Hyperreal Differential Order in Douyin’s Tuwei Ecology: A Socio-Technical Systems Analysis of Algorithmic Visibility

Abstract

Recommendation systems reorganize culture by linking classification, visibility, and value. This article analyzes Douyin’s #tuwei ecology as a socio-technical system in which platform architecture, creator formatting, audience evaluation, and urban-rural symbolic hierarchies interact through recursive feedback. We formalize this system as four coupled operators—commensuration, sign packaging, metricized valuation, and reception feedback—and derive two observable indices: an item-level transfer-efficiency ratio and a class-level diffusion–appreciation asymmetry index. Using 123,300 public #tuwei videos collected between May and November 2024, multimodal coding of 500 videos, and sentiment analysis of 61,284 comments, we identify a stable diffusion-appreciation asymmetry. Practice-oriented rural videos receive relatively strong approval after encounter, but they diffuse less widely than spectacle-oriented performances optimized for rapid recognition, ironic forwarding, and interaction density. Corpus-level tests (${\chi }^2$, N = 123,300; Cramér’s $V\approx 0.32$) and an outlier-sensitivity analysis confirm the pattern independently of the coded-sample size. To explain this pattern, we propose hyperreal differential order (HDO), a systems concept describing how platform infrastructures repackage embedded cultural practices into feed-compatible sign forms and then differentially amplify them across segmented attention circuits. We position HDO against algorithmic visibility, platformization, and metric power, showing what each leaves underspecified. The study offers a mechanism-based account of how short-video systems reorganize vernacular culture, symbolic power, and interpretive authority without relying on overt exclusion.

1. Introduction

Recommendation-driven media environments turn cultural circulation into a systems problem. On short-video platforms, tags, feeds, and engagement metrics convert heterogeneous practices into comparable signals that can be ranked, recombined, and amplified [1,2,3,4,5,6]. What becomes visible is therefore not simply what exists or what audiences privately prefer. It is what can be made legible to platform systems and what can travel through them at speed.

This transformation is especially consequential when platform logics intersect with pre-existing social hierarchies. In China, the urban-rural divide has long structured symbolic value as well as economic opportunity. Rurality can be celebrated as cultural origin, yet it is also routinely marked as backward, rustic, or lacking refinement. The vernacular label “tuwei” condenses this ambivalence. It can signal earthy humor, grassroots creativity, and intimacy, but it can also function as a pejorative shorthand for tastelessness or low status [7]. On Douyin, where exposure is organized primarily through recommendation rather than follower subscription, this ambivalent sign has become a visible and analytically rich genre cluster [8,9].

To orient readers unfamiliar with the case, three features of the empirical setting are worth stating at the outset. First, Douyin is the largest short-video platform in China, with hundreds of millions of daily active users, and its distribution is governed primarily by an interest-based recommender rather than by social-graph following. Second, the #tuwei tag is not a marginal niche but a large, highly interactive classification zone that aggregates everything from staged “social shake” performances to slow rural-craft demonstrations under one label. Third, the social meaning of “tu” (earth/soil) carries deep historical sediment in Chinese culture, ranging from an agrarian philosophical root to a modern marker of urban–rural distinction. The empirical puzzle that motivates this study is that two very different renderings of rurality share the same tag yet circulate very unequally.

Existing platform scholarship shows that recommendation systems are valuation infrastructures rather than neutral conduits [10,11,12]. Critical studies of algorithms have also emphasized the limits of transparency-centered approaches and the importance of tracing system effects through observable outputs rather than through inaccessible source code alone [13,14,15]. Meanwhile, work on digital rurality and short-video culture demonstrates that platformized representation can reshape what counts as authentic, entertaining, or socially valuable in peripheral settings [9,16]. Two gaps remain. First, recommendation is often treated as background infrastructure rather than as an active organizing mechanism that couples classification, circulation, and reception. Second, representational studies often describe rural images without specifying how some semiotic forms become systemically more diffusable than others.

A systems perspective is useful because it directs attention to components, linkages, outputs, and feedback rather than to single variables in isolation. The issue is not whether Douyin explicitly suppresses one kind of rural content. The issue is how a socio-technical system selectively rewards particular combinations of signs, temporalities, and affects [5]. This framing also moves platform research closer to systems analysis by asking how local interactions accumulate into stable ordering effects.

Against this background, the objectives of this study are fourfold. (i) To describe, at scale, the distribution of appreciation and diffusion across competing renderings of rurality within a single platform classification field. (ii) To develop and formally specify a mid-range mechanism—hyperreal differential order—that explains how culturally embedded practices are reformatted into platform-compatible sign surfaces and then differentially amplified. (iii) To distinguish this mechanism conceptually from neighboring constructs (algorithmic visibility, platformization, and metric power). (iv) To identify system leverage points for governance that do not depend on access to proprietary ranking parameters.

These objectives are pursued through three research questions. RQ1: Within the #tuwei field, which semiotic renderings of rurality are more likely to diffuse widely? RQ2: Which affective and formal configurations are more compatible with engagement-driven metrics? RQ3: How do recurrent visibility asymmetries reorganize interpretive authority over rural representation?

The paper makes three contributions. Empirically, it provides large-scale evidence of patterned visibility asymmetry inside a major short-video platform. Theoretically, it introduces hyperreal differential order (HDO) to explain how culturally embedded practices are reformatted into platform-compatible sign surfaces and then differentially amplified. Methodologically, it shows how circulation effects can be reconstructed without direct access to proprietary ranking parameters. Together, these contributions align with Systems’ interest in complex social systems in which people, processes, and technologies interact recursively [1,17,18].

The remainder of the paper is organized as follows. Section 2 reviews the relevant literature and states the research gap. Section 3 develops the theoretical framework, defines the #tuwei ecology as a socio-technical system, formalizes the HDO mechanism, and positions it against neighboring concepts. Section 4 describes the data source, sampling, variable operationalization, and scope of inference. Section 5 reports the results, including robustness analyses. Section 6 discusses the findings in relation to the literature and derives governance leverage points. Section 7 concludes with limitations and directions for future research.

