How News Media Frame Regulatory Conflict in Emerging Industries: Evidence from Structural Topic Modeling of South Korean Press Coverage, 2018–2022

Abstract

Regulatory reforms in emerging industries require renegotiating the distribution of burdens between incumbents and new entrants, and the news media constitute the principal arena through which the public encounters and evaluates such conflicts. This study examines how regulation-related conflicts are represented in South Korean news media, applying structural topic modeling (STM) to 21,266 articles collected from BigKinds for 2018–2022, the formative period of Korea’s regulatory sandbox and the peak of its mobility-platform disputes. The analysis derives 30 topics and shows that coverage is dominated by announcements of regulatory innovation agendas and new-industry promotion, while concrete processes of conflict resolution are nearly invisible. Resolution-relevant topics declined over the observation period and appeared mainly in regional newspapers and IT–science sections rather than in mainstream national outlets. We interpret this pattern as an invisible middle of regulatory conflict: the media report the outbreak and formal conclusion of disputes, but not the negotiation through which regulatory bargains are reworked. This informational gap deprives governments of the informed public support needed to legitimize negotiated regulatory change, and we draw implications for how regulatory authorities should communicate reform as conflict mediation rather than industrial promotion.

1. Introduction

Regulation inevitably creates winners and losers. Whatever public purpose a regulatory policy serves, it restricts some actors’ behavior while protecting or advantaging others. This distributive character makes conflict an intrinsic feature of regulatory politics (Lowi, 1964; Stigler, 1971; Wilson, 1980). The classical public interest theory of regulation—the view that regulation exists to correct market failures and protect citizens from social harms (Baldwin et al., 2011)—has long obscured this conflictual dimension: if regulation serves the public good, the costs borne by regulated parties appear to be a justified and largely invisible sacrifice. Yet in contemporary societies where interest groups are densely interconnected, the legitimacy of a regulatory intervention can no longer be taken for granted simply because it claims to prevent harm. How regulatory conflict is surfaced, contested, and resolved has therefore become a central concern of regulatory governance research (Koop & Lodge, 2017; Levi-Faur, 2013).

This tension is particularly acute when new technology-based industries emerge because innovation routinely outpaces existing regulatory frameworks, forcing a redistribution of rents between incumbents and entrants (Ranchordas, 2015; Thelen, 2018). South Korea offers a revealing case. Since the late 2010s, the Korean government has pursued an ambitious regulatory reform agenda, including the introduction of a regulatory sandbox in 2019, to accommodate AI-driven and platform-based industries. Nevertheless, conflicts between incumbents and new entrants have repeatedly resulted in stalemates or de facto prohibitions. Uber’s ride-sharing service was banned after fierce opposition from the taxi industry; the domestic van-hailing service Tada was shut down by legislative amendment despite a court acquittal; and short-term rentals via Airbnb remain largely illegal for urban residential properties. In each case, what failed was not the recognition that regulatory change was needed, but the process for resolving the underlying conflict of interests among stakeholders.

Ways in which such conflicts are represented in the news media matter for their resolution. The mass media remain the primary channel through which citizens encounter regulatory issues, and media coverage shapes both the public salience of an issue and the perceived legitimacy of competing claims (Entman, 1993; McCombs & Shaw, 1972). When media coverage highlights conflict but renders the negotiation and resolution process invisible, the public is left without the information needed to evaluate or support compromise. To be sure, regulatory scholars have not ignored the media altogether: reputation-based accounts treat the media as a key audience shaping regulatory behavior (Carpenter, 2010; Maor, 2014), and the social amplification of risk framework highlights how media coverage magnifies public concern over regulated hazards (Kasperson et al., 1988). Yet this work centers on regulators and risks, not on how conflicts over regulation are framed for the public; the mainstream of regulatory scholarship remains focused on regulatory design, compliance, and institutional reform (Ayres & Braithwaite, 1992; Black, 2002; Majone, 1994). Likewise, while computational text analysis has been used to quantify the volume and complexity of regulatory texts themselves (Al-Ubaydli & McLaughlin, 2017), systematic, large-scale analysis of news discourse on regulatory conflict remains rare.

This study addresses this gap by analyzing how regulation-related conflict—particularly conflict surrounding emerging industries—is represented in South Korean news media. Drawing on 21,226 news articles collected from BigKinds, Korea’s national news archive, for the period 2018–2022 (the formative years of Korea’s regulatory sandbox and the peak of the mobility-platform disputes), we apply Structural Topic Modeling (STM) to ask three questions. RQ1: What thematic frames dominate media coverage at the intersection of regulation and conflict? RQ2: How does the prevalence of these frames vary over time, across media outlets, and across news sections? RQ3: To what extent is information about conflict resolution—as opposed to the mere existence of conflict—visible in this coverage?

The study makes two contributions. Substantively, it shows that while regulatory reform efforts and industry-related conflicts are heavily covered, concrete conflict-resolution processes are nearly invisible, have declined over time, and are relegated to regional outlets and IT sections—an informational gap that we argue constrains the government’s capacity to legitimize negotiated regulatory change. Methodologically, it demonstrates how covariate-aware topic modeling can extend regulatory governance research beyond case studies toward systematic analysis of public regulatory discourse. The remainder of the paper proceeds as follows: Section 2 develops the theoretical framework linking regulation and social conflict; Section 3 describes the data and methods; Section 4 presents the results; Section 5 discusses implications and limitations; and Section 6 concludes with limitations and suggestions for future research.

