment of a random sample of 100 comments from each platform revealed a high degree of similarity in netizen engagement behavior across platforms. This suggests that the sample exhibits limited heterogeneity. Therefore, this study employs quota sampling to extract 150 comments from each platform for each incident. Of these, 50 comments per platform per incident are used for subsequent coding saturation checks, resulting in a final dataset of 900 comments for analysis.
In addition to examining popularity-boosting comments in comment sections, the author also collected discussions on the internet about these behaviors. Using keywords “boosting popularity (顶热度)”, “comment for boosting popularity (顶热度专用评论)”, and “boosting popularity specific (顶热度专用)”, after excluding data with empty content, data was gathered from Douyin, Xiaohongshu, and Weibo, totaling 5642 items from April 14, 2023, to May 15, 2024. The data includes 102 entries from Xiaohongshu, 190 entries from Weibo, and 5350 entries from Douyin. Through preliminary reading, it was observed that despite significant variations in the number of data collected across different platforms, the content exhibited high similarity. This suggests that the sample exhibits limited heterogeneity. Consequently, the same quota sampling strategy was implemented. Specifically, all 102 entries from Xiaohongshu and 190 entries from Weibo were included in the analysis. From the Douyin platform, 300 entries were randomly selected, with 200 used for analysis and 100 for saturation checks. Consequently, the final dataset included 492 entries for analysis.
In this study, a total of 1392 data entries were included for coding, of which 900 were popularity-boosting comments, and an additional 492 were discussions by netizens on social media regarding this phenomenon. The author, in collaboration with an employed coder, conducted open coding, axial coding, and selective coding. The coding was performed within the theoretical frameworks of algorithmic folk theory and algorithmic resistance. Initially, two coders independently carried out the coding, then discussed discrepancies to reach a consensus. The coding was finalized after several rounds.
It is important to note that although the data collected for this research was publicly available online, ethical considerations were paramount. Prior to coding, personal identifiers such as commenter's usernames and real names were anonymized. Additionally, all Chinese quotations were translated by the author, and any personal information in the comments was redacted to maintain confidentiality.
Findings and discussion
It is commendable that the practice of popularity-boosting by netizens effectively maintains the lifecycle of posts (Figure 1). Some threads receive continuous updates in their comment sections for as long as five months, thereby reaching a broader audience. In the three cases included in this study, many netizens learned about the incidents two, three, or even four months after they occurred. They noted that despite being active social media users, they had not heard about these events until they saw the posts.
Comment distribution percentages by month for the fire hydrant, Garcinol, and Zhongshan second hospital incidents, across Weibo, Xiaohongshu, and Douyin. Note: The percentage is calculated by dividing the number of comments for each event in each month on each platform by the total number of comments for that event on that platform.
Popularity-boosting comments indeed provide a certain level of online social information and emotional support for the individuals involved, aiding them in coping with the stress of the events (Rains and Young, 2009; Wright, 2016). In terms of informational support, netizens are dedicated to clarifying misinformation and biased narratives, directing public attention to the critical aspects of the events. For instance, in the “fire hydrant incident”, popularity-boosting comments repeatedly emphasized not only the fire caused by fireworks but more importantly the inoperability of the fire hydrant, the property management's compliance with fire safety inspections, and the broader concerns about similar issues nationwide. On the emotional support front, the act of engaging in popularity-boosting commenting itself represents emotional support for the affected individuals. Some netizens expressed, “Even just having more comments would help alleviate this feeling of helplessness that is overwhelming me”; others mentioned, “I wonder if my efforts to keep the post active are useful, but even such comments make me hope more people would help. It's so frustrating.” Additionally, popularity-boosting comments often provide supportive and comforting words, demonstrating empathy with phrases like “hang in there” and “justice must be served for the victims.”
Furthermore, the study reveals that netizens’ algorithmic folk theories exhibit characteristics of cross-platform sharing. Netizens disseminate the same algorithmic folk theories and advocate similar forms of algorithmic resistance across various platforms. This contrasts with specific groups like influencers, who develop more nuanced folk theories and targeted strategies for multi-platform algorithmic resistance.