2. Literature Review

This section reviews three bodies of work that the study connects—platforms as valuation infrastructures and the production of algorithmic visibility; the cultural-theoretic vocabulary of vernacular culture, symbolic power, and interpretive authority; and digital rurality and the urban–rural symbolic order in China—before stating the research gap.

2.1. Platforms as Valuation Infrastructures and Algorithmic Visibility

A first strand treats recommendation systems not as neutral pipes but as infrastructures that classify, rank, and value content [1,2,3,4]. Platforms perform editorial functions through automated curation, deciding what is surfaced to whom and thereby acting as gatekeepers of visibility [1]. Bucher’s influential account frames this as the “threat of invisibility”: creators orient their practice toward being seen by the system, and ranking becomes a form of power that operates by distributing exposure rather than by overt prohibition [6,7]. Adjacent work on the platformization of cultural production shows how platform infrastructures, markets, and governance penetrate creative industries and reshape the cultural commodity itself [10].

Recent empirical research sharpens this picture for short-video systems. Studies of Chinese platforms describe how creators are cultivated in stages through traffic rewards, algorithmic visibility, and advertiser-facing incentives, producing a standardized production line that trades autonomy for reach [16]. A systematic review of content creation under algorithmic environments identifies recurring dynamics—market rationality underpinning visibility, folk theorization of the algorithm, normative control of creators, and the partial subversion of these arrangements—and warns that algorithms fed by an unequal field can replicate and amplify offline inequalities [19]. Work on everyday creator practice further shows that visibility is actively negotiated rather than passively received, with users developing tactics of resistance and adaptation to platform power [20,21]. Audits of engagement-based ranking, finally, demonstrate that optimizing for revealed engagement can amplify emotionally charged and divisive content beyond what users say they value on reflection [22]. Together, this literature establishes that visibility is produced, uneven, and consequential—but it tends to treat visibility as an item- or creator-level outcome rather than as the differential routing of comparable units within a single cultural field.

2.2. Vernacular Culture, Symbolic Power, and Interpretive Authority

A second strand supplies the cultural-theoretic vocabulary that the analysis reorganizes. By vernacular culture we mean the everyday, grassroots, often non-institutional expressive practices of ordinary people—dialect, craft, folk humor, and situated knowledge—whose authority derives from lived participation rather than from credentialed institutions [23]. Bakhtin’s account of the carnivalesque is useful here because it foregrounds how popular, bodily, and grotesque registers both invert and reproduce hierarchies; #tuwei spectacle draws much of its energy from precisely this carnival grammar [23].

By symbolic power we mean, following Bourdieu, the capacity to impose legitimate categories of perception so that an unequal classification of taste comes to appear natural [24]. Symbolic power works less by exclusion than by making distinctions—refined versus vulgar, modern versus backward—seem self-evident. In platform settings this classificatory work is partly delegated to infrastructures: hashtags, adjacency, and ranking become operative categories that distribute legitimacy [11,18]. Couldry and Mejias’s notion of data colonialism extends this to the extraction of social life as data, including the affective and semantic labor of cultural producers [25].

By interpretive authority we mean the socially distributed right to define what a cultural sign means and whose rendering counts as authentic. Fricker’s analysis of epistemic injustice clarifies the stakes: groups can be wronged specifically in their capacity as knowers and meaning-makers when their testimony is systematically discounted or when the shared interpretive resources needed to render their experience intelligible are unequally available [26]. Reframed for platforms, the question is not only who is seen, but whose definition of “real” rurality is allowed to scale. These three concepts—vernacular culture, symbolic power, and interpretive authority—are precisely the terms that, we argue, short-video systems reorganize; Section 3 and Section 6 specify how.

2.3. Digital Rurality and the Urban–Rural Symbolic Order in China

A third strand concerns rurality in Chinese digital culture. Fei Xiaotong’s classic account of Chinese society as a “differential order” (chaxu geju)—a rippling, ego-centered web of relations anchored in land and kinship—remains the touchstone for understanding the urban–rural symbolic environment [7]. The post-reform period layered onto this a sharp urban–rural distinction in which the rural became, simultaneously, a site of nostalgic authenticity and a marker of backwardness. This ambivalence is exactly what the “tu” sign carries into the platform era.

Empirical platform research documents how this ambivalence is reworked online. Studies of Douyin’s “platformization of rurality” show how platform architecture, power, and marginal creativity jointly shape what rural creators can make visible [9], while accounts of TikTok’s infrastructuralization in China trace how the app’s technical layers transform the conditions of cultural production [8]. Research on rural creators and rural women influencers cautions that visibility is not the same as empowerment: rural imagery is often emotionally resonant yet idealized, and exposure can subject creators to moral evaluation and performance pressures rather than granting genuine agency [16]. Chinese-language scholarship has analyzed #tuwei specifically as a site of subcultural resistance and aesthetic stigma and has examined the cultural politics of algorithmic society for short-video content production. What this work establishes is that the urban–rural divide supplies an already unequal symbolic environment into which platform operations are inserted. What it less often specifies is the mechanism by which some rural renderings, under shared classification, become systematically more diffusable than others.

2.4. Research Gap

Across these strands, three limitations recur. First, recommendation is frequently treated as background infrastructure rather than as an active organizing mechanism that couples classification, circulation, and reception within a single field. Second, representational studies richly describe rural images but rarely specify, and almost never formalize, why some semiotic forms travel further than others under identical tagging. Third, the cultural vocabulary of symbolic power and interpretive authority is seldom connected to measurable circulation outcomes. The present study addresses this gap by (a) treating the #tuwei tag as a bounded socio-technical system, (b) formalizing a coupling mechanism—hyperreal differential order—that links commensuration, packaging, valuation, and reception, and (c) testing its observable implications against large-scale platform traces while remaining explicit about the limits of inference. This positions the study to explain visibility asymmetry as an emergent system property rather than as the expression of a single hidden rule.

