The erosion of the gatekeeping function of photojournalists in the digital age

Abstract

In today’s digital news media environment, where the number of professional staff photojournalists is declining, this study examines the extent to which images outsourced from external sources, including vested groups, are published, and how the visual quality of images varies by photo source: 1) staff photojournalist, 2) wire services, 3) social media, 4) news sources, and 5) citizens. Computational and statistical analyses of 7,651 online news photos published between January 1, 2024, and December 31, 2024, by five major South Korean newspapers, revealed that the quality of visual journalism, based on photo characteristics, varied by photo source. The results also indicate that both the topics and the accuracy of visual representation varied by photo source: photos taken by staff photojournalists were more closely aligned with the article content than those from other sources. Furthermore, photos from social media and news sources tended to depict human faces with positive emotions, whereas those from wire services more often showed negative emotions. This suggests that the gatekeeping role of photojournalists as news gatherers may be weakening, enabling external sources, including vested interest groups, to promote their own agendas.

Keywords

visual journalism gatekeeping photo sources photo quality computational analysis

Introduction

In the digital age, the current trend in visual journalism worldwide is a decrease in images provided by staff photojournalists, alongside an increase in the use of images from external sources, such as wire services, citizens, and even news sources (Ferrucci et al., 2020; Mortensen and Gade, 2018). Although this phenomenon is largely attributed to the financial crises of news organizations, which have led to a reduction in the number of staff photojournalists (Mortensen and Gade, 2018; Thomson, 2018), advancements in camera technology have also led news managers to anticipate that nonprofessional journalists may increasingly supplant the work of skilled professional (Greenwood and Thomas, 2015; Mast and Hanegreefs, 2015).

However, the effects of this phenomenon may be more serious, weakening professional photojournalists’ gatekeeping roles and resulting in a decline in the quality of photojournalism. Gatekeeping refers to the process by which journalists, editors, and news organizations select and omit items, including photographs, based on the concept of news value (Shoemaker et al., 2009; White, 1950). Therefore, the gatekeeping role involves journalists’ professional decisions about which events to report based on their editorial judgment. The increased use of photographs taken by external sources, including non-professional photographers, has led to negative journalism effects because non-professionals may disregard professional photojournalistic rules that ‘have come to define quality photojournalistic images’ (Lough and Mortensen, 2023: p. 2017).

Given the well-documented power of images (Bissell, 2000) and the ability of professional staff photojournalists to produce images with news value and greater representativeness (Klein-Avraham and Reich, 2016; Lough and McIntyre, 2019), recent studies have highlighted the displacement of news by citizen-generated content (Greenwood and Thomas, 2015; Mast and Hanegreefs, 2015), the stock photography industry (Mortensen et al., 2023), social media, and even news sources. Although scholars have noted an increasing number of news organizations incorporate photographs provided by interest groups, including public relations and news sources (Schwalbe et al., 2015), few studies have examined the extent to which images provided by non-staff photographers displace the work of staff photojournalists, particularly in online news environments.

Therefore, drawing on prior research, this study examined whether and, to what extent, the quality of visual journalism differs depending on photo sources, particularly between professional staff photojournalists and non-staff external sources. To that end, this exploratory study analysed 7,651 lead photographs published online between January 1, 2024, and December 31, 2024, by five major South Korean newspapers by using computational approaches. In South Korea, more than seven out of 10 people consume their daily news through online platforms—one of the highest levels reported in global surveys conducted by the Reuters Institute for the Study of Journalism (Newman et al., 2021). Thus, South Korea provides an appropriate context for examining news photographs published online.

The findings of this study contribute to journalism research by suggesting that the gatekeeping role by staff photojournalists has eroded, and by documenting the distinct differences in quality between online news photographs taken by staff photojournalists and those captured by non-staff professionals.

