Source-linked AI summary
Making Sense of Audience Change: Causal Attribution and Boundary Management on Xiaohongshu
Zhiyi Zhou, Kyrie Zhou
TL;DR
Recommendation-driven platforms can reshape who becomes visible and how community boundaries are negotiated, yet Xiaohongshu has no single platform-wide membership boundary. This study analyzes discussions of a perceived World Cup-related audience influx and finds a strong association between users’ attributions and their boundary-management strategies.
Problem
The study asks how users make sense of perceived community change on a recommendation-driven platform and what strategies they use to negotiate its boundaries.
Source: §1 · pp. 1–2Method
The authors analyzed 262 comments from eight purposively selected high-engagement threads using thematic analysis, structured coding, and permutation-based association analysis.
Source: §1 · p. 2Results
Among 70 comments with explicit attributions and strategies, attribution and strategy were strongly associated (Cramer’s V = .583): algorithmic attributions concentrated in technical management, while social-actor attributions aligned more with enforcement or socialization.
Source: §1 · p. 2Takeaways & Limitations
Audience acquisition can create governance consequences, while personalization can function as informal social-boundary infrastructure and gendered membership remains contested.
Source: §1 · p. 2, §5 · p. 17The purposively selected, high-engagement threads and 262 comments are not representative of Xiaohongshu users or platform-wide discussion.
Source: §5.6 · p. 22
Abstract
from arXiv · showhide
Recommendation-driven platforms can reshape not only what users see, but also who becomes visible and how community boundaries are negotiated. We examine Xiaohongshu discussions surrounding a perceived World Cup-related audience influx to understand how users attributed community change and how those explanations related to boundary-management strategies. We analyzed 262 comments from eight high-engagement threads using researcher-led thematic analysis, structured coding, and permutation-based association analysis. Users described personalization as both social separation and a boundary that could fail, interpreted platform growth through audience acquisition and commercialization, and contested whether gender should define community membership. Among 70 comments with explicit attributions and strategies, attribution and strategy were strongly associated (Cramer's V = .583): algorithmic attributions concentrated in technical boundary management, while social-actor attributions more often aligned with social enforcement or normative socialization. These findings connect causal sensemaking with boundary management and show how audience restructuring can create governance consequences on recommendation-driven platforms.
1 Introduction
The paper examines how Xiaohongshu users explained a perceived World Cup-related audience shift and negotiated belonging on a recommendation-driven platform without a formal platform-wide membership boundary. Its central finding is that perceived problem location aligned with articulated intervention site.
1 Introduction
- Users’ explanations for a perceived platform-associated audience shift and the strategies they articulated for negotiating boundaries remain understudied when membership criteria are informal and contested.Potential criteria included gender, conduct, interests, platform history, and everyday use.Source: §1 · p. 2
- The study combines thematic analysis, structured coding, and permutation-based association analysis of public discussions from eight high-engagement threads.The final analytic corpus contained 262 comments.Source: §1 · p. 2
- Among 70 comments with explicit attributions and strategies, attribution and strategy were strongly associated (p < .001, Cramér’s V = .583).Algorithmic attributions concentrated in technical boundary management, while social-actor attributions more often aligned with social enforcement or normative socialization.Source: §1 · p. 2
- Users described personalization as social separation that could fail, interpreted platform growth through audience acquisition and commercialization, and contested whether gender should define membership.Attribution was not significantly associated with evaluative stance (p = .182, Cramér’s V = .200).Source: §1 · p. 2
- The paper connects causal sensemaking with boundary management and frames audience acquisition as having governance consequences on recommendation-driven platforms.It also treats personalization as social boundary infrastructure and gendered membership as meaningful but contested.Source: §1 · p. 2
2 Related Work
Related work treats community boundaries as produced through norms, governance, audience management, and recommendation. The paper addresses how users publicly negotiate an informal platform-wide boundary during perceived audience change and whether causal explanations align with boundary-management strategies.