2. Theoretical Framework

2.1. Regulation and the Asymmetric Distribution of Burdens

Regulation, in its most basic sense, is a policy instrument through which government intentionally restricts or steers the behavior of citizens and enterprises (Koop & Lodge, 2017; Thompson & Jones, 1982). What makes regulation inherently conflictual, however, is not its definition but the structure of its consequences: the actors whose conduct is restricted and the actors who benefit from that restriction are rarely the same. Safety regulation burdens producers to protect consumers; environmental regulation burdens emitters to protect residents; licensing regulation burdens potential entrants to protect the stability—or the rents—of incumbents. Regulated parties bear concentrated compliance costs, while the benefits of regulation are typically diffused across society (Wilson, 1980). This asymmetry has predictable political consequences. As the logic of collective action suggests, small groups bearing concentrated costs mobilize far more effectively than large groups enjoying diffuse benefits (Olson, 1965), so regulatory decisions are persistently contested by those who must carry their burden. Lowi’s (1964, 1972) policy typology captures the same insight from the opposite direction: because regulatory policy applies coercion directly and immediately to individual conduct, it generates higher levels of social conflict than distributive policy, where burdens are barely visible. Conflict, in short, is not a side effect of poorly designed regulation; it is built into the distributive structure of regulation itself.

2.2. Regulation and Innovative Entrants

This distributive structure becomes particularly contentious when new, innovation-driven entrants appear in a regulated market. Existing regulatory frameworks are written around the incumbent industry—its technology, its business model, and its risk profile—and incumbents have organized their operations and sunk their investments under those rules. For entrants, the same rules operate as de facto entry barriers, whether or not they were intended as such (Djankov et al., 2002; Stigler, 1971). A new service that does not fit existing licensing categories is often simply illegal by default, regardless of its social value. The Korean ride-hailing service Tada is a case in point: operating vans with drivers through a rental-car exemption, it attracted over a million users but was faced with taxi-industry protests and criminal prosecution, and was ultimately shut down by a 2020 amendment that closed the exemption—before any renegotiation of the underlying licensing framework had taken place. Accommodating it, therefore, requires rewriting the rules under which incumbents built their businesses, and incumbents experience this rewriting not as neutral modernization but as a state-sanctioned redistribution of rents. This is why conflicts over platform services such as ride-hailing and short-term rentals have unfolded as intense distributive struggles in country after country (Thelen, 2018; Uzunca et al., 2018): the conflict is structured as incumbents defending the existing regulatory bargain against entrants demanding a new one, with the government caught in between.

2.3. Regulatory Lag and the Hollowing-Out of Regulation

The temporal dimension aggravates this conflict. Regulatory lag—the structural gap between the pace of technological change and the pace of regulatory adaptation—is a well-documented phenomenon, often discussed as the “pacing problem” (Marchant et al., 2011; Ranchordas, 2015). Yet it is insufficient to understand regulatory lag merely as technology moving too fast or regulatory reform moving too slowly. The more fundamental problem is that existing regulation rests on premises that are no longer valid. A regulation designed around a particular level of technology, a particular industrial structure, and a particular market order can fail to achieve its own purpose the moment those premises collapse—a condition Brownsword and Goodwin (2012) term “regulatory disconnection.” What follows is not simply that regulation falls behind, but that regulation persists in form while being hollowed out in substance: the rule remains on the books, yet it no longer protects the public interest it was meant to serve.

The consequences of this hollowing-out are concrete. A regulation designed to remedy consumer harm may fail to reach platform-based services, or rules established to ensure fair competition may apply only to the incumbent industry, while new entrants compete outside their scope without comparable constraints, resulting in a structure of reverse discrimination. Short-term rental platforms illustrate the point: Korean lodging rules, premised on conventional accommodation businesses, permit urban home-sharing only for foreign guests, leaving services such as Airbnb in a gray zone where identical transactions are legal or illegal depending on the guest’s nationality—a rule whose original premises no longer match the market it governs.

For regulated parties, it becomes unclear which rules apply to them, creating legal uncertainty; for regulators, a double dilemma emerges: strict enforcement of existing provisions blocks innovation, while lax enforcement erodes the consistency of the rule of law (Cortez, 2014). A regulation that has lost its design premises thus risks becoming a norm that imposes high costs without delivering protection, hindering innovation, and failing to safeguard citizens. Regulatory innovation, properly understood, is therefore not a matter of speed but of redesigning the underlying logic and operating structure of regulation itself, and instruments such as regulatory sandboxes represent institutional attempts to achieve this redesign (Zetzsche et al., 2017). Crucially, however, redesign of this kind redistributes burdens and benefits anew, which means it cannot proceed without confronting—and resolving—the conflicts of interest it provokes.

2.4. Why Media Representation Matters

If regulatory conflict is structural and its resolution requires renegotiating the regulatory bargain among incumbents, entrants, and the state, then how this conflict is presented to the public becomes consequential. The news media remain the principal arena through which citizens encounter regulatory issues, and media coverage shapes both the salience of those issues and the perceived legitimacy of competing claims (Entman, 1993; McCombs & Shaw, 1972). When coverage reports the existence of conflict but renders the negotiation and resolution process invisible, a representational gap emerges that we call the invisible middle of regulatory conflict. The concept is distinct from, though it builds on, two established ideas. Agenda-setting research shows that media coverage determines which issues gain public salience (McCombs & Shaw, 1972), and framing research shows that episodic presentation crowds out thematic context (Entman, 1993; Iyengar, 1991). The invisible middle names a third, stage-specific pattern: within a single conflict’s life cycle, the outbreak of confrontation and its formal conclusion—an agreement signed and a service banned—are visible, while the intermediate phase in which stakeholders actually negotiate the redistribution of burdens remains unreported. Where the middle is invisible, public support for compromise has little informational ground to stand on.

Seen through the lens of interactive marketing in new media, this gap is ultimately a failure of interaction. Research on interactive marketing emphasizes that communication creates value when it operates as dialogue rather than one-way transmission (Peltier et al., 2024; Wang, 2024), and studies of dialogical communication show that organizations build legitimacy on social media precisely by visibly engaging competing voices rather than broadcasting their own message (Herrada-Lores et al., 2025). Government communication of regulatory reform that consists of promotional announcements—innovation agendas and support budgets—follows the one-way model; rendering the negotiation process visible would constitute its dialogical counterpart. This perspective informs both our empirical expectations and the policy implications developed in Section 5.3.