3. Theoretical Framework: Hyperreal Differential Order

This section develops the framework in four steps. Section 3.1 argues for treating the #tuwei ecology as a socio-technical system and justifies its four constituent components. Section 3.2 integrates two cultural theories—hyperreality and the differential order—into the concept of hyperreal differential order (HDO). Section 3.3 positions HDO against neighboring concepts. Section 3.4 gives a formal specification and derives the testable propositions.

3.1. The #Tuwei Ecology as a Socio-Technical System

We define Douyin’s #tuwei ecology as a bounded socio-technical system composed of four interacting components. The choice of these four—and the claim that they constitute a system rather than a list—rests on the criterion that distinguishes a system from an aggregate: the components are mutually constituting and linked by feedback such that the behavior of the whole cannot be read off any one part [17]. The four components are: (1) platform architecture (interfaces, the interest-based recommender, hashtag commensuration, and metric currencies such as likes, comments, and shares); (2) creator formatting strategies (the packaging of situated practice into short, repeatable, feed-compatible units); (3) audience evaluation practices (viewing, liking, commenting, sharing, and the affective repertoires expressed in comments) [27]; and (4) historically sedimented symbolic hierarchies (the urban–rural distinction that supplies an already unequal evaluative environment) [7,24,28].

These components form a system because they are coupled by recursive feedback rather than chained in a one-way sequence. Architecture conditions what formatting is rewarded; formatting shapes what audiences encounter; audience evaluation generates the metric signals that the recommender ingests; and the resulting visibility re-enters creator expectations and platform optimization, while the urban–rural hierarchy biases each step by supplying the categories through which “real” and “vulgar” rurality are perceived [12,14]. Removing any component dissolves the phenomenon: without commensuration there is no shared field; without metricized evaluation there is no selective amplification; without sedimented hierarchy there is no culturally loaded direction to the asymmetry. The system is therefore the appropriate unit of analysis, and the diffusion–appreciation asymmetry we observe is its emergent output [17,25].

3.2. Hyperreality and the Differential Order

To theorize the recursive process, we integrate two traditions. From Baudrillard we take hyperreality: the operational detachment through which embedded practices circulate as stylized surfaces whose value increasingly derives from circulation rather than from situated reference [29,30]. On the platform, the “golden seconds” logic of feed capture compresses an agrarian narrative into an instantly recognizable sign; the practice becomes a simulacrum optimized for the scroll. From Fei Xiaotong we take the differential order: the patterned, ego-centered inequality of relational distance that structures who is close, legible, and valued [7]. We argue that the platform performs a hyperreal cutting of the differential order: it detaches rural signs from their situated reference and simultaneously re-sorts them into a metricized hierarchy of circulation, so that some sign forms repeatedly enter high-velocity pathways while others remain in bounded circuits of appreciation [24].

Hyperreal differential order (HDO) names this configuration. It can be stated compactly: HDO is the system property by which a recommendation infrastructure (i) repackages embedded cultural practice into feed-compatible sign surfaces (the hyperreal moment) and (ii) differentially amplifies those surfaces across segmented attention circuits according to a culturally loaded ordering (the differential-order moment), with (iii) reception feedback stabilizing the result. HDO is thus a mechanism concept rather than a metaphor: it specifies how classification, ranking, and reception translate local interactions into stable visibility asymmetry [12,14,25,31].

3.3. Conceptual Positioning: HDO, Algorithmic Visibility, Platformization, and Metric Power

HDO must be distinguished from several established constructs, and the distinction is one of scope and unit of analysis rather than of mere relabeling. Algorithmic visibility (Bucher) centers on whether content and creators are made (in)visible by ranking; platformization (Nieborg and Poell) centers on how platform infrastructures and markets penetrate cultural production; metric power (Beer) centers on how metrics measure, govern, and discipline. Each is indispensable, and HDO draws on all three. What each leaves underspecified is the within-field, semiotically differentiated routing of comparable units coupled to a status hierarchy and stabilized by reception. HDO’s contribution is to integrate semiotic transformation (hyperreality), relational ordering (differential order), and infrastructural feedback into a single testable mechanism operating inside one classification field.

Table 1. Conceptual positioning of hyperreal differential order (HDO) relative to neighboring constructs.

Construct	Core Focus	What It Leaves Underspecified	HDO’s Added Value
Algorithmic visibility [6,22]	Whether/how content and creators are made (in)visible by ranking; the threat of invisibility	How visibility differences couple with semiotic transformation and cultural hierarchy; the appreciation-vs-diffusion split	Specifies the semiotic and relational mechanism, not only exposure
Platformization [10]	Penetration of platform infrastructures and markets into cultural production	The differential routing of comparable units within one field; reception feedback	Operates inside a single classificatory field and models differential amplification
Metric power [3]	How metrics measure, govern, and discipline conduct	How metric valuation becomes culturally patterned diffusion asymmetry across a status order	Links metric valuation to hyperreal detachment and differential order
HDO (this paper)	How commensuration + packaging + valuation + reception jointly reformat embedded practice and differentially amplify it	—	Integrates semiotic transformation, relational ordering, and feedback into one testable mechanism

summarizes the comparison.

3.4. Formal Specification and Propositions

Consistent with a systems-analytic approach, we specify HDO formally before deriving its observable implications. Let the #tuwei field be a set of videos indexed by i. For each video we observe cumulative likes $L_i$, comments $C_i$, and shares $S_i$ at capture, and assign a semiotic-regime indicator $g_i$, where $g_i=1$ denotes grotesque-carnivalesque spectacle and $g_i=0$ denotes rural-practice authenticity. We treat likes as a low-friction signal of appreciation and shares as a higher-velocity signal of outward transfer into new attention circuits.