Literature review

Journalistic gatekeeping and digital media

Gatekeeping is perceived as crucial in journalism because it is fundamental to the news production process. The concept of gatekeeping, initially proposed by Lewin (1947), refers to ‘the process of selecting, writing, editing, positioning, scheduling, repeating and otherwise massaging information to become news’ (Shoemaker et al., 2009: p. 73). Gatekeeping theory is one of the oldest in journalism studies and it has been conceptualized as a process carried out by individuals and shaped by organizational structures. Especially in early studies conducted until the 1970s, researchers influenced by White (1950) understood gatekeeping as a process influenced by individual journalists.

However, subsequent studies have shown that gatekeeping may not be the responsibility of a single journalist or editor but rather a process that operates across multiple levels within a news organization. For example, beginning in the 1980s, Gans (1979), working from the perspective of the sociology of news work, helped initiate the study of gatekeeping and multiple gatekeepers in their organizational context (Shoemaker and Vos, 2009). Gatekeeping is now generally understood not only as an individual journalist’s selection of information but also as a process in which information gathered from various sources passes through multiple levels of gates within a news organization before being shaped into a product for public dissemination. Thus, similar to news stories, photographs published in newspapers also undergo a gatekeeping process.

In the digital media environment, however, scholars have expressed concerns regarding the role of gatekeeping in journalism. From a withering perspective, researchers highlight that traditional ‘gates’ and gatekeepers are struggling to sustain their long-standing role of selecting, from an ever-expanding pool of information providers in the digital media environment, that which is deemed sufficiently valuable to communicate to the public (Singer, 2006). In contrast, from a shifting-role perspective, scholars contend that although journalists’ gatekeeping responsibilities have fundamentally transformed—from producing original content to functioning as curators—this evolution does not signal a wholesale decline in journalistic gatekeeping roles (Bakker, 2014; Schwalbe et al., 2015).

For example, Bakker (2014) analysed job advertisements for journalists and editors posted on Villamedia.nl over the course of approximately 1 month, as well as job postings from 62 major Dutch media companies. They argued that producing original content is no longer the primary task of contemporary journalists. Instead, journalists are increasingly functioning as ‘harvesters’, ‘managers’, and ‘curators’ of information generated outside their own organizations. In the context of photojournalism, Schwalbe et al. (2015) drew on findings from an online survey and interviews with elite photojournalists, approximately half of whom reported that visual journalists’ roles were shifting from content creation to content curation. They proposed that ‘the traditional gatekeepers and gate watchers in legacy newsrooms are evolving into ‘gatecheckers’ who select, verify, and curate visuals’ (p. 476).

Specifically, during crisis events such as terror attacks, the active participation of citizens or amateur photographers is becoming increasingly important in the process of news production (e.g., Mast and Hanegreefs, 2015) because they are ‘the person first on the scene with a camera’ (Allan, 2014: p. 134) or acting as a ‘citizen witnessing’ (Allan, 2013). However, they are not professionals, and many scholars have thus raised concerns about ethical and normative issues, criticizing this type of participation as ‘me-first journalism’, ‘unsubstantiated amateur footage’, or ‘unreliable crowd-sourced material’ (Fry, 2013, cited in Allan, 2014: p. 136). In contrast, several scholars have proposed that professional photojournalism could benefit from citizen witnesses in the digital media environment (Allan and Peters, 2015). This is not a matter of replacement but of complementarity; previous studies on citizen photojournalism have defined it as involving ‘amateurs who do not identify as journalists and are not bounded by journalistic norms, standards, guidelines, code of ethics or procedures, but create and produce images, video and multimedia that are integrated into traditional media reports’ (Brennen and Brennen, 2015: p. 520). As this type of journalism has increased, the gatekeeping role of professional journalists, who can curate or verify the content, has become more important.