Moreover, netizens’ actions for visibility in algorithmic resistance unfold along two paths: Within and beyond algorithmic frameworks with the ultimate aim of increasing dissemination and breaking through algorithmic filter bubbles.
Algorithmic folk theories and algorithmic resistance within algorithmic frameworks
Consensus folk theory: Increasing engagement to enhance visibility
Social media platforms’ recommendation algorithms operate on the “traffic pool” logic, where content visibility is directly influenced by user interactions (Zhao, 2022). Upon posting on a social media platform, the algorithm initially assigns the content to a preliminary traffic pool based on factors like content tags and the poster's social relationships, exposing it to a select user cohort. The content's future dissemination into larger traffic pools depends on its performance metrics during this initial exposure, including shares, likes, comments, and follows. If these engagement indicators suggest strong user interaction, the algorithm deems the content high-quality and increases its visibility by recommending it to a broader audience. Conversely, if the content fails to meet these benchmarks, its visibility remains confined to the initial pool.
Therefore, given the logic of algorithmic distribution through traffic pools, netizens generally believe that increasing interaction metrics is crucial for breaking through the hierarchical constraints of traffic pools, thus achieving widespread dissemination and enhancing visibility. Consequently, netizens actively engage in and advocate for more people to like, save, share, and comment on posts, in order to enhance the posts’ quantitative metrics. This, in turn, helps the posts enter larger traffic pools, offering them the chance to reach a broader audience.
However, it is noteworthy that the ensuing disturbances brought by popularity-boosting practices to individuals can lead to some netizens avoiding hot topics, thereby hindering their sustained and deep participation. This occurs because, for individuals, while supporting posts through quantitative metrics, they also signal to the platform their personal content preferences. Subsequently, the personalized recommendation system relentlessly cycles through similar content, causing annoyance to the individual. Despite netizens’ attempts to escape this cycle by clicking “not interested,” the outcomes are often unsatisfactory. As one shared, “After initially liking it and then marking several as uninteresting, it doesn’t take long before they start pushing the same stuff at me again, which is really painful.” This leads some netizens into a dilemma between supporting and ignoring posts. Driven by sympathy for the individuals involved in online hot topics, they wish to participate and lend their support. However, considering the subsequent annoyance of endless similar content pushed by the platform, they are cautious about whether to continue their support. Ultimately, some choose to withdraw due to the distress caused by the algorithmic recommendations, opting to ignore such content and no longer participate or read further. Thus, the platforms’ personalized recommendation algorithm initially deters some netizens from participating in algorithmic resistance collectively.
Conflicting folk theories I: Do comment quality or comment quantity more significantly influence visibility?
Although netizens commonly believe that increasing the number of comments can enhance the posts’ interaction metrics, thereby boosting their visibility, there is a divergence of opinions regarding which type of comments actually achieve this objective.
One folk theory posits that algorithms primarily determine whether to promote posts to a broader audience based on the quantity of comments, thus deeming the quality of comments irrelevant. Proponents of this theory assert that “the volume of data is key.” Consequently, their algorithmic resistance strategies focus on increasing comment volume. This is achieved by mass-replicating specific comments designed for popularity boosting, utilizing minimal effort from commenters to maximize interaction metrics. Specifically, this includes three types of comments (Table 1).
Three types of popularity-boosting comments.
Note: For the third type, it is posited that the frequent appearance of specific keywords prompts algorithms to recognize these terms as trending, thereby enhancing the visibility of related posts containing these keywords. Typically, this involves repeatedly mentioning core keywords in the comments or elaborating on narrative details pertinent to the event.
Another folk theory contends that the quality of comments is crucial for evaluating posts, as it determines whether a platform will promote the post to a broader traffic pool or restrict its reach. Proponents argue that “a large volume of repetitive, meaningless comments causes the platform to classify the post as low-quality content,” thereby preventing it from accessing higher levels of traffic. Consequently, they advocate for the submission of substantive comments and oppose the practice of posting numerous repetitive, insubstantial comments for the sake of boosting popularity.
Conflicting folk theories II: Can diversified keywords or thematic centrality better break through filter bubbles?