2.1 Community Norms and Gendered Boundary Negotiation
- Community boundaries are maintained through everyday norms and sociotechnical governance, while Xiaohongshu research documents recommendation-mediated audience control and contested membership criteria.The remaining gap concerns public negotiation of membership in a broader platform community when gender, conduct, interests, and platform culture compete as bases for belonging.Source: §2.1 · pp. 2–3
2.2 Audience Management and Algorithmic Sensemaking on Recommendation-Driven Platforms
- Audience management research links uncertain visibility and algorithmic sensemaking to individual and collective responses, while personalization can produce both social proximity and separation.The paper extends this work by asking whether causal framings are systematically associated with articulated boundary management.Source: §2.2 · pp. 3–4
2.3 Audience Change, Newcomer Influx, and Platform Governance
- Studies of influx, migration, and governance show that participation changes reshape norm reproduction, but less often examine public sensemaking of perceived platform-associated audience shifts.The paper focuses on how users distribute responsibility while negotiating an informal boundary whose meaning is contested.Source: §2.3 · pp. 4–5, Figure 1 · p. 5
3 Methods
The study uses an event-focused corpus of public Xiaohongshu discussions, combining researcher-led thematic analysis with separate structured operational coding and quantitative association analysis. The methods distinguish interpretive themes from coded variables and assess coding reproducibility and robustness.
3.1 Study Context and Data Collection
- The study examined public discussions around World Cup content as a focal case of perceived audience change on a recommendation-driven platform.Researchers purposively selected eight high-engagement posts; the source set was not treated as representative of Xiaohongshu activity as a whole.Source: §3.1 · pp. 5–6
- Researchers collected 836 comment records from eight source posts, removed within-thread exact duplicates to obtain 808 comments, and retained source identifiers.Collected counts refer to deduplicated public comments and replies; Table 1 overviews the source discussions and their contributions to the corpus.Source: §3.1 · p. 6, Table 1 · p. 6
3.2 Research Ethics and Data Protection
- The study analyzed public posts without user interaction, omitted profile-level identifiers, and translated and lightly paraphrased quotations to reduce identification risks.The verbatim dataset was not publicly released because aggregating and redistributing user-generated content can pose privacy risks.Source: §3.2 · p. 6
3.3 Corpus Construction and Relevance Screening
- Comment-level relevance screening reduced 297 candidate comments to a final corpus of 262, including 146 core relevant and 116 contextual comments.All subsequent analyses used this corpus.Source: §3.3 · p. 7
3.4 Researcher Led Thematic Analysis
- Thematic analysis iteratively clustered 28 initial codes into four broader areas and retained contradictory cases that constrained interpretations.The first author rechecked themes against the final corpus; all four remained represented across multiple threads.Source: §3.4 · pp. 7–8
3.5 Operational Coding
- A separate operational codebook coded stance, explicit causal attribution, and articulated boundary-management strategy as mutually exclusive variables.The codebook used conservative textual evidence, and “none” indicated no explicit attribution or strategy in the comment rather than absence outside the text.Source: §3.5 · p. 8
3.6 Reliability and Robustness Checks
- Raw agreement/κ were .983/.963 for relevance, .983/.971 for stance, .950/.937 for attribution, .967/.962 for fine-grained strategy, and .967/.957 for broad strategy.These statistics assess reproducibility of the operational categories, not the interpretive themes.Source: §3.6 · p. 8
- Held-out classifier Macro F1 was .606 for attribution, .660 for fine-grained strategy, and .719 for broader strategy families.Performance varied across labels, particularly infrequent norm-oriented strategies; classifier predictions served as supplementary validation and workflow support, not the sole basis for analysis.Source: §3.6 · pp. 8–9
4 Findings
The findings describe four overlapping patterns in how users made sense of a perceived World Cup-related audience influx. They concern recommendation and visibility, platform growth, participation norms and boundary management, and contested gendered belonging.
4 Findings
- Users described recommendation as keeping audiences in relatively separate content environments or making them unexpectedly visible to one another.The themes overlap within comments rather than defining mutually exclusive groups.Source: §4 · p. 9
- Users interpreted World Cup investment as part of a broader effort to expand Xiaohongshu’s audience, often connecting growth to male users, commercialization, and platform development.Source: §4 · p. 9
- Perceived participation changes brought expectations about appropriate conduct to the foreground and prompted users to articulate different boundary-management responses.Source: §4 · p. 9
- Users disagreed over whether Xiaohongshu was a gendered community and whether gender itself should determine belonging.Source: §4 · p. 9
4.1 Personalization as Separation and Its Breakdown
Users experienced personalized feeds as keeping social and interest circles apart, but also described unwanted exposure that persisted despite attempts to retrain or filter recommendations.