We model the system as four coupled operators applied to a situated practice $x_i$:

$$\begin{array}{cc}\hfill \mathrm{commensuration}:& {\tilde{x}}_i=\kappa \left(x_i\right)\hfill \\ \hfill \mathrm{packaging}:& {\pi }_i=\pi ({\tilde{x}}_i,g_i)\in [0, 1]\hfill \\ \hfill \mathrm{valuation}:& w_i=exp({\beta }_{\pi }{\pi }_i+{\beta }_{\alpha }{\alpha }_i)\hfill \\ \hfill \mathrm{reception}\mathrm{feedback}:& {\pi }_i^{(t+1)}=\phi \left({\pi }_i^{\left(t\right)},{\alpha }_i^{\left(t\right)}\right)\hfill \end{array}$$

Here $\kappa$ maps heterogeneous practices into a common comparability space via the hashtag, so that all #tuwei videos become comparable content units; ${\pi }_i$ is a packaging-compatibility score capturing how compressed, punchline-oriented, and feed-legible a sign package is (operationalized through the four multimodal dimensions in Section 4); ${\alpha }_i$ is reception ambivalence, the share of ambivalent/ironic responses relative to stable positive appreciation; and $w_i$ is the visibility weight that converts packaging and reception into realized diffusion. The parameters ${\beta }_{\pi }$ and ${\beta }_{\alpha }$ are assumed positive: the system rewards feed-compatible packaging and interaction-dense, debate-generating reception.

Two observable quantities follow. At the item level, define the transfer-efficiency ratio

$${\rho }_i=S_i/L_i,$$

the outward propagation generated per unit of low-friction approval. Because $S_i$ is increasing in $w_i$, the model implies $\mathsf{\partial }\rho /\mathsf{\partial }\pi >0$ and $\mathsf{\partial }\rho /\mathsf{\partial }\alpha >0$. At the class level, define the diffusion–appreciation asymmetry index for a content class k as

$${\mathrm{\Phi }}_k={\sigma }_k^L/{\sigma }_k^S,$$

where ${\sigma }_k^L$ and ${\sigma }_k^S$ are class k’s shares of total likes and total shares in the field. ${\mathrm{\Phi }}_k>1$ indicates an appreciation-rich but diffusion-poor (contained) class; ${\mathrm{\Phi }}_k<1$ indicates a diffusion-rich (amplified) class. HDO predicts $\mathrm{\Phi }>1$ for rural-practice content and $\mathrm{\Phi }<1$ for spectacle. Finally, because $\phi$ carries positive feedback (${\beta }_{\pi },{\beta }_{\alpha }>0$), the system has a reinforcing fixed point at high $\pi$: repeated coupling drives the field toward spectacle as its dominant attractor even when practice content is positively evaluated after encounter.As shown in Figure 1, the above operators $\kappa$, $\pi$, w, and $\phi$ are visualized via two coupled feedback loops.

This formalization yields three propositions, which Section 5 tests:

P1. Spectacle-compatible semiotic packages should convert attention into diffusion more efficiently than process-oriented packages grounded in cumulative labor and situated narration.

P2. Ambivalent or ironic reception should be more diffusion-productive than stable positive appreciation because it generates debate, forwarding, and reflexive positioning.

P3. Repeated coupling among classification, ranking, and reception should stabilize spectacle as the dominant public face of #tuwei, even when practice-oriented videos are positively evaluated after encounter.

4. Materials and Methods

4.1. Data Source: Newrank Collection Mechanism, Coverage, and Biases

The empirical entry point is the Douyin hashtag #tuwei, treated as a platform-native classification device. Data were obtained through Newrank, a licensed commercial analytics service that aggregates publicly visible Douyin metadata [32]. Because the validity of the analysis depends on the properties of this source, we describe its collection mechanism, coverage, update frequency, and known biases in turn.

Collection mechanism. Newrank is a third-party data-service provider that indexes publicly available short-video metadata across major Chinese platforms, including Douyin. For a given hashtag it returns the set of public videos carrying that tag together with structured metadata—upload time, caption text, the full hashtag set, and cumulative public engagement counts (likes, comments, collections, shares). Newrank functions as an observational interface onto public outputs; it does not expose impression counts, recommendation weights, or any internal ranking signal. Following digital methods and critical data studies, we therefore treat these records as mediated traces produced through layered infrastructures of capture, formatting, and access rather than as direct windows onto platform ground truth [14,33,34].

Coverage. Coverage is restricted to content that was public and undeleted at the time of indexing. Private accounts, removed or shadow-restricted posts, and content that never received the tag are out of scope; exposure (how often a video was served) is not observable at all. The source therefore captures the population of publicly visible #tuwei outputs, which is the appropriate population for a study of public circulation, but it cannot speak to suppressed or never-surfaced content.

Update frequency and snapshot nature. Engagement counts in the service are refreshed periodically and were retrieved as cumulative snapshots at the time of export rather than as continuous trajectories. We treat all counts as values at capture, not as final lifetime values, and our distributional analyses are robust to this because they compare regimes measured under the same snapshot regime.

Potential biases and mitigation. Four biases follow from the above: (i) a visibility/survivorship bias toward public, non-removed content; (ii) a snapshot bias because counts are cumulative at capture; (iii) a platform-mediation bias because Newrank’s own indexing and sampling rules are proprietary and not fully disclosed; and (iv) a self-labeling bias because creators choose whether to apply the tag. We do not claim these are eliminated. To bound their effect on distributional conclusions, we manually verified a random subsample of 200 videos within 48 h of export: public accessibility and hashtag consistency exceeded 95%, and engagement discrepancies between the service and the live app remained within low single-digit percentages. These checks support the dataset’s adequacy for system-level distributional analysis while leaving exposure-level claims explicitly out of scope.

After this verification, the initial export of 131,300 records associated with #tuwei between 1 May and 30 November 2024 (China Standard Time) was cleaned by removing duplicates, incomplete entries, and explicit advertising content, yielding a final analytic corpus of 123,300 videos. For each video we retained upload time, caption text, hashtag set, and cumulative counts of likes, comments, and shares at capture.