Building on this framework, we examine how the connection between regulation and social conflict—specifically, conflict surrounding newly emerging industries—is represented in South Korean news media, asking which themes dominate coverage, how their salience changes over time, and where in the media landscape they appear. Answering these questions requires identifying thematic patterns across a corpus far too large for manual reading, while relating those patterns to document characteristics such as publication year, outlet, and news section. Structural topic modeling is suited to precisely this task: it recovers latent themes at scale and estimates how their prevalence varies with such covariates, thereby operationalizing our concern with the differential visibility of conflict phases as measurable differences in topic prevalence. The following section describes the data and model in detail.

3. Materials and Methods

The data used in the analysis here were collected from search results on BigKinds (www.bigkinds.or.kr), the Korea Press Foundation’s systematically organized media database. The URLs contained in the search results were used to directly collect the full text of each article. Specifically, multiple keyword combinations were used in the BigKinds search to maximize the diversity of the final dataset. News articles were collected by searching for keyword pairs closely related to ‘regulation–conflict’; all collected data were then combined, and duplicate documents were removed before analysis.

Four keyword pairs were ultimately used: (1) regulation–conflict, (2) regulation–collision, (3) new industry–conflict, and (4) new industry–collision. We also tested other terms with meanings similar to “conflict” or “collision”—including legal challenges, legal disputes, and stakeholder conflict—as well as terms denoting specific emerging services that had come to symbolize regulatory conflict over new industries in Korea, most notably Tada and Airbnb. Tada, a ride-hailing service, was in many respects the catalyst for public debate on new-industry regulation, while Airbnb represented the emerging platform economy more broadly; both were among the most prominent new-industry services of the period. Even for these emblematic cases, however, the searches retrieved very few documents, most of them very short and therefore excluded from the analysis. This outcome is itself informative: because even the terms most closely identified with new-industry regulatory conflict yielded too little text for reliable estimation, the broader “regulation”/“new industry” combinations were necessary to assemble a corpus of sufficient size.

The decision to anchor the search on conflict-denoting terms was likewise deliberate. Broader policy terms such as governance, policy reform, innovation policy, or platform regulation, if used without a conflict anchor, would have retrieved large volumes of general policy coverage and shifted the corpus away from conflict toward policy news at large. Pairing the domain terms with “conflict” and “collision” instead keeps the corpus centered on the phenomenon of interest—regulatory conflict in emerging industries.

After the texts were collected, tokenization was performed using the Bareun engine (https://bareun.ai), a Korean morphological analyzer also used for the Korea Press Foundation’s BigKinds service. Since 2019, the engine has been developed on diverse datasets from the National Institute of the Korean Language and the BigKinds service. Following tokenization, standard stop words—particles, endings, and bare numbers—were removed, and one-character tokens were also discarded, as they provided little useful information for the analysis.

The method used in this study is Structural Topic Modeling (STM), which extends the Latent Dirichlet Allocation (LDA) approach commonly used in topic modeling by incorporating the Correlated Topic Model (CTM)—which allows correlations among the derived topics—and the Dirichlet-Multinomial Regression (DMR) model, which allows the characteristics of the documents in the dataset to inform topic derivation. Originally developed by political scientists Roberts et al. (2013), STM makes it possible to visualize how patterns of topic prevalence differ across document characteristics, providing considerable assistance in understanding the context in which each topic arises (Mostafa et al., 2025; Roberts et al., 2016, 2020; Rodriguez & Storer, 2020). In our analysis, publication year, the media outlet, and the news category were included as document-level prevalence covariates to identify how the prevalence of each topic varied across these characteristics. The analysis was done in R statistical language, version 4.3.1 and stm package version 1.3.6.

The mathematical representation of this model is as follows. Equation (1) shows that, when estimating the probability that each document (d) contains each topic (K), the Dirichlet-Multinomial Regression method is applied to incorporate document characteristics ($X_d$) into the model while simultaneously accounting for relationships among topics (Σ). Equation (2) relates to the content of each document: the probability that a word belonging to a specific topic is distributed within each document (${\beta }_{d,k}$) is assumed to be decomposed into the baseline distribution of words within the document (m), the deviation attributable to a specific topic (${\kappa }_K$), the deviation attributable to document characteristics (${\kappa }_{X_d}$), and the deviation arising from the interaction between a specific topic and document characteristics (${\kappa }_{X_d,K}$). This model also allows one to verify the effect that document characteristics have on the topics contained in each document. Finally, Equation (3) represents the distribution of each word within a given topic, enabling the model to capture how topic characteristics manifest in each document through word distributions (Roberts et al., 2013, 2014, 2016).

$${\theta }_d|X_{d\gamma },\mathsf{\Sigma }~LogisticNormal(X_{d\gamma },\mathsf{\Sigma })$$

(1)

$${\beta }_{d,K}\propto \exp (m+k_K+k_{X_d}+k_{X_{d}K})$$

(2)

$$z_{d,n}|{\theta }_d~Multinomial({\theta }_d)$$

(3)

This model builds on the strength of conventional LDA models—which clearly define the distribution of latent topics as a set of manifest words—by enabling analysis of how variables associated with each document influence the distribution of topics within a regression framework. This makes it possible to analyze how the context in which documents were produced affects the distribution of topics, thereby enabling a wide range of interpretations. For example, when analyzing how a topic changes over time, the conventional analytical framework requires dividing data by time period and conducting separate analyses before comparing the distributions or frequencies. With STM, all of this analysis becomes possible within a single modeling framework, providing a clearer picture of how consistent topics change over time. This analysis also leverages these strengths to separately examine the patterns of change in each topic over time, how topics vary by media outlet, and how they vary by topic, while simultaneously analyzing the context in which regulation and conflict are mentioned alongside the distribution of each topic. Because structural topic modeling estimates the relative prevalence of latent themes rather than directly measuring what the media omit, we treat the scarcity of resolution-relevant topics as evidence of low salience in coverage, not as proof of its absence.