Although the academic community continues to debate whether personalized recommendation algorithms exacerbate filter bubbles (Mattis et al., 2024), with perspectives varying between those asserting that such algorithms intensify filter bubbles (Dylko et al., 2017; Törnberg, 2018) and those suggesting that personalized recommendations can diversify content and effectively mitigate algorithmic filter bubbles (Fletcher and Nielsen, 2017; Haim et al., 2018; Möller et al., 2018), the reality of information silos created by algorithms remains evident in netizens’ actual media consumption experiences. A case in point is the 2023 ban of the Douyin celebrity “Xiucai (秀才),” a popular influencer with over 10.54 million followers, dubbed the “harvester of middle-aged women” and particularly favored by women over 40, who was ultimately banned due to alleged tax violations. Shockingly, despite his popularity, many young, frequent social media netizens had never heard of him nor encountered his content on short video platforms until the ban made headlines.
During the process of popularity-boosting, netizens not only strive to increase dissemination volume but also seek to diversify the recipient base, aiming to spread content across different social circles. In addressing how to break through algorithmic filter bubbles, two conflicting folk theories have emerged. One theory suggests that diversifying the words used in comments can enrich the post's tags, thus promoting it to a broader audience. This theory is rooted in the matching mechanism between content and user tags within distribution algorithms (Zhao, 2022). Essentially, it implies that the more varied the content-related tags, the more likely they are to match with diverse user groups. Therefore, netizens attempt to enrich tags by commenting on various unrelated content. By leveraging tag-matching algorithms, they aim to distribute content to a wider user base, thereby breaking through algorithmic filter bubbles. For instance: Help top, thank you! Might be better: more diverse keywords (break the information cocoon, divert to other areas) Genshin Impact, Honkai Impact 3rd, Honkai: Star Rail, Onmyoji, Onmyoji Arena, Honkai School 2, Kings of Glory, Happy Popping, …, Xiaolongbao, Char Siu, Braised Pork, Duck Blood Vermicelli Soup, …, Jay Chou, Li Ronghao, Mayday, concert.
Therefore, proponents of this folk theory argue that “genuine and concentrated discussions will enhance the importance of the topic, enabling the post to penetrate various circles and break through the algorithmic filter bubble.” They thus encourage high-quality, meaningful, and in-depth comments centered on the post itself. Conversely, they believe that overly chaotic keyword comments may classify a post as low-quality content due to its lack of focus, thus reducing the likelihood of breaking through algorithmic filter bubbles. Consequently, they oppose meaningless keyword stuffing in the comment section, considering it as “internet water army” that disrupts the focus of the post, thereby reducing its circulation and even obscuring the event itself.
Conflicting algorithmic folk theories and the ambiguity of algorithms
The formation of conflicting folk theories about algorithms stems from platforms’ ambiguous descriptions of their mechanics, leading to disputed attributions among netizens about the relationship between algorithmic resistance’ actions and outcomes.
The study reviewing the dissemination rules published by social media platforms indicates that, although these platforms ostensibly disclose how their algorithms operate, they in fact provide only vague descriptions (Zhao, 2022). Specifically, these platforms do not clarify how their algorithms specifically enhance or diminish post visibility, nor do they elucidate how these metrics are adjusted in various contexts. This fundamental ambiguity allows for covert interventions by platforms and states, particularly in contentious online viral events. Such interventions are invisible “black boxes” to the public, who cannot see whether and to what extent they occur. This ambiguity also creates space for the public to develop differentiated, even conflicting, folk theories. Although seemingly contradictory, both sets of conflicting algorithmic folk theories are supported by anecdotal evidence of successful applications.
For the folk theory that prioritizes the quantity and diversity of keywords, this is validated by the case described at the beginning of the text, where a girl sought to download a video of her deceased grandmother. Beneath this post, nearly all the comments were popularity-boosting generic ones such as “up up up…” and “Boost popularity only!!! Boost popularity only!!!…”. These comments successfully pushed the post into a larger traffic pool, thus increasing its visibility and ultimately helping the girl contact the software service provider. Unlike what the folk theory focusing on quality and keyword centrality suggests, these conventional popularity-boosting comments did not classify the content as low quality, thereby not limiting its dissemination reach or breadth.