News value and visual journalism

Fundamentally, the gatekeeper’s role is to determine what should be reported based on newsworthiness. Since Galtung and Ruge’s (1965) seminal study on news factors, news value has remained an enduring topic in journalism research. According to Bednarek and Caple (2012, 2017), news values have been conceptualized in two primary ways: (1) a material perspective, which locates news values in the characteristics of actual events and people (Galtung and Ruge, 1965); and (2) a cognitive perspective, which views news values as residing in the minds of journalists (Donsbach, 2004; Harcup & O’Neill, 2001; Harrison, 2010). The material perspective emphasizes inherent attributes of events or actors, such as impact, proximity, prominence, and negativity. By contrast, the cognitive perspective conceives news values as professional or organizationally learned conventions.

Third perspective is the discursive perspective that focuses not only on events or actors, but also on how and why they are constructed as newsworthy (Bednarek and Caple, 2017). From this viewpoint, news values are constituted through news factors and discursive devices, such as language and images. For example, Bednarek and Caple (2012) analysed news articles published in newspapers such as The Sydney Morning Herald, and demonstrated that evaluative language (e.g., celebrity, momentous), superlatives (e.g., up to, more than), and historical comparisons function as semiotic devices that contribute to the construction of newsworthiness.

Aligned with the discursive perspective that news values are constructed primarily through ‘what is presented in images (content) and to a lesser extent through how the image is technically realized’ (Caple and Bednarek, 2015: p. 445), this study proposes that photojournalism embodies both journalistic news values and distinctive visual qualities. Accordingly, the quality of photojournalism should be evaluated along two dimensions: news factors, understood as attention to events or individuals from a material perspective (Galtung and Ruge, 1965), and the visual features of the photographic work itself (Lough and Mortensen, 2023; Mortensen and Gade, 2018). Although these two elements are not always clearly delineated, assessments of newsworthiness often consider the significance of the news topic, human components, the representation of diversity, including race and gender (Kim and Smith, 2005; Lough and Mortensen, 2023). For instance, Kim and Smith (2005), through a content analysis of 60 years of Pulitzer Prize-winning photographs (1942–2002), found that award-winning domestic news images were predominantly characterized by a negative valence, with a primary focus on social issues such as crime and poverty.

Quality differences in visual journalism depending on image providers

There has been an increased reliance on external providers, such as wire services, freelancers, citizens, and even stock imagery companies, referring to businesses that manufacture photographic images primarily for use in marketing promotions and advertising (Greenwood and Thomas, 2015; Mortensen et al., 2023; Thomson, 2018).

Moreover, with the advancement of digital technology, influential interest groups traditionally regarded as news sources—such as elite politicians and government officials, can now interact directly with news audiences by disseminating information and uploading self-produced images online. As news reporters are increasingly bypassed by elite sources in social media and digital media environments (Carlson et al., 2021), a similar trend may be observed in the field of photojournalism.

Consequently, scholars in visual journalism have expressed concerns that this trend may affect not only the quality of news images, but also audience trust in the news. Professional staff photojournalists’ work has been found to differ from that of nonprofessional photojournalists in several aspects (Greenwood and Thomas, 2015; Klein-Avraham and Reich, 2016; Nilsson, 2017). Based on professional and rigorous gatekeeping practices recognized as a fundamental aspect of journalism and intrinsically linked to journalism’s principles of objectivity, impartiality, and autonomy (Shoemaker and Vos, 2009; Singer, 2006), photojournalists should be autonomous and independent to ensure the unbiased and objective selection and representation of subjects, fulfilling their primary obligation—the pursuit of truth. For example, in an online survey on the factors that affect the gatekeeping role of visual journalists in the US (Schwalbe et al., 2015), 84% of respondents agreed or strongly agreed that objectivity and fairness is the most important factor; followed by the visual itself (47%).

Similarly, photojournalists prioritize the accurate representation of reality, a principle that Wheeler and Gleason (1995) describe as ‘an idea at the heart of photojournalism ethics’ (p. 8). Consistent with this perspective, Brennen (2010) proposed that news photographs document ‘truthful depictions of an objective reality’ (p. 73). By analysing images in the print edition of The Middletown (NY) Times Herald-Record in the U.S. before and after the newspaper laid off its entire photography staff, Mortensen and Gade (2018) also said, ‘they take different news images than non-professionals, and these images capture what is newsworthy’ (p. 1005). Furthermore, through an experimental study, Mortensen et al. (2023) demonstrated that participants perceived stock photographs as significantly less credible than those taken by staff photojournalists.