4. Results

The annual distribution of documents used in the final analysis is shown in

Table 1. Annual Distribution of Documents.

Year	Number of Documents
2018	3266
2019	8430
2020	4154
2021	3808
2022	1608

below. Considering the total number of news articles published in a given year, it is readily apparent that articles containing both ‘regulation’ and ‘conflict’ as keywords simultaneously are not numerous.

To proceed with analysis based on these articles, it was first necessary to determine the total number of topics. This is the most important parameter set by the analyst in the practical implementation of the model; the challenge is that it is impossible to precisely determine the number of topics latently present in the documents. Therefore, the number of topics was varied continuously across multiple models, and the number that produced the best model fit was selected.

The fit statistics used were: residuals from the model; exclusivity, which quantifies how distinctly each derived topic differs from others; semantic coherence, which evaluates the consistency of each derived topic; and held-out likelihood, which estimates the likelihood of held-out words using a model estimated after removing some words. As shown in Figure 1, which presents results for topic counts ranging from 5 to 70, three of the four criteria improved as the number of topics increased, while semantic coherence was highest at 15 topics and declined consistently thereafter. To resolve this trade-off, five researchers, including the authors, independently read and labeled the topics produced by models with 20, 25, 30, 35, and 40 topics and compared them across models. This comparison indicated that the corpus supports a stable set of interpretable themes in the high twenties: fewer topics left distinct themes merged, whereas increasing the number beyond 30 recovered no additional reproducible themes and yielded only progressively more idiosyncratic ones. The 30-topic model was therefore selected as the most parsimonious specification that captured this stable set, and the substantive patterns central to our analysis recurred across the alternative solutions.

The topics derived from the final model through the procedure described above are summarized in

Table 2. Topics from Analysis Results.

Topic No.	Proportion in Docs	Key Words	Derived Topic
Topic 5	8.30%	Japan, Korea–Japan, our country, government, conflict, bilateral, relations, position, export regulation, measures, prime minister, Moon Jae-in, dialogue, Abe, situation	Tensions in Korea–Japan Relations
Topic 17	7.00%	regulation, government, minister, measures, Ministry of Economy and Finance, system, new industry, sector, regulatory innovation, ministry, promotion, improvement, relaxation, expansion, support	Regulatory Innovation Efforts Related to New Industries
Topic 8	6.60%	new industry, industry, build, foster, establish, support, project, enterprise, foundation, sector, development, future, center, region, promote	Provision of Diverse Policy Support to Actively Foster New Industries
Topic 6	6.10%	Gyeonggi, US–China, outlook, impact, economy, quarter, decline, situation, exports, our country, global, rise, decrease, possibility, growth rate	Heightened Uncertainty and Economic Sluggishness Due to Negative Changes in the International Environment
Topic 24	5.90%	National Assembly, bill, amendment, regulation, controversy, law, lawmaker, provision, concern, revision, conduct, legislation, government, claim, point	Various Conflicts Arising in the Legislative Process Related to Regulation
Topic 1	5.40%	Democratic Party, president, Democratic Party of Korea, Moon Jae-in, lawmaker, presidential election, National Assembly, candidate, election, people, opposition, ruling, Blue House, leader, government	Election-Related News
Topic 30	4.90%	enterprise, economy, government, new industry, regulation, industry, our country, investment, companies, competitiveness, chairman, global, innovation, growth, large enterprise	Efforts to Enhance International Competitiveness Through New Industry Regulatory Innovation
Topic 2	4.50%	project, region, residents, regulation, conflict, officials, nearby, residents, plan, scale, development, facility, construction, promotion, construction	Resident Conflicts Due to Regulations in Regional Development Processes
Topic 14	4.50%	Japan, exports, semiconductor, materials, export regulation, our country, Korea–Japan, measures, items, government, regulation, enterprise, conflict, retaliation, economy	Trade-Related Conflict Due to Tensions in Korea–Japan Relations
Topic 25	4.30%	United States, Trump, China, president, Donald, US–China, administration, world, Biden, government, Xi Jinping, regulation, local time, trade, conflict	US–China Relations Conflict
Topic 7	4.30%	technology, service, market, United States, enterprise, domestic, world, global, development, artificial intelligence, data, project, platform, sector, our country	Growth of Domestic Companies in Various New Industry Sectors Needed
Topic 4	4.30%	regulation, COVID, accident, government, conflict, situation, police, one day, Seoul, occurrence, first, nationwide, photo, safety, distance	Social Distancing-Related Situations
Topic 3	3.50%	service, taxi, industry, project, platform, conflict, taxi industry, government, regulation, vehicle, representative, carpooling, mobility, innovation, startup	Mobility-Related Conflict Situations
Topic 10	3.40%	region, development, policy, role, metropolitan area, support, conflict, change, resolution, population, center, local, administration, future, citizen	Need for Balanced Regional Development
Topic 12	3.30%	jobs, minimum wage, employment, increase, policy, income, labor, economy, enterprise, growth, wage, government, fiscal, union, worker	Debate Among Various Disagreements Over Economic Policy
Topic 19	3.30%	real estate, housing, apartment, housing prices, Seoul, supply, market, price, government, reconstruction, regulation, household, countermeasures, speculation, loan	Real Estate-Related Issues
Topic 27	2.90%	era, our country, nation, state, history, reality, value, United States, future, century, world, politics, past, world, change	Meaningless Topic
Topic 9	2.70%	finance, bank, stocks, enterprise, investment, market, funds, regulation, assets, transactions, investors, investor, stock price, stock market, company	Situations Requiring Regulation in Stock Markets or Asset Investment
Topic 11	1.90%	attendance, Seoul, topic, attended, event, host, held, representative, discussion, attendees, forum, from left, Gwangju, debate, afternoon	Meaningless Topic
Topic 15	1.80%	production, energy, factory, supply, electric vehicle, China, proportion, automobile, world, demand, price, our country, United States, Europe, product	Growing Demand for Electric Vehicles Due to Interest in Eco-Friendliness
Topic 26	1.60%	I think, I heard, (speech markers), interview, question, really, first, I think, explanation, me too, let’s do it, really, probably, isn’t it	Meaningless Topic
Topic 23	1.50%	free, introduction, sponsorship, truth, marginalized, courage, Hankyoreh, persistently, create, always, shadowed, happiness, together, walk, flutter	Meaningless Topic
Topic 18	1.50%	North Korea, Korean Peninsula, inter-Korean, security, relations, diplomacy, people, politics, political, war, national interest, citizens, necessary, past, itself	Inter-Korean Relations Situation
Topic 13	1.20%	glass, necessary, effect, expert, impact, experts, aspect, help, interview, role, large, professional, result, possible, predict	Meaningless Topic
Topic 21	1.10%	allocation, next year, budget, supplementary, small business owners, budget bill, fiscal, vitality, budget, people’s livelihood, people, National Assembly, full-scale, vulnerable, social safety net	Direction of Government Support After COVID-19
Topic 29	1.00%	fairness, nation, politics, Republic of Korea, power, respect, regime, people, president, jobs, state, market, taxes, world, vested interests	Meaningless Topic
Topic 16	0.90%	Korean Peninsula, peace, inter-Korean, Korean laborers, World Cultural Heritage, World Heritage, Sado Mine, concealing, nation, withdraw, forced labor, historical distortion, prosperity, oldest, history	Issues Related to Japan’s Historical Distortion
Topic 20	0.90%	respect, people, Republic of Korea, Korean Peninsula, era, peace, innovation, hope, nation, economy, cooperation, inter-Korean, sector, jobs, center	Meaningless Topic
Topic 28	0.70%	art, naturally, visit, work, learn, college student, event, music, splendid, performance, experience, try, creative, exhibition, ordinary	Meaningless Topic
Topic 22	0.60%	national petition, reason, Blue House, deservedly, lifetime, Joseon, caught, strangled, discernment, tears, democracy, constitution, pain, nation, evaluated	Meaningless Topic