Conversely, the folk theory emphasizing post quality and centrality, which includes rewards for quality and centrality and penalties for quantity and dispersed keywords, has also been validated through some bloggers’ practices. For example, a video blogger noted that his casually posted videos usually perform well in terms of likes and views. However, videos about the “Garcinol Incident” and “Cancer Cases at Zhongshan Second Hospital” received at least a hundred times fewer views than his other videos. The blogger attributed this significant discrepancy to the latter videos’ comments focusing on boosting popularity through irrelevant discussions rather than substantial interactions. This case has persuaded some netizens, even converting those who previously believed in the folk theory of quantity and diverse keywords, to shift their understanding of algorithms and adjust their algorithmic resistance behaviors.
However, whether platforms and states penalize traffic involving posts with specific sensitive topic keywords during particular viral online events remains unclear due to the vague rules of platform algorithms, making it difficult for netizens to discern.
The disruption of netizens’ continuity and consistency in algorithmic resistance
The inundation of comments that merely aim to maintain post visibility, devoid of substantial content, can diminish some netizens’ engagement and sustained interest in pertinent social issues. The comment section, conceptualized as a dynamic textual environment around posts (Zhao, 2022), serves as a platform for civic engagement and social interaction (Stroud et al., 2015). It provides citizens with an opportunity to express views on social events while exposing them to diverse perspectives (Reich, 2011; Ziegele et al., 2018). Research indicates that user comments are a crucial component of the information valued in online news environments. In controversial social issues, some users might prioritize reading others’ comments to gauge diverse viewpoints before engaging with the main content (Zhao, 2022). Thus, the comment section is considered a vital arena for the exchange of opinions (Ziegele et al., 2018). However, when comments that solely enhance visibility begin to dominate these spaces, originally intended for public discourse (Toepfl and Litvinenko, 2018), they transform the space into a battleground against algorithm manipulation, disrupting meaningful public discussions.
Consequently, some netizens resist such algorithm-driven behaviors by boycotting them. For example, one expressed frustration: “I understand that everyone uses it to boost visibility out of good intentions, but it's just too long and meaningless! It overshadows many more meaningful comments. Every time I see a bunch of these, it really annoys me.” Similarly, another remarked, “It's really bothersome when genuinely heartfelt comments get buried. Seeing this affects my reading experience.”
Some netizens, due to the difficulty of finding meaningful information in the comment sections, even choose to completely withdraw from public discourse. For instance, one said, “Seeing these makes me not even want to scroll any further; I just exit.” Similarly, another commented, This so-called boosting popularity comments only leads to impatience in searching through other comments! Once I was determined and scrolled through dozens of such comments, which left me frustrated. Ever since then, whenever I see these kinds of comments, I simply exit and don’t look back.
Additionally, conflicting algorithmic folk theories incite internal strife among netizens, hindering further negotiations toward a unified folk theory. Netizens with different theories often engage in attacks on social media, leading to caution when expressing views on algorithmic folk theories. Netizens only feel secure in expressing their algorithmic folk theories and resistance strategies when they are certain that the majority of posts and comments align with their views; otherwise, they face the risk of confrontation. For instance, a netizen noted, “I expressed my opinion in those comment sections, and then I was attacked by others.”
Moreover, these conflicting algorithmic folk theories exacerbate disunity in netizens’ algorithmic resistance efforts, leading to power struggles within the group and weakening collective resistance. Netizens holding folk theories that prioritize quality and topic-centricity express strong aversion to posts focusing merely on quantity and keyword diversity. Consequently, upon encountering such posts, some of these individuals opt to disengage entirely, neither viewing nor contributing high-quality comments to enhance the posts’ visibility. In contrast, those advocating for diversity in keywords and post quantity to broaden reach adopt more inclusive resistance practices, accommodating various forms of algorithmic resistance. This disparity leads to a withdrawal of participation in algorithmic resistance among those prioritizing quality, as their motivation depends not only on the content but also on whether others meet their expectations. This withdrawal significantly reduces the number of participants in collective algorithmic resistance efforts.