4.1.1 Personalized feeds kept different circles apart.
- Personalized feeds let different interests and social circles coexist with limited contact, so platform-wide audience growth did not necessarily change what users encountered.Users compared distinct homepages and described some feeds as largely free of the conflict others reported.Source: §4.1.1 · p. 10
4.1.2 Users tried to repair unwanted exposure when separation failed.
- Some users found personalization unreliable: World Cup or contentious content continued appearing despite repeated “not interested” feedback, prompting blocking or other filtering attempts.Users also described changing preferences, blocking terms or accounts, and searching for preferred topics to retrain feeds.Source: §4.1.1 · p. 10, §4.1.2 · p. 11
- Users disagreed about whether recommendation reliably maintained separation, describing both successful filtering and repeated failure rather than a uniformly effective or ineffective system.Filtering was an attempt to keep unwanted topics outside regular feed experience; frustration followed when that separation did not hold.Source: §4.1.1 · p. 10, §4.1.2 · p. 11
4.2 World Cup Investment and Audience Growth
Users interpreted the World Cup investment as a way to attract or retain audiences, especially male users, and debated the consequences of that perceived growth strategy.
4.2 World Cup Investment and Audience Growth
- Users often interpreted the World Cup investment as an effort to acquire or retain users, particularly men, whom they viewed as an audience the female-skewed platform had attracted less successfully.The theme appeared in 82 comments across seven source threads.Source: §4.2 · p. 11, §4.2.1 · p. 12
4.2.1 The World Cup was understood as an audience acquisition event.
- Commenters viewed the World Cup as an entry point for new users, while some expected event-driven participation to fade after the tournament.One commenter predicted at least half of the World Cup audience would leave afterward.Source: §4.2.1 · p. 12
4.2.2 Users connected audience growth with commercialization.
- Users linked audience expansion to commercialization, but differed over whether growth served the platform well or threatened its existing community.They connected male users with advertising, ecommerce, and consumption categories, while the paper notes these were user interpretations, not evidence of internal platform decisions.Source: §4.2.2 · p. 12
4.3 Community Norms and Boundary Management
Users debated acceptable participation and used a range of strategies to maintain a tolerable interaction environment, from accommodation to sanction and exit.
4.3 Community Norms and Boundary Management
- Perceived audience change made expectations about acceptable behavior and disagreement explicit, with users advocating civility as well as blocking, reporting, confrontation, exclusion, and withdrawal.The theme appeared in 76 comments across all eight source threads.Source: §4.3 · p. 13, §4.3.1 · p. 13
4.3.1 Users articulated conduct based expectations for participation.
- Some users treated respectful conduct, rather than gender, as the condition for participation and sought civil discussion without requiring newcomers to share existing users’ identities.Other commenters welcomed new participation, complicating a uniform account of existing users defending a stable community against outsiders.Source: §4.3.1 · p. 13
4.3.2 Boundary management involved several different repertoires.
- Boundary management involved coexisting strategies—not a documented escalation sequence—including technical filtering, norm education, blocking or reporting, confrontation, and leaving.Among 262 comments, 97 articulated a strategy; algorithmic separation appeared in 28 and content filtering in 24, while blocking or reporting appeared in 15.Source: §4.3.2 · pp. 13–14, §4.3.1 · p. 13
4.4 Gendered Membership Was Contested
Users contested whether gender defined Xiaohongshu membership, invoking the platform’s existing identity while also proposing conduct, interests, history, or utility as grounds for belonging.
4.4.1 Some users treated the existing gender composition as part of community identity.
- Users disagreed over whether the platform’s female-skewed audience was a meaningful community identity or a membership boundary worth protecting.Some viewed gender composition as ordinary platform differentiation; others challenged the idea that Xiaohongshu remained women-centered.Source: §4.4.1 · pp. 14–15
4.4.2 Users also defined belonging through behavior, interests, and everyday utility.
- Other users detached belonging from gender, emphasizing respectful behavior, shared interests, longstanding participation, and Xiaohongshu’s everyday usefulness.Some male users described established participation and practical value, while commenters also disputed whether football interest mapped neatly onto gender.Source: §4.4.2 · pp. 15–16
- Personalized feeds could make users experience the same platform as socially homogeneous, while the discussions showed that its audience and membership criteria remained contested.Figure 2 reports an association between attribution families and boundary-management strategies among 70 comments; the structural-platform category has only six observations and warrants caution.Source: §4.4.2 · p. 16, Figure 2 · p. 16
4.5 Attribution and Boundary Management Strategies
Among comments that explicitly named both a cause and a response, attribution and boundary-management strategy were strongly associated. Algorithmic explanations tended to pair with technical responses, while social-actor explanations more often paired with responses aimed at user interaction.