. The “Derived Topics” in the rightmost column were assigned by the authors based on independent readings of the keywords and representative documents for each topic, with disagreements resolved through discussion. “Proportion in Documents” is a model parameter expressing the probability that a given topic appears in a document. As Table 2 shows, several topics appear unrelated to the phenomenon of interest. This can be attributed in part to the way the notions of “conflict” and “dispute” are expressed in Korean. Although the two carry somewhat different connotations in English, both are rendered by the single Korean word “갈등 (galdeung).” Because this word applies wherever differing opinions or interests collide, it is readily associated with contexts such as international tensions, elections, and the COVID-19 pandemic, which are consequently drawn into the corpus. A small number of “meaningless” topics likewise appear to stem from residual noise that our text-cleaning process could not fully eliminate. This pattern reflects a deliberate trade-off in our search strategy. Because narrower, industry-specific terms yielded too little text for reliable estimation, we adopted broader keyword combinations to secure an adequate corpus, accepting that some tangential topics would inevitably be retrieved. The reading-and-labeling procedure described above allows us to identify such topics and set them aside, so that the interpretation focuses on topics genuinely relevant to regulatory conflict over new industries.

A notable observation is that, while several regulatory conflicts are fairly well known, and many regulatory changes are needed for newly emerging industries to thrive, there is almost no mention of specific efforts to resolve these conflicts. However, policy efforts to address these conflicts were not entirely absent during this period: the Presidential Committee on the Fourth Industrial Revolution operated “regulatory hackathons” that brought stakeholders together to negotiate contested regulations, and the Ministry of Economy and Finance promoted a mediation scheme known as the “One Step Model.” This finding may therefore be evidence that existing policy efforts have not received significant public attention and have not brought about meaningful change. That said, given that conflict resolution efforts rarely produce visible results in the short term, it would be difficult to conclude that policy efforts were meaningless simply because they failed to generate change significant enough to attract media coverage.

Considering only the regulation-related topics, the topic estimated to appear with the highest probability was the need for and efforts toward regulatory innovation related to new industries. However, as shown in Table 2, there is little news related to any specific alternatives. This point is also readily confirmed in

Table 3. Representative Documents Related to Topic 17.

Document Content
[Chungbuk Ilbo, Reporter Lee Jong-eok] Boeun County has announced it will identify and improve unreasonable regulations hidden in residents’ daily lives and workplaces. The county announced on the 6th that it will solicit ‘Boeun County People’s Livelihood Regulatory Innovation Proposals’ to receive direct suggestions from residents on regulations that inconvenience or are unreasonable for residents, and improve them. The proposal categories are five fields closely related to residents’ daily lives and economic activities: △social welfare △daily life △employment/jobs △small business/SME support △new industry fields. Accepted proposals will go through review by the Boeun County Regulatory Reform Committee, with outstanding items selected and prize money awarded in March. For proposals that can be resolved independently, the county plans to address them immediately through revision of Boeun County autonomous regulations. For cases requiring legal or institutional improvement, the county plans to actively work to reflect them in civil complaint and administrative systems by formally submitting them to relevant central ministries.
[Yesan] Yesan County will hold the ‘2021 Yesan County People’s Livelihood Regulatory Improvement Contest’ to discover and improve unreasonable regulations hidden in daily life and work sites, targeting the entire population. The contest period runs until next month’s 5th, and any interested citizen may apply. Contest categories are five people’s livelihood regulations: △civic welfare △daily life △employment/jobs △small business/SME support △new industries. Entries excluding simple petitions and complaints related to taxes, subsidies, and fines will be selected through a preliminary qualification review and final review by the Yesan County Regulatory Reform Committee, with prizes including certificates and Yesan Love gift certificates (Best Award: 1 person, 500,000 won; Excellence Award: 2 people, 300,000 won each; Encouragement Award: 4 people, 100,000 won each). Applications can be submitted by downloading the form from the Yesan County website’s notices and announcements section and submitting by mail or email.
The 2018 Customs Administration Regulatory Reform Public Proposal Contest is accepting entries until the 31st. Contest areas include regulatory reform proposals related to customs administration such as job creation, new technology and new industry promotion matters, and inconveniences for citizens and businesses; anyone including companies, general citizens, and public officials may participate. Applications are accepted until the 31st through the People’s Proposal section of the People’s Shinmungo system and by email, with final results to be announced at the end of July after document screening. The Best Award (1 entry) receives the Commissioner of Korea Customs Service Award and a prize of 1 million won. The Excellence Award (3 entries) receives 500,000 won each, the Encouragement Award (6 entries) receives 300,000 won each, with a total of 10 entries receiving awards.