Algorithmic resistance beyond algorithmic frameworks
Boosting dissemination by creating viral memes
In addition to leveraging the information distribution rules within the algorithmic framework to achieve widespread dissemination of specific posts, they extend beyond this framework by actively exploring new angles conducive to viral transmission. Research suggests that novelty is a crucial driver for the extended reach of information (Vosoughi et al., 2018), and content that deviates from the norm often garners more public interest (Shoemaker et al., 1991). In online discussions, netizens amplify the peculiar aspects of events, creating trending memes to attract internet celebrities and spark broad participation, thereby expanding the event's visibility. For instance, in the garcinol incident, researchers used this costly compound in swine feed to study its effects on pig growth. Netizens were shocked by the high costs involved, deduced from the published data, leading to garcinol being humorously proposed as a new unit of measure. Satirical comments such as defining a “pig day” as costing 432,000 RMB—the cost of garcinol for one pig per day—proliferated. Similar comments included, “My monthly living expenses are just 2.5 milligrams of garcinol, less than a pig's.”
After the creation of these viral memes, they indeed sparked continued engagement from netizens. Influencer accounts, driven by both a sense of public responsibility and the desire to capitalize on the surge of online traffic, consistently posted content featuring these trending memes. The broader netizen community, in turn, eagerly utilized these memes to express their frustrations and dissatisfaction with various aspects of life. For example, in the case of the Zhongshan Second Hospital incident, where the research team demonstrated a lack of concern for the health of graduate students, many doctoral students sarcastically referred to themselves as “laboratory consumables” to criticize the pressures of research and the various issues within the academic training system. Similarly, salespeople using the tactic of gifting pigs raised on garcinol to maintain client relationships served as a satire of the garcinol incident. Regardless of the form these expressions took, the ultimate aim was to maintain public attention on the relevant events, voice the public's demand for an investigation into the truth, and advocate for actions by the parties involved in pursuit of justice.
However, it is worth noting that this meme-creation strategy is not applicable to all trending events. This approach is more effective when the event contains elements of irony. In contrast, for events with predominantly tragic emotional tones, such as the “fire hydrant” incident—a real tragedy—this strategy is not suitable. Consequently, in such cases, netizens do not employ humor to drive viral dissemination.
Compared to other forms of algorithmic resistance, the primary advantage of this strategy is its avoidance of provoking netizen dissatisfaction during the observed period. Netizens are willing to engage in and support discussions on trending online issues within the context of satirical entertainment consumption.
Manually breaking the filter bubble by commenting on related and unrelated posts
Additionally, in terms of breaking through algorithmic filter bubbles, netizens do not act solely within the algorithmic framework. They also engage in resistance activities by manually circumventing these filter bubbles. This involves two main methods. The first method involves integrating recent viral events and posting about them in the comment sections of related videos or posts. This strategy aims to facilitate the dissemination of concurrent viral events by allowing netizens to discover multiple trending topics through a single comment section. For instance, in a post about a fire hydrant incident, netizens might not only see comments related to the fire hydrant but also integrated comments mentioning other viral events, covering all event keywords like “garcinol, Baby Come Home, Kunming Zoo, Sun Yat-sen Second Hospital, Guizhou forest fire, 731…” This approach ensures that once netizens are aware of one trending event, they can easily learn about other recent trending events through the comment section. The second method involves commenting on recent hot topics or event summaries under unrelated popular posts, thereby spreading content through the comments.
Manually circumventing algorithmic filter bubbles by integrating comments on trending online events and posting comments about trending events in unrelated popular posts has proven effective. In online ethnographic observations, many netizens reported discovering trending online events through such comments. For example, one mentioned that they did not encounter content related to the garcinol incident but found relevant keywords in the comments while browsing Baidu videos. They subsequently searched for the incident on Douyin. Another one discovered content about the Zhongshan Second Hospital incident in the comments section of news about the 731 incident and learned about the incident's details through searching.
Moreover, this practice holds significant positive implications for sustaining public discourse on related events. Under the algorithm-driven content distribution mechanisms of social media platforms, the lifecycle of a post is limited; when it lacks sustained interaction data, it ceases to be promoted. However, the practice of netizens manually circulating trending events in the comment sections of popular posts can, to some extent, mitigate the limitations imposed by the post's lifecycle on the ongoing dissemination of events. Through this manual intervention, netizens effectively counteract the restrictions of algorithms in the distribution and promotion of posts, thereby extending the lifespan of discussions surrounding the event.