Building on the preceding discussion, this study examines the impact of images provided by external sources on visual quality and the first research question (RQ1) investigates the extent to which photographs produced by external sources, including both non-staff photographers and news sources, are utilized in South Korean newspapers.

RQ1 : To what extent do South Korean newspapers incorporate photographs from external sources? Specifically, what categories of image sources are represented in online news photographs published by South Korean newspapers?

Additionally, this study examines the quality of news photographs based on their news factors and visual features. In terms of news factors, the topic and accuracy of the images were assessed. Regarding visual techniques, this study considered the dominant emotion conveyed, the presence of people, and the vividness of the work. Specifically, this study highlights the contrast between images supplied by traditional news sources with vested interests and those produced by staff photojournalists (RQ2).

RQ2a : To what degree do online news photographs published in South Korean newspapers reflect news factors, such as the topic and accuracy of the images, and how do these attributes vary depending on the type of photograph producer?

RQ2b : To what degree do online news photographs published in South Korean newspapers exhibit visual features including the dominant emotion, human component, and vividness? In addition, how do these features vary depending on the type of photograph producer?

Methods

Data

Photos

Naver.com, the most popular online news aggregator in South Korea, was used to collect photographs of news articles. For each newspaper, Naver provides a daily list of the top three most-read news articles, categorized into six sections: Politics, Economy, Society, World, IT/Science, and Lifestyle/Culture. Among these six sections, this study focuses on articles from the hard news sections. Photos contained in the top three most-read articles per section were collected each day from news articles published by the five major newspapers¹ in South Korea–Dong-A Ilbo, Chosun Ilbo, JoongAng Ilbo, Kyunghyang Shinmun, and Hankyoreh–between January 1, 2024, and December 31, 2024 using a Python-based Web data scraper. Only the first photograph of each news article was used. Additional information on each news article, including the publication date, title, publisher, and article content, was collected. Brief descriptions and source information were also collected for each photograph. The number of photos initially collected after removing duplicates is reported in Table 1.

Table 1.

The number of photos by newspaper.

Chosun	Dong-A	Hankyoreh	JoongAng	Kyunghyang	Total
3,658	3,168	3,590	3,852	3,244	17,512

Photo sources

To examine the differences in visual qualities of photos according to the source, we categorized photo sources into five categories: staff photojournalists, wire services, social media, news sources, and citizens. The ‘staff photojournalist’ source indicates that photos were taken by a photographer employed by the newspaper, while the photos of the ‘wire service’ source were provided by a wire agency. Photographs acquired from social media were classified as social media photographs. In this study, a news source refers to the origin of the information used in a news article, such as individual people (including politicians), institutional or PR materials, and representatives from organizations such as government, companies, or other institutions who provide the basis for the news article. Finally, the photos provided by the readers were categorized as ‘citizens’.

Among the 17,515 initially collected photographs, 4,824 that did not belong to one of the five sources or did not have source information were removed from the dataset. The sources of the removed photographs included captures from TV screens and non-social-media webpages. A two-step approach was employed to identify the source of each photograph accurately. First, the source of a photo was identified using computational methods—specifically, a program developed by the authors using Python that extracted source information from the article. Subsequently, two human coders verified whether the source of each image had been correctly classified, and any images that were not accurately classified were corrected using the source information acquired from the newspaper. Subsequently, two human coders double-checked whether the sources were correctly identified. The numbers of photos by source are listed in Table 2.

Table 2.

The number of photos by source and newspaper.