, which lists representative documents that best capture this topic. Most articles mention the possibility of regulatory change due to new industries or other changes, but concrete solutions or directions for change are rarely found in most articles.

The same point can be readily confirmed in the second topic, ‘Provision of Diverse Policy Support to Actively Foster New Industries.’ As seen in

Table 4. Representative Documents Related to Topic 8.

Document Content
Gyeongnam Province announced it plans to invest 142.6 billion won across 7 sectors this year to revitalize the sluggish regional industry. This comes as traditional manufacturing industries such as shipbuilding and machinery in South Gyeongsang Province face a downturn, requiring structural upgrading of existing industries and discovery and cultivation of new industries to respond to the Fourth Industrial Revolution… [content continues with details of investment in smart manufacturing, robotics, aerospace, and other sectors]
South Jeolla Province announced that 46 new and ongoing projects totaling 129 billion won in government budget for economic and industrial sectors have been allocated in the 2020 government budget to invigorate new growth industries… [content continues with details of investment in e-mobility, drones, vaccine industry, energy new industries, shipbuilding, and other sectors]
Ulsan City has begun in earnest to build smart manufacturing industries fusing Fourth Industrial Revolution technologies with existing major industries such as automobiles, shipbuilding, and petrochemicals, to upgrade existing major industries and foster new industries… [content continues with details of the U-Smart Industry 4.0 roadmap and various strategic projects]

, which lists representative documents for this topic, despite various mentions of policy support, these mentions relate only to budget allocation or priority support areas, with conflict arising from regulation mentioned minimally.

The next topic, ‘Various Conflicts Arising in the National Assembly Legislative Process Related to Regulation,’ emphasizes only that various opinions conflict within the National Assembly rather than conflicts related to regulation per se, with little mention of regulation itself. Topics such as ‘Efforts to Enhance International Competitiveness Through New Industry Regulatory Innovation’ and ‘Resident Conflicts Due to Regulations in Regional Development Processes’ also contained few specific mentions of solutions to regulation-related conflicts.

The results for representative documents related to ‘Mobility-Related Conflict Situations,’ which is most directly relevant, are shown in

Table 5. Representative Documents Related to Topic 3.

Document Content
The Taxi-Carpooling Social Grand Compromise Body signed an agreement allowing carpooling services during commute hours. This came 45 days after the four taxi associations including the National Taxi Transport Business Association, carpooling operator Kakao Mobility, Democratic Party of Korea Taxi-Carpooling TF Chairman Jeon Hyun-hee, and the Ministry of Land, Infrastructure and Transport put their heads together. It is commendable that the taxi industry, which had been conducting hardline struggles demanding ‘abolition of carpooling,’ and the carpooling industry resolved their conflict and opposition to reach some form of agreement. However, it remains uncertain whether this agreement can reignite the Korean-style ride-sharing market. The core of this agreement—allowing carpooling during commute hours—could actually become a constraint regulating the ride-sharing business… [article continues discussing limitations of the agreement and reactions from various stakeholders]
Carpooling companies such as Poolus and SOCAR said on the 7th regarding the Taxi-Carpooling Social Grand Compromise Body’s agreement: ‘The seeds of innovation through vehicle sharing have been completely cut. It will remain as a bad precedent going forward.’ Poolus, the largest domestic carpooling company, stated: ‘A conclusion with no practical effect has emerged. Citizens will not agree that late-night hours when taxis are hardest to find cannot use carpooling.’ The most taxi-scarce time is late night after midnight. The essence of carpooling is drivers and passengers with the same destination sharing a ride to reduce costs. Yet the most high-demand late-night hours were excluded… [article continues with reactions from various stakeholders including SOCAR CEO’s criticism of the agreement]
The prosecution ruled on the 28th that ‘Tada,’ VCNC’s minivan hailing service, violated the Passenger Transport Service Act. VCNC’s CEO Park Jae-wook and the parent company SOCAR’s CEO Lee Jae-woong were indicted without detention. The government, ruling party, and now prosecution appear to be pressuring Tada. Forecasts suggest the mobility (transportation) industry branded as ‘sharing economy’ will contract. The Seoul Central District Prosecutors’ Office Criminal Division 5 announced it had indicted the two CEOs and two corporations for violating the Passenger Transport Service Act without detention on the 28th… [article continues with details of the prosecution and reactions]

. Here too, there is almost no reporting on specific methods of resolution; rather, reports indicate that proposed compromise agreements were not accepted, and legal resolution was ultimately relied upon. From the perspective of conflict resolution, such reporting reflects a reality in which resolution remains distant, while the complex negotiations among stakeholders—the primary method of resolving these conflicts—are not being properly reported.

If interpreting each derived topic was a strength of conventional topic modeling, the STM used in this study has the additional advantage of analyzing the relationships among these topics and the influence of other contextual factors on their emergence. Figure 2 shows a network visualization of the correlations among the 30 topics analyzed here.