Although such practices have their merits, the repetition of comments related to trending topics—including keywords and summaries of the events—across various posts on different platforms has also caused annoyance among some netizens. As one individual lamented, “I … also get annoyed seeing trend-chasing comments under completely unrelated videos … After seeing it replicated everywhere, I become so frustrated; it really triggers a contrarian response.” When netizens grow weary of pervasive, popularity-boosting comments to the point of losing interest in the related social issues themselves and thus abandon discussions about the event, these comments not only fail to achieve their initial purpose but also lead to a loss of support from a segment of netizens.
Conclusion
This study focuses on the collective algorithmic resistance actions of Chinese netizens in the comment sections of social media, employing the method of netnography and analyzing them from the theoretical perspectives of algorithmic resistance and folk theories of algorithms.
The study found that algorithmic resistance among a broader range of netizens primarily unfolds along two paths: Within and beyond the algorithmic framework. Within the framework, netizens develop both consensual and contentious algorithmic folk theories based on their practical experiences. While there is general agreement that increasing interaction volume can enhance dissemination, there is no consensus on which specific types of interaction effectively boost dissemination and break through algorithmic filter bubbles. Contentious folk theories lead to internal attacks and power struggles within netizen communities, hindering the negotiation process for a unified theory and further impeding the consistency and sustainability of algorithmic resistance actions, thereby weakening the collective strength of the resistance. In the path beyond the algorithmic framework, netizens break free from algorithmic constraints by creatively generating viral memes and manually disseminating information. This approach disrupts personalized recommendation algorithm filter bubbles, expanding the scope and prolonging the duration of event dissemination. While the creation of viral memes did not provoke dissatisfaction among netizens, the manual dissemination of trending events, however, generated frustration among some.
Therefore, whether within or beyond the algorithmic framework (excluding meme creation), unlike individuals, special groups, or TikTok mutual-aid users, there has been no consistent collective action among netizens, characterized by prominent internal conflicts and a loose form of algorithmic resistance. In actions intended to increase post dissemination through heightened interaction, some netizens disengage due to the ongoing recommendation of information by social media platform algorithms triggered by likes, comments, favorites, and shares. Others withdraw because folk theories emphasizing quality and thematic centrality fail to gain broader acceptance and support. Additionally, some netizens abandon attention and participation in online trending events due to comments that focus on data volume yet distract from meaningful public discussions, even calling for platforms to intervene and suppress similar algorithmic resistance actions. Lastly, some disengage due to annoyance at frequently encountering event keywords and summaries in comment sections.
Overall, the process of netizens’ algorithm resistance is fraught with multiple overlapping conflicts, lacking a robust mechanism for negotiation. Moreover, they tend to direct dissatisfaction experience toward internal disputes, overlooking external demands, such as collectively lobbying platforms to disclose more detailed algorithm rules or to provide specific support for significant public interest events to garner broader societal attention. The hybrid form of connective action proposed by Bennett and Segerberg (2012) offers useful insights in this context, where netizen groups could form public-interest organizations to aggregate dispersed individualized actions, thus creating larger-scale algorithmic resistance groups that integrate various forces. However, in practical terms, whether the collective algorithmic resistance of netizens can sustain over time (Couldry, 2014) or develop an internal negotiation system to ease internal conflicts and power struggles, ultimately forming a cohesive collective force, remains to be seen.
The theoretical contributions of this study are primarily threefold. First, in the realm of algorithmic resistance, this study enriches the conceptual understanding of algorithmic resistance by expanding the scope of research from within the algorithmic framework to beyond it. Previous studies have regarded algorithmic resistance as resistance through algorithmic tools (Bonini and Treré, 2024: 23), with a focus on resistance behaviors within the algorithmic framework, exploring how netizens exploit algorithmic rules to achieve specific objectives. However, this study reveals that netizens’ resistance behaviors are not always confined to the boundaries of algorithmic rules. Instead, they also employ more innovative and original methods to transcend these limitations and resist algorithms. The agency of netizens is increasingly highlighted in their interactions with algorithms. This finding provides a new perspective for enriching the concept of algorithmic resistance and offers a fresh approach for more comprehensive future investigations into the interactions between algorithms and humans.