	C^a	D	H	J	K	Total	C	D	H	J	K	Total
Citizen	14	12	4	11	9	50	14	12	4	11	9	50
Social media	600	364	176	493	132	1,765	600	364	176	493	132	1,765
Staff photographer	290	370	543	351	875	2,429	290	370	543	351	875	2,429
News source	180	131	208	182	206	907	180	131	208	182	206	907
Wire service	1,573	1,278	1,744	1,632	1,313	7,540	502	414	584	565	435	2,500
Total	2,657	2,155	2,675	2,669	2,535	12,691	1,586	1,291	1,515	1,602	1,657	7,651
	Before							After

^aC, Chosun; D, Dong-A; H, Hankyoreh; J, JoongAng; K, Kyunghyang.

The numbers on the left show the original distribution of the number of photographs by source. As shown, the number of photos provided by the wire services (7,540) was much higher than the number of photos provided by other sources. To make the distribution more even and reduce the total number of photos to be analysed, 2,500 wire service photos were randomly selected without altering the number of photos from other sources, as shown in the table on the right side. Table 2 also shows the number of photos by newspaper.

The number of photos by source and section is presented in Table 3. The proportion of photographs obtained from each specific source varied across sections.

Table 3.

Number of photos by source and section.

Section	Citizen	Social media	Staff photographer	News source	Wire service	Total
Economy	18 (1.2)	137 (8.8)	476 (30.5)	419 (26.8)	512 (32.8)	1562 (100)
Society	25 (1.2)	580 (28.9)	801 (39.9)	122 (6.1)	478 (23.8)	2006 (100)
World	1 (0.1)	819 (48.5)	134 (7.9)	29 (1.7)	707 (41.8)	1690 (100)
Politics	6 (0.3)	229 (9.6)	1018 (42.5)	337 (14.1)	803 (33.6)	2393 (100)
Total	50 (0.7)	1765 (23.1)	2429 (31.7)	907 (11.9)	2500 (32.7)	7651 (100)

*The numbers in parentheses indicate percentages.

Visual qualities: News factors and visual features

The differences in the visual qualities of photographs were analysed using computer vision and text mining techniques. Based on previous studies (Bednarek and Caple, 2017; Greenwood and Thomas, 2015; Mast and Hanegreefs, 2015), this study examined the visual quality of photographs using two components, news factors and visual features. Specifically, since previous studies have defined news factors as characteristics located in actual events (e.g., what is presented in images), this study measured each photo’s topic and accuracy of representation (i.e., the similarity between a photo and the article content). In addition, because visual features have been defined as ‘how the image is technically realized’ (Caple and Bednarek, 2015: p. 445), this study analysed the presence of people in the frame, the dominant emotion, and vividness of the image.

Topics of a photo

The topics represented in each photo were identified using topic modelling methods with a two-step approach. First, a short text description was generated for each photo using multimodal LLaMa 3.2², a multimodal large language model. Multimodal LLaMa 3.2 is a large language model developed by Meta in 2024 that is designed to process and generate responses based on both textual and visual inputs (Meta, 2024).

After generating a text description for each photo, a topic modelling technique was applied. BERTopic (Grootendorst, 2022) was used. BERTopic is a Transformer-based topic modeling algorithm that was released in 2022. According to the author, BERTopic performs better than traditional probability-based models such as LDA (Latent Dirichlet Allocation).

By applying this model to the text descriptions of the photos in the dataset, 106 topics were identified. Subsequently, similar topics were first identified using BERTopic. Next, two human coders verified whether the topics classified as similar were indeed similar, and those confirmed to be similar by the human coders were merged into a single topic, resulting in 61 topics. Appendix 1 presents the top five words representing each topic.

Accurate representation: Similarity between a photo and the article that embeds it

To measure the similarity between a photo and the article content, the Contrastive Language-Image Pretraining (CLIP) model developed by OpenAI (Radford et al., 2021) was used. The primary purpose of CLIP is to calculate the similarity between an image and its corresponding text. As the original content was too long for use with CLIP, the content was first summarized using the KoBART model.³ The KoBART model is a Korean version of the Bidirectional and Auto-Regressive Transformer (BART) model developed by SKT, a leading telecommunications company in South Korea. BART is a sequence-to-sequence model based on both the encoder and decoder parts of the transformer and was developed by Facebook AI in 2019 (Lewis et al., 2019). The summarized text was then fed into CLIP, along with the corresponding photo, which resulted in a similarity score between them.