Figure 2 uses a gravity model representation in which topics of higher importance appear toward the center of the figure, and topics with higher proportions are represented by larger circles. Overall, one can observe that regulation–conflict-related topics appear on the periphery. Topics appearing at the center—such as Topic 17, ‘Regulatory Innovation Efforts Related to New Industries,’ and Topic 24, ‘Various Conflicts Arising in the Legislative Process Related to Regulation’—are connected to various topics in terms of overall regulatory change. However, the considerable overlap between these two topics suggests they convey only that legislative changes are being attempted in response to the need for such changes, with almost no content on other aspects of conflict resolution. Meanwhile, Topic 3, ‘Mobility-Related Conflict Situations,’ in the right corner, is the most meaningful topic related to regulation–conflict, but its close connection to Topic 2, ‘Resident Conflicts Due to Regulations in Regional Development Processes,’ and Topic 7, ‘Growth of Domestic Companies in Various New Industry Sectors Needed,’ suggests this topic does not contain much information related to conflict resolution. A close examination of this analysis ultimately shows that information related to resolution of regulation–conflict was not prominently featured. The generally dispersed distribution of related topics also indicates that information related to regulation–conflict is being reported in a relatively fragmented manner.

The next aspect to examine in relation to the context of topics is change over time—the most fundamental dimension of change. This examines how much each topic changes over time. While more complex patterns of change over time are certainly possible, here the relationship with time was modeled as linear. Looking at Figure 3, the topic most relevant to this research’s subject, ‘Mobility-Related Conflict Situations,’ can be somewhat understood as declining in frequency, given that it is a conflict for which the main discussions were held in the past. However, the tendency for media coverage of topics such as ‘Regulatory Innovation Efforts Related to New Industries’ and ‘Efforts to Enhance International Competitiveness Through New Industry Regulatory Innovation’ to decline is concerning, given that there are few resolved cases of conflicts related to new industry regulation. Meanwhile, the increasing mentions of topics such as ‘Provision of Diverse Policy Support to Actively Foster New Industries’ and ‘Growth of Domestic Companies in Various New Industry Sectors Needed’ are understandable trends, but the declining mentions over time of topics directly connected to problem resolution—such as ‘Efforts to Enhance International Competitiveness Through New Industry Regulatory Innovation’ and ‘Regulatory Innovation Efforts Related to New Industries’—are not desirable changes from the perspective of conflict resolution.

Figure 4 analyzes topic emergence patterns by media outlet to examine how topics vary by the medium through which they are conveyed. One of the most easily identified findings is that topics representing substantive government efforts—such as ‘Regulatory Innovation Efforts Related to New Industries’ and ‘Efforts to Enhance International Competitiveness Through New Industry Regulatory Innovation’—are concentrated only in certain regional newspapers and not prominently mentioned in other media. This is consistent with the finding that these topics are mentioned less frequently over time. Meanwhile, the topic of ‘Various Conflicts Arising in the Legislative Process Related to Regulation’ appears to be mentioned quite frequently across a relatively diverse range of media. The topic ‘Growth of Domestic Companies in Various New Industry Sectors Needed’ is heavily mentioned in certain central newspapers while largely absent from broadcasting, and ‘Mobility-Related Conflict Situations’ is reported through a relatively diverse range of media including broadcasting and central newspapers. ‘Situations Requiring Regulation in Stock Markets or Asset Investment’ is found to be mentioned with unusual frequency in specific outlets. The fact that ‘Provision of Diverse Policy Support to Actively Foster New Industries’ is frequently mentioned only in regional newspapers also implies that necessary information from the perspective of regulation–conflict is not being adequately conveyed.

Figure 5 examines patterns of topic emergence according to the subject classification of each news article, showing which section of news coverage each topic primarily appeared in. While there are about 30 actual classifications, these have been reorganized into 8 broader categories for ease of interpretation. What stands out here is that topics such as ‘Regulatory Innovation Efforts Related to New Industries’ and ‘Efforts to Enhance International Competitiveness Through New Industry Regulatory Innovation’ appear in the ‘local’ news section and in articles covering ‘IT–Science General.’ Typically, if the narrative around these topics had been regulatory innovation efforts related to conflict, it would have been more likely to appear in ‘Society General’ or ‘Politics General.’ The fact that they appear in other sections suggests the context in which these topics were mentioned was somewhat different. The fact that ‘Provision of Diverse Policy Support to Actively Foster New Industries’ is mentioned overwhelmingly in the ‘local’ classification also implies that information of a somewhat different nature from overall institutional improvement is being conveyed. The fact that topics such as ‘Growth of Domestic Companies in Various New Industry Sectors Needed’ and ‘Efforts to Enhance International Competitiveness Through New Industry Regulatory Innovation’ are primarily conveyed in ‘IT–Science General’ also indicates the high likelihood that the broad topic of new industries and regulation has been reported from a somewhat different angle than institutional improvement.

5. Discussion

5.1. Summary of Findings

The analysis yields three findings. First, regarding RQ1, media coverage at the intersection of regulation and conflict is dominated by topics announcing the need for regulatory innovation and promoting new industries, while topics conveying concrete processes of conflict resolution are nearly absent. Even in the mobility dispute—the most prominent regulatory conflict of the period—coverage focused on the confrontation itself and its legal outcomes rather than on the negotiation processes through which stakeholders sought compromise. Second, regarding RQ2, the visibility of resolution-relevant topics declined over the observation period, and such topics were concentrated in regional newspapers and IT–science sections rather than in broadcasting and major national dailies. Third, regarding RQ3, the placement of regulatory innovation topics in the local and IT sections—rather than in the politics or society sections, where conflict and its mediation are typically covered—suggests that regulatory reform was framed primarily as industrial promotion news, not as a matter of resolving competing social claims.