Secondly, this study provides a detailed analysis of the complex power struggle dynamics within netizen communities during algorithmic resistance. This offers a valuable model for future research on netizen interactions across different groups and cultural contexts. Furthermore, it presents a preliminary conceptualization of analyzing these dynamics from the perspective of connective action. This provides an insightful framework for future investigations into netizens’ collective algorithmic resistance behaviors.
Finally, regarding algorithmic folk theories, these theories not only guide netizens’ algorithmic resistance behaviors but also shape the continuity and consistency of collective algorithmic resistance among netizens through their inherent conflicts. This provides insights into expanding the negotiation of algorithmic folk theories among netizens and examining the dynamic power struggles between platform algorithms, netizens, and within netizen groups. The ambiguity and lack of clarity in the disclosure of platform algorithmic rules provide ample space for the formation of diverse and conflicting algorithmic folk theories. However, previous studies have primarily focused on how netizens summarize folk theories through intuitive and experiential methods, as well as how certain consensus-driven folk theories guide their algorithmic behaviors. There is a lack of in-depth analysis on how conflicting folk theories within netizen communities lead to subsequent behavioral conflicts and power struggles. Future research could explore this aspect in greater depth through multi-scenario analyses.
Moreover, this study also holds significant practical implications for the algorithmic resistance of netizens, as well as for the formulation and adjustment of national policies. For netizens, initially faced with a fragmented algorithmic resistance, it is crucial during the resistance process to establish negotiation mechanisms. This involves negotiating both algorithmic folk theories and resistance strategies, aiming to build a consensus based on thorough communication, reconcile internal conflicts and power struggles, and thus enhance the cohesion and consistency of collectivity. Moreover, when facing internal conflicts over folk theories or the impact of popularity-boosting resistance actions on public discussions, netizens should not solely direct criticism inward but also recognize the role of platforms. It is vital to progressively form a negotiating force with platforms to shift from exploiting to rewriting algorithmic rules to better serve the public interest.
For policymakers, the state should pay attention to netizen algorithmic resistance behaviors and the related decline in public event engagement triggered by the platforms’ ambiguous algorithmic rules. It should further prompt platforms to establish negotiation mechanisms with netizens and provide a comprehensive and detailed explanation of their algorithmic rules, ensuring that netizens are informed of their rights regarding these rules, thereby maintaining harmony in social media interactions.
This study faces limitations, primarily focusing on netizen commentary behaviors and discussions, while insufficiently addressing platform attitudes and reactions, as well as the state's regulatory role. This limitation makes it challenging to comprehensively describe the visibility competition among netizens, platforms, and the state. The difficulty in accessing platform algorithm personnel, especially under the current controversy over personalized recommendation algorithms in Chinese society, contributes to this limitation. Interviews with algorithm personnel require strict approval procedures, and even if obtained, the richness of insights is questionable due to the tech companies’ scripted responses, with little else permitted for disclosure. Furthermore, for the state, apart from publicly released documents, there is a lack of additional referenceable material. Therefore, in the future, encouraging more open public discussions on algorithm issues by both platforms and the state could facilitate deeper research and debate on these topics.
Future studies could progress in several key areas. Firstly, incorporating a broader range of social media platforms would validate or refine this study's findings, thereby ensuring their broad applicability and reliability. Secondly, future studies could delve deeper into resistance strategies beyond the algorithmic framework and internal power struggles within algorithmic resistance, across different application scenarios and cultural contexts, to better understand the dynamics of power struggles between algorithms and netizens, and among netizens themselves. Additionally, the formation and impact of folk theories should be scrutinized more closely, focusing on subtle differences and conflicts within algorithmic folk theories to analyze power struggles among netizens holding different algorithmic folk theories. Lastly, the platforms’ flexible manipulation of algorithms and their ambiguous descriptions also require further exploration. Despite challenges in accessing platform data, this remains a critical area of research. Addressing these issues would significantly advance our understanding of the complex interactions among social media platforms, algorithms, and netizens.
Footnotes
Acknowledgements
ORCID iD