4.2. System Boundary and Scope of Inference

The system boundary is deliberately modest. We analyze public content, platform-visible metrics, and public comments. We do not claim access to proprietary ranking parameters, impression logs, or recommendation weights. Consequently, we are explicit that this design supports system-analytic and associational claims rather than internal-causal ones; in particular, it cannot establish that the recommender alone “caused” the observed differences. We cannot, from public traces alone, fully separate user-preference effects, network effects, and ranking effects.

Two features of the design nonetheless make a pure-preference explanation insufficient on its own. First, the key outcome is the diffusion–appreciation asymmetry: practice content that is in fact approved after encounter (high likes) still diffuses comparatively little. A simple “audiences just prefer spectacle” account does not explain why approved content fails to convert approval into transfer. Second, the asymmetry is observed at the level of comparable units under one tag, holding classification constant. These features locate the explanandum at the system interface—the coupling of packaging, valuation, and reception—rather than in any single internal rule. We therefore phrase findings throughout as patterned outcomes consistent with the HDO mechanism, not as proof of an internal algorithmic cause.

4.3. Field Mapping, Sampling, and Multimodal Coding

Field-level aggregation alone cannot identify the semiotic mechanisms through which rurality becomes differentially diffusable, so we combined corpus mapping with close multimodal coding.

First, we constructed a thematic subset using the keyword phrase “xiangcun shenghuo” (“rural life”), yielding 2578 videos. This subset serves as a conservative proxy for explicitly self-labeled rural-practice content and is used for comparative contextualization rather than as an exhaustive definition of rural authenticity.

Second, we drew a stratified random sample of 500 videos from the full corpus, stratified across upload month (May–November 2024) and likes-based quintiles within each month, ensuring coverage of both highly visible and ordinary posts. Under this ecologically faithful design, the two regimes were unequally represented (44 rural-practice versus 456 spectacle-oriented videos). We emphasize that this imbalance is itself a substantive finding: within the #tuwei tag, spectacle is simply far more prevalent than practice. The stratified sample was designed to capture system regularities, not to estimate balanced category parameters.

Third, two features of the design address the concern that 44 coded practice videos are too few for stable inference. The coded practice set serves a qualitative purpose for which it is adequate—identifying the ideal-typical semiotic regimes and characterizing their packaging and reception—rather than estimating population parameters. Quantitative claims about differential circulation do not rest on this cell: they are anchored in the full corpus ($N=\mathrm{123,300}$) and in the larger, explicitly practice-oriented “rural life” subset ($n=2578$), while the regime contrasts within the coded sample are evaluated with non-parametric tests and effect sizes appropriate for unequal group sizes. As reported in Section 5, the direction and significance of every key contrast are independently corroborated by corpus-level tests that do not depend on the coded subsample.

The coding framework drew on multimodal discourse analysis [35]. We coded four dimensions—visual grammar, auditory semiotics, embodied practice, and temporal regime—which jointly operationalize the packaging-compatibility score ${\pi }_i$ in the formal model. Through pilot coding and adjudication, two ideal-typical semiotic regimes were identified: grotesque-carnivalesque spectacle and rural-practice authenticity. These are analytical constructs, not normative judgments. Three trained coders independently coded the sample; mean pairwise Cohen’s kappa before adjudication was 0.81, indicating substantial agreement.

1.

Grotesque-carnivalesque spectacle: intensified filters, exaggerated embodiment, compressed temporality, punchline-oriented framing, and rapid feed legibility (high $\pi$).

2.

Rural-practice authenticity: coherent rural settings, process-oriented labor, contextual narration, and cumulative temporality tied to material transformation (low $\pi$).

4.4. Operationalization and Measurement

Table 2. Operationalization of the formal HDO variables.

Variable	Construct	Operationalization	Source/Measure
$L_i,C_i,S_i$	Appreciation, interaction, transfer	Cumulative likes, comments, shares at capture	Newrank metadata
${\pi }_i$	Packaging compatibility	Mean of standardized codes on visual grammar, auditory semiotics, embodied practice, temporal regime; higher = more feed-compatible	Multimodal coding (mean Cohen’s $\kappa$ = 0.81)
${\alpha }_i$	Reception ambivalence	Share of ambivalent/ironic comments relative to stable positive comments	3-class sentiment model + qualitative repertoire analysis
${\rho }_i=S_i/L_i$	Transfer efficiency	Shares per unit of likes; item-level diffusion of approval	Derived
${\mathrm{\Phi }}_k=\sigma L_k/\sigma S_k$	Diffusion–appreciation asymmetry	Class share of likes ÷ class share of shares; >1 = contained, <1 = amplified	Derived (corpus)

maps each formal variable introduced in Section 3.4 to its operational measure and data source. The engagement counts ($L_i$, $C_i$, $S_i$) are taken directly from the platform metadata. Packaging compatibility (${\pi }_i$) is computed as the mean of the standardized multimodal codes across the four coded dimensions, so that higher values denote more feed-compatible packages. Reception ambivalence (${\alpha }_i$) is the share of ambivalent or ironic comments relative to stable positive comments. The two indices that the empirical analysis reports—item-level transfer efficiency (${\rho }_i$) and the class-level diffusion–appreciation asymmetry (${\mathrm{\Phi }}_k$)—follow directly from these measures. The remainder of this subsection specifies the circulation and reception measures used to estimate them.

Given the heavy-tailed distribution of platform engagement, we prioritized distributional comparison over mean-based inference. At the corpus level we calculated total engagement, per-video averages, and the shares-to-likes ratio $\rho$, interpreting $\rho$ as a practical proxy for outward propagation beyond immediate approval: likes are a low-friction signal of appreciation, whereas shares require higher-velocity transfer into other attention circuits. At the coded-sample level we compared the two regimes using non-parametric tests appropriate for skewed data (Mann–Whitney for continuous engagement measures, with medians and effect sizes; chi-square for categorical sentiment comparisons, emphasizing effect magnitude alongside significance).