4.5 Attribution and Boundary Management Strategies
- The attribution–strategy association was statistically significant under a 50,000-permutation test (p < .001; Cramér’s V = .583) and remained similar in magnitude when any one of eight source discussions was excluded.Leave-one-source-out Cramér’s V values ranged from .561 to .626.Source: §4.5 · pp. 16–17
- 38 of 41 comments (92.7%) attributing the problem to recommendation articulated technical boundary management.No comments in this group used normative socialization as their primary strategy.Source: §4.5 · p. 17
- Among 23 social-actor attributions, 12 (52.2%) used social enforcement, six (26.1%) normative socialization, and two (8.7%) technical management.These comments more often located both the problem and response in interaction between users.Source: §4.5 · p. 17
- The six structural-platform comments with an explicit strategy were too few to support substantive interpretation of that category’s strategy distribution.The six comments were divided across technical management, social enforcement, normative socialization, and withdrawal.Source: §4.5 · p. 17
- Among 173 comments with an explicit attribution, the analysis did not detect a significant association between attribution and evaluative stance (p= .182, Cramér’s V= .200).The clearer pattern concerned users’ articulated responses to perceived causes, rather than whether they evaluated change positively or negatively.Source: §4.5 · p. 17
5 Discussion
The discussion connects perceived audience change with governance, recommendation-mediated boundaries, causal explanations, and contested membership. It interprets users’ accounts as evidence of how people understand and negotiate change, while recognizing that the observational, event-focused corpus limits broader claims.
5.1 Platform Growth as Community Governance
- Users interpreted the World Cup investment as audience acquisition, making platform growth a governance concern because perceived shifts in participation can renew negotiation over norms and membership.These comments reveal users’ interpretations, not Xiaohongshu’s internal goals or decision making.Source: §5.1 · pp. 17–18
5.2 Recommendation Systems as Social Boundary Infrastructure
- Personalization served as user-experienced social boundary infrastructure: it could separate audiences, but unwanted exposure could make that separation seem unreliable.Users described “not interested” feedback and feed retraining as boundary repair; the framing does not imply fixed groups or objective demographic separation.Source: §5.2 · pp. 18–19
5.3 Causal Sensemaking and User Boundary Management
- Attribution and strategy were strongly associated (Cramer’s V= .583), aligning perceived problem locations with intervention sites, but the observational data cannot establish causal direction.The corpus also showed no detected significant attribution–stance association; the more pronounced relationship concerned articulated responses.Source: §5.3 · pp. 19–20
5.4 Gendered Membership as Negotiated Community Identity
- Gender was prominent in accounts of platform identity, but users disagreed over whether it should define membership, with some emphasizing conduct, interests, or everyday use instead.The paper treats gendered platform identity as negotiated rather than inferable directly from audience composition.Source: §5.4 · pp. 20–21
5.5 Design Implications
- The design implications favor multiple intelligible intervention pathways and revisable norm communication, without inferring users’ causal models or automatically encoding demographic membership criteria.The authors frame these as design opportunities, not interventions validated by the study.Source: §5.5 · pp. 21–22
5.6 Limitations and Future Work
- The eight-thread, 262-comment corpus is event-focused and nonrepresentative; public comments do not establish users’ full beliefs or behavior, platform decisions, or recommendation mechanisms.The attribution–strategy association is observational, and the small number of source threads and cross-sectional data constrain inference about its development.Source: §5.6 · pp. 22–23
6 Conclusion
The study examines how users interpreted a perceived World Cup-related audience influx on Xiaohongshu and negotiated community boundaries. It finds a strong association between attributed causes and articulated strategies, alongside contested views of personalization and gendered membership.
6 Conclusion
- Across 262 comments, users described personalization as social separation that could fail, disputed gender as a membership rule, and strongly associated algorithmic attributions with technical boundary management.Social-actor attributions more often aligned with normative or enforcement responses.Source: §6 · p. 23