5.2. Theoretical Interpretation: The Invisible Middle of Regulatory Conflict

These findings can be interpreted through the framework developed in Section 2. If regulatory conflict is structurally rooted in the asymmetric distribution of burdens between regulated incumbents and beneficiaries, and intensified by regulatory disconnection in emerging industries, then its resolution requires renegotiating the regulatory bargain among incumbents, entrants, and the state. What the media render visible, however, is only the two endpoints of this process: the outbreak of conflict and its formal conclusion (an agreement signed, a service banned, an executive indicted). The middle—the protracted, complex negotiation among stakeholders in which the actual redistribution of burdens is worked out—remains invisible. This pattern is consistent with the well-documented tendency of news media toward episodic rather than thematic framing (Iyengar, 1991): conflict events are newsworthy, while slow-moving negotiation processes are not. The declining coverage over time, despite the conflicts remaining unresolved, likewise mirrors the issue-attention cycle (Downs, 1972), in which public attention recedes long before the underlying problem is addressed.

The coverage patterns we document are suggestive of an informational environment poorly suited to sustaining negotiated regulatory change. As Wilson’s (1980) typology suggests, disputes such as the mobility case approximate interest-group politics: concentrated incumbents and concentrated entrants contest the rules while the diffuse public—formally the beneficiary of both consumer protection and innovation—stands outside the arena. Where media coverage does not supply the public with information about what is being negotiated and what compromise would entail, the diffuse public may be less able to function as a counterweight, so that the discourse is left largely to the concentrated parties. We emphasize that this is an inference about the likely consequences of the coverage patterns we observe, not a relationship our data directly measure. The finding that government conflict-resolution initiatives, such as hackathons of the Presidential Committee on the Fourth Industrial Revolution and the ‘One Step Model’ of the Ministry of Economy and Finance, left almost no trace in the corpus may thus be read in two ways: as evidence that these initiatives failed to produce results visible enough to warrant coverage, or as evidence that resolution processes are systematically under-reported regardless of their substance. Given that conflict resolution rarely produces short-term visible outcomes, the truth likely combines both.

5.3. Policy Implications

Three implications follow for governments seeking to manage regulatory conflict in emerging industries. First, communication about regulatory reform may need to shift from promotion to mediation. In practice, this means that regulators’ public messaging would foreground the distributive stakes of reform—who bears the costs of regulatory change and what claims must be addressed—rather than leading with innovation agendas and support budgets. This differs from existing regulatory-innovation discourse not in subject but in framing: the same sandbox program can be communicated as an industrial-promotion announcement (its current dominant frame) or as an account of competing claims being reconciled. In the vocabulary of interactive marketing, this is a shift from one-way promotional messaging to dialogical communication, in which the public can observe competing claims being engaged rather than a finished message being delivered (Herrada-Lores et al., 2025; Wang, 2024).

Second, the negotiation process itself can be made deliberately visible, with the diffuse public as the primary audience. Relevant regulators or institutionalized compromise bodies could generate the process information the media otherwise omit; mandating that such bodies report interim progress at regular intervals—rather than communicating only when a final agreement is signed—would supply the missing middle of the public narrative and give the public a stake in the process before its conclusion. To be clear, our argument concerns the visibility of the process, not the exposure of its content: since publicity can harden bargaining positions (Elster, 1998; Stasavage, 2004), what should be communicated is not the substance of negotiations but basic process information, such that talks are underway, who participates, and what stage has been reached. Periodic reporting of this kind preserves the confidentiality that bargaining requires while giving the public grounds to regard an eventual compromise as legitimate.

Third, because regulatory-innovation coverage is concentrated in regional and IT-specialized outlets, mainstream political and society desks appear to lack ready frames for treating regulatory reform as a conflict-resolution story. Regulators can lower the cost of adopting such frames by providing background briefings that explicitly lay out stakeholder trade-offs and negotiation status, and by designating spokespersons responsible for the conflict-resolution narrative rather than for innovation promotion. This can be a low-cost intervention that targets journalists as intermediaries to the diffuse public.

6. Conclusions

Applying structural topic modeling to 21,266 South Korean news articles (2018–2022), this study finds that while regulatory innovation and new-industry promotion are heavily covered, the processes by which regulatory conflicts are negotiated and resolved are nearly invisible, decline in salience over time, and are relegated to peripheral outlets and sections. Here “nearly invisible” refers to the low estimated prevalence of resolution-relevant topics in the corpus—their small proportion in documents relative to other topics—rather than to a direct measurement that such reporting is absent. We argue that this informational gap is not incidental: it reflects the episodic logic of news production interacting with the structural character of regulatory conflict, and it might weaken governments’ capacity to legitimize negotiated regulatory change.

Beyond the Korean case, the study makes two contributions to regulatory-governance research. Substantively, it shifts attention from the design of regulatory-reform instruments to the public representation of regulatory conflict—a dimension that scholarship focused on regulators and regulated industries has largely passed over. Methodologically, it shows how covariate-aware topic modeling can move the study of regulatory politics beyond single-case narratives toward systematic analysis of public discourse at scale.

These findings should be read in light of several limitations. Media coverage is an indirect measure of public perception, and audience-level data would be needed to trace how the patterns documented here translate into individual perceptions; the keyword-based corpus may underrepresent disputes framed in other terms; the single-country, single-period design leaves open whether the invisibility of resolution processes is general or specific to Korea’s institutions; and topic interpretation rests on researcher judgment, for which comparable validation methods remain scarce. Future work could link media framing to audience reception, extend the window beyond 2022, and employ embedding- or LLM-based methods such as BERTopic to validate the topic structure identified here. Relatedly, although the topic structure proved reproducible across STM specifications with different numbers of topics, we did not undertake external validation such as manual coding of a random article sample with formal inter-coder agreement statistics or comparison with a different family of topic models. Such complementary procedures would further strengthen confidence in the identified themes.