To analyze reception, we collected comments from the coded videos between 18 and 20 December 2024, combining up to 25 hot comments with up to 100 chronological comments per video; after cleaning, the reception corpus comprised 61,284 comments. A manually annotated dataset of 5000 comments trained a three-class sentiment model (positive, negative, ambivalent/ironic); an overlap subset of 800 comments achieved Krippendorff’s alpha = 0.84. The fine-tuned Chinese RoBERTa model reached 92.3% accuracy and 0.91 macro-F1 on the held-out test set. Because irony is pragmatically unstable, we built an irony/Internet-slang lexicon and used model outputs to guide qualitative repertoire analysis rather than to replace it [36,37].

4.5. Ethical Considerations

All analyzed materials were publicly accessible at the time of collection. No private accounts, private messages, or non-public data were accessed. Following AoIR guidance, we report aggregate patterns and paraphrase illustrative comments to reduce traceability [38]. The study therefore addresses system effects at the level of public circulation rather than individual diagnosis or personalized inference.

5. Results

5.1. Differential Circulation Across the #Tuwei System

At the field level, the #tuwei corpus is expansive and highly interactive. Across the 123,300 cleaned videos, total engagement reached approximately 173 million likes, 17.33 million comments, and 87.29 million shares. The size of the field indicates that #tuwei is a stable classification zone inside Douyin’s recommendation environment rather than a marginal fringe.

Once the explicitly practice-oriented “rural life” subset is isolated, a differentiated pattern appears. Although this subset accounts for only 2.1% of all videos, it captures 3.8% of total likes but only 1.5% of total shares. Applying the asymmetry index, ${\mathrm{\Phi }}_{\mathrm{practice}}\approx 3.8/1.5\approx 2.5>1$: the class is appreciation-rich but diffusion-poor, exactly the contained position HDO predicts.

Table 3. Aggregate circulation metrics for the #tuwei ecology (May–November 2024).

Dataset	Video Count	Total Likes	Total Comments	Total Shares
Full #tuwei corpus	123,300	≈173,000,000	≈17,330,000	≈87,290,000
“Rural life” subset	2578	6,520,000	429,500	1,303,000

summarizes the divergence. Table 3 summarizes this divergence.

Per-video averages clarify the asymmetry further. Rural-practice videos generate more likes per video and slightly more comments per video than the corpus baseline, but they generate fewer shares per video and a far lower shares-to-likes ratio (

Table 4. Per-video engagement and diffusion indicators.

Dataset	Likes/Video	Comments/Video	Shares/Video	Shares/Likes
Full #tuwei corpus	≈1403	≈141	≈708	≈0.50
“Rural life” subset	≈2529	≈167	≈505	≈0.20

). In other words, viewers who encounter these posts often respond positively, yet that approval is less likely to be converted into outward movement across the system.

This divergence is not reducible to a handful of viral cases. Median and interquartile comparisons point in the same direction, and the pattern persists when the coded sample is stratified by engagement quintile. Practice-oriented videos cluster more often in mid-visibility tiers, whereas spectacle-oriented videos remain overrepresented in higher diffusion tiers. The result is consistent with P1: practice-centered rurality attracts approval after encounter, but spectacle-compatible packages convert that approval into wider transfer more efficiently.

Corpus-Level Confirmation and Robustness

To confirm that the asymmetry does not depend on the coded-sample size, we conducted two corpus-level checks. First, a chi-square comparison of interaction volumes across the two regimes over the full corpus is significant for likes, comments, and shares (${\chi }^2$ well above critical values; all $p<0.001$), with a Cramér’s V of approximately 0.32 for likes—a medium effect by Cohen’s convention—indicating that the urban–rural divergence in circulation is a systematic, not random, feature of the field. Second, because 21 spectacle videos exceeded one million likes (maximum $\approx 2.43$ million), we re-ran the comparison after removing all posts above one million likes (remaining $N\approx \mathrm{123,279}$). Statistical significance ($p<0.001$) and effect size (Cramér’s $V\approx 0.31$) were essentially unchanged, showing that the asymmetry is not an artifact of a few outliers. These corpus-level results address the small-cell concern directly: the central pattern holds on the full population independently of any subsample.

The coded-sample contrast points in the same direction: rural-practice videos show a substantially lower shares-to-likes ratio than spectacle videos, with the Mann–Whitney comparison significant and a small-to-medium effect size, which is the appropriate statistic to report given the unequal group sizes. The convergence of the full-corpus tests, the outlier-sensitivity check, and the coded-sample contrast provides triangulated support for P1 that does not rest on the 44-video cell alone.

5.2. Semiotic Regimes and Format Compatibility

The circulation asymmetry is tied to a semiotic asymmetry. In the ecologically faithful stratified sample, grotesque-carnivalesque spectacle accounts for the large majority of videos and rural-practice authenticity for a small minority; because the sample was stratified to capture visibility tiers rather than to estimate population shares, this ratio indexes everyday recognizability within the tag rather than a population parameter. The two regimes differ systematically in how they package rural signs (

Table 5. Semiotic stratification under the #tuwei label.

Modal Dimension	Grotesque-Carnivalesque Spectacle	Rural-Practice Authenticity
Visual grammar	Intensified filters, artificial lighting, exaggerated bodily performance, and decontextualized scenes.	Natural light, coherent rural locales, and scenes anchored in material environment.
Auditory semiotics	Loud and repetitive electronic motifs; dialect as punctuating exclamation.	Ambient soundscapes or folk motifs; dialect as narration and knowledge transmission.
Embodied practice	Repetitive gestures simulating activity without material transformation.	Sequential and goal-oriented labor or craft practice with visible outcomes.
Temporal regime	Accelerated and loop-friendly time aligned with rapid attention capture.	Cumulative time aligned with process, seasonality, and demonstration.

).

As shown in Figure 2, spectacle-oriented content has a higher shares-to-likes ratio $\rho$ than rural-practice content.

These differences matter because recommendation systems reward not only topics but also temporal and affective formatting. Spectacle-oriented videos make meaning available almost immediately; their signs are compressed, repeatable, and optimized for the first seconds of playback (high $\pi$). Practice-centered videos unfold more slowly, depending on sequential demonstration and attention to transformation rather than punchline (low $\pi$). The divergence is therefore not purely semantic; it is temporal and infrastructural. Under shared hashtag commensuration, the system treats these heterogeneous forms as comparable content units, yet comparability does not imply equal diffusion potential: spectacle is more compatible with scroll, replay, and ironic forwarding. The empirical implication is that rurality becomes recognizable as a format spectrum rather than as a situated social world—a tool shown without sustained labor, a kitchen shown without extended preparation, a dialect phrase used as punchline. The system does not erase rurality; it reformats it.

5.3. Reception Feedback and Affective Productivity

Reception analysis shows that diffusion advantage is reinforced by different affective repertoires. Practice-oriented videos are predominantly positive in the comment corpus (approximately 82% positive, 13% ambivalent/ironic, and 5% negative). Spectacle-oriented videos generate a markedly different pattern: approximately 58% ambivalent/ironic, 30% negative, and only 12% positive. A chi-square test confirms that the distributions differ significantly (p < 0.001).

The contrast is not best described as approval versus rejection. Practice-oriented videos are praised as “real,” “calming,” or “healing,” with comments emphasizing skill, patience, and everyday dignity. Spectacle-oriented videos generate moralized fascination: viewers call them vulgar, excessive, or embarrassing, yet still replay, debate, and forward them as jokes. This combination of stigma and attraction is consistent with carnivalesque and affective accounts of contentious attention [23,36,37]. From a systems perspective, ambivalence can be more diffusion-productive than stable admiration: it generates longer comment threads, more reflexive positioning, and more opportunities for recirculation. This pattern supports P2—the system rewards affect that travels, not merely affect that approves—and converges with audit evidence that engagement-based ranking amplifies emotionally charged, divisive content beyond stated user preference [22].

This is the key feedback pathway in the HDO model (the operator $\phi$). Reception does not merely respond to system outputs; it helps stabilize them. Once spectacle is repeatedly recognized as the most shareable form of #tuwei, creators and audiences treat it as the default grammar of the tag; practice-oriented forms remain visible, but more often as an admired niche than as the field’s dominant public face. Creator adaptation to perceived algorithmic preference—documented in folk-theory and creator-agency research [19,20,21]—supplies the imitation step that closes the reinforcing loop.

5.4. Mechanism Synthesis

Taken together, the evidence supports all three propositions at the level of patterned system effects rather than internal algorithmic proof. P1 is supported by the higher transfer efficiency ($\rho$) of spectacle-compatible packages, corroborated by corpus-level tests and the coded-sample contrast. P2 is supported by the greater recirculation productivity of ambivalent reception (high $\alpha$). P3 is supported by the way those two processes, once coupled through $\phi$, stabilize spectacle as the public face of the tag. In mechanism terms: hashtag classification ($\kappa$) makes heterogeneous rural signs commensurable; ranking (w) advantages packages that trigger fast recognition and dense interaction; reception ($\alpha$) feeds back into expectations about what works; and creator adaptation reproduces the same public grammar. The resulting system does not silence rural practice. It differentially amplifies it, reproducing symbolic inequality through ordinary platform operations rather than overt suppression.

6. Discussion

6.1. HDO as a Systems Explanation

The main theoretical value of HDO is that it explains visibility asymmetry as an emergent property of a socio-technical system rather than as the expression of a single hidden rule. The concept links semiotic transformation, relational ordering, and infrastructural feedback: hyperreality identifies how embedded practices are turned into circulating surfaces [29,30]; differential order identifies how those surfaces are unevenly distributed across attention circuits [7,24]; and the formal model in Section 3.4 specifies how classification, ranking, and reception translate local interactions into stable asymmetry [12,14]. This coupling explains why the results cannot be reduced to audience preference in a simple sense: viewers do not merely choose among fixed cultural objects but encounter content already filtered by classification, sequencing, and behavioral inference. What looks like organic popularity may be a recursive product of infrastructural affordances and culturally sedimented judgments, and the diffusion–appreciation split is one indicator of that recursion.

Reading the findings against the literature sharpens the contribution. Our results are consistent with, and extend, three recent lines of work. First, they parallel audit evidence that engagement-optimized ranking amplifies divisive, emotionally charged content beyond reflective preference [22]; we show the cultural analogue inside one tag, where ambivalent reception is more diffusion-productive than stable appreciation. Second, they corroborate accounts of staged creator cultivation on Chinese short-video platforms, in which traffic rewards and algorithmic visibility standardize production at the expense of diversity [16], by demonstrating the semiotic consequence of that standardization for rural representation. Third, they specify a mechanism behind systematic-review findings that algorithms fed an unequal field replicate offline inequality [19] and behind everyday creator negotiations of platform power [20,21,39]. Where these literatures largely treat visibility as item- or creator-level, HDO supplies the missing within-field, semiotically differentiated routing mechanism.

6.2. Vernacular Culture, Symbolic Power, and Interpretive Authority Reconfigured

Returning to the three concepts introduced in Section 2.2, the analysis specifies how short-video systems reorganize each. Vernacular culture is reorganized as format: grassroots rural expression is admitted to scale chiefly in its carnivalesque, feed-compatible register, while its process-oriented, knowledge-bearing register is contained in low-velocity circuits. Symbolic power is reorganized as routing: in Bourdieusian terms, domination works by making unequal classifications appear natural [24], and on platforms this naturalization occurs through routine coupling among tags, metrics, and feedback—the system never declares one rurality more legitimate; it simply routes one further, until routinized circulation acquires the appearance of common sense. Interpretive authority is reorganized as adjacency and scale: rural creators may supply the raw materials of vernacular culture, yet the renderings that travel are those most compatible with urban irony and algorithmic circulation, so the right to define “real” rurality migrates toward whoever commands feed-legible packaging. This is an epistemic injustice in Fricker’s sense as much as a visibility problem [26], and it is the analytic answer to RQ3.

6.3. Platform Governance and System Leverage Points

These findings contribute to platform-governance debates. The usual policy language of opacity, bias, or moderation is necessary but insufficient [40,41,42]. The #tuwei case suggests that governance must also address jurisdiction over recognizability: who defines categories, who benefits from adjacency, and which interaction forms are disproportionately rewarded. In short-video systems, symbolic ordering is itself a governance function. A systems perspective highlights leverage points that do not require full access to a proprietary ranking model; they target the high-gain junctions where ordering is produced—classification, diffusion weighting, and auditability [13,14,15,42,43]. Recent proposals to design recommenders for societal good, and regulatory instruments such as the EU Digital Services Act’s audit and transparency requirements, indicate that such junction-level intervention is increasingly feasible [43].

Table 6. Potential leverage points for reducing differential visibility in short-video systems.

Leverage Point	System Location	Expected Effect on Visibility Ordering
Classification governance	Hashtags and adjacent recommendations	Reduces automatic conflation of rural practice with derisive spectacle and increases semantic specificity at the entry point of circulation.
Diffusion weighting	Ranking rules and reshare incentives	Lowers the dominance of irony-rich recirculation by giving greater weight to sustained attention and constructive appreciation.
Auditability and access	Platform metrics and research interfaces	Makes it possible to distinguish exposure from evaluation and to test whether appreciation-rich content is systematically under-diffused.

summarizes these leverage points and their likely effects.

Three implications follow. First, category governance matters: hashtags and recommendation adjacencies are operative infrastructures of meaning, not merely descriptive metadata. Second, interpretive authority is unevenly distributed, an epistemic as well as a visibility issue [26]. Third, the system extracts more than behavioral data; it also extracts semantic and affective labor by turning situated cultural practice into monetizable circulation [25,44,45]. China-specific research on algorithm anxiety, infrastructural platformization, and the marginal creativity of rural creators is sharpened by our mechanism account: the problem is not simply that rural users participate under unequal conditions, but that the system rewards some rural sign packages as high-velocity spectacle while containing others as low-velocity authenticity [8,9,35].

6.4. Boundary Conditions and Transferability

Several limits define the scope of the claims. First, the analysis relies on public traces and a third-party dataset and therefore cannot recover impression-level recommendation pathways; engagement counts are snapshots, not full trajectories. Second, third-party aggregation mediates observability, and the source’s internal coverage rules are proprietary. Third, without impression logs or internal model parameters, user-preference, network, and ranking effects cannot be fully disentangled, so claims are system-analytic rather than causal in the narrow experimental sense (Section 4.2). Fourth, the study is bounded to one tag, one platform, and one time window; we do not claim that all rural representation on Douyin follows the same pattern.

At the same time, the analytical logic is portable. HDO should be strongest where four conditions coincide: open classification, short-form compression, interaction-weighted ranking, and culturally loaded peripheral subjects. Any platform system combining these features may produce similar asymmetries, so the framework should travel as a middle-range mechanism to other vernacular tags, other marginalized cultural fields, and other recommendation-based systems.

7. Conclusions

This study argued that Douyin’s #tuwei ecology is best understood as a socio-technical system in which cultural signs, platform interfaces, and audience feedback co-produce unequal visibility. Practice-centered rural videos are neither absent nor uniformly devalued; they attract meaningful appreciation, yet they diffuse less widely than spectacle-oriented forms that better match the platform’s temporal, affective, and classificatory logic. Hyperreal differential order names this configuration and, in its formal specification, makes it measurable: an item-level transfer-efficiency ratio and a class-level diffusion–appreciation asymmetry index together capture how recommendation systems repackage embedded practice into circulating sign surfaces and then distribute those surfaces unevenly. The resulting hierarchy is reproduced not through explicit prohibition but through ordinary operations of commensuration, packaging, valuation, and feedback.

For systems research, the broader implication is that algorithmic governance should be analyzed not only as a problem of model design or bias detection, but as a problem of recursive symbolic ordering. Platforms govern culture by structuring what can become recognizable at scale; the struggle over “tuwei” is therefore also a struggle over interpretive jurisdiction—whose meanings travel, under what formatting constraints, and with what social consequences. The value of HDO lies in making that ordering analytically tractable and normatively contestable.

8. Limitations

As detailed in Section 4.2 and Section 6.4, the principal limitations are the reliance on public, snapshot traces from a third-party source whose internal coverage is proprietary; the absence of impression logs, which precludes narrow causal attribution and the full separation of preference, network, and ranking effects; and the single-tag, single-platform, single-window scope. The semiotic regimes are analytical ideal types, and the small number of coded practice videos reflects their genuine scarcity within the tag; quantitative inference is therefore anchored in the full corpus and the larger practice-oriented subset rather than in the coded cell.

9. Future Research

Three directions follow. First, longitudinal capture of engagement trajectories—rather than snapshots—would allow estimation of the feedback operator $\phi$ over time and a direct test of the predicted reinforcing fixed point. Second, combining creator interviews and a balanced multi-platform design (for example, comparing Douyin with platforms whose ranking weights sustained attention differently) would test where the HDO mechanism holds, where it weakens, and which leverage points most effectively redistribute visibility. Third, partnerships that provide privacy-preserving access to exposure data, or independent audits under transparency regimes such as the DSA, would let researchers distinguish exposure from evaluation and convert the present associational findings into stronger causal tests. Extending HDO to other culturally loaded tags (for example, “guochao” or regional-craft fields) would further probe the boundary conditions of the mechanism.