Whether a person is present in a photo

To detect a person in a photo, the YOLOv11 object detection model released by Ultralytics in 2024 (Ultralytics, 2024) was employed. Among the versions of YOLOv11, the ‘nano’ version (YOLO11n) was utilized. YOLOv11 was pre-trained using the MS COCO (Microsoft Common Objects in Context) dataset (Ultralytics, 2024). The MS COCO dataset is a widely used large-scale dataset in computer vision, primarily used for object detection. It contains 80 object categories, such as persons, cars, buses, and dogs. This study used a pre-trained model without fine-tuning, as the primary purpose was to detect persons in the photo.

Dominant emotion

The dominant emotion reflected on the faces in the photographs was assessed. To measure the emotion of a face in a photo, this study employed DeepFace, a face-recognition program developed by Meta (Serengil, 2024). DeepFace achieved an accuracy of 97.25% on the Social Face Classification (SFC) dataset (Li, 2025). Using DeepFace, this study identified the main emotion of a face in a photo. DeepFace extracts seven emotions —‘angry’, ‘sad’, ‘disgust’, ‘fear’, ‘surprise’, ‘happy’, and ‘neutral’—with percentages indicating the extent to which each emotion is presented in a face. First, the emotions with the highest percentage in each face were identified. If a photo contained more than two faces, its dominant emotion was determined as most frequently occurring dominant emotion among the faces. In addition, the seven emotions were recategorized into three categories—positive (happy), negative (angry, sad, disgust, fear), and neutral. This is mainly because there is little difference between negative emotions, such as fear and anger. For these emotions, DeepFace often failed to make accurate distinctions. The ‘surprise’ emotions was excluded from the final measurement because it can either represent a positive or negative emotion, and it was infrequently observed in the dataset. Two human coders verified whether the emotion results of all the photos in the dataset derived using DeepFace had been correctly classified. Any misclassified images were reclassified based on the coders’ consensus. The number of photographs displaying each emotion is presented in Table 4.

Table 4.

The number of photos for each emotion.

Positive	Negative	Neutral	Surprise	Not applicable	Total
1233	419	2795	34	3170	7651

Vividness: Saturation and value

Vividness refers to the extent to which a visual stimulus is perceived as clear, detailed, and perceptually rich. It can be approximated using low-level visual features, such as saturation and value, because these directly shape sensory intensity and clarity (Kind, 2017). In particular, saturation reflects the richness and intensity of colors, while value (brightness) determines the visibility and clarity of visual details. Higher saturation and appropriate brightness increase perceptual information and make images appear more vivid. Highly saturated images tend to have bright and vivid colours, whereas low-saturation images are more likely to have subdued colours. Value refers to the lightness of a colour. An image with a high value has a lighter colour, whereas an image with a low value has a darker colour. OpenCV was employed to measure the saturation and value of each photograph⁴.

Statistical analyses

A series of statistical analyses to answer the research questions were employed. To assess whether the topic and dominant emotion of a photo differed according to its source, a Chi-square test of independence was conducted to examine the relationship between categorical variables. In addition, a set of regression analyses was conducted for the following characteristics of the photographs: 1) similarity between a photo and the article text in which it appears, 2) whether a person is included in a photo, 3) saturation, and 4) value. The following regression model was used for the analysis:

y = b_{0} + b \cdot p h o t o_s o u r c e_d u m m i e s + θ \cdot p u b l i s h e r_d u m m i e s + δ \cdot s e c t i o n_d u m m i e s + e

(1)

where y represents the photo characteristics mentioned in the previous sentence.

p h o t o_s o u r c e_d u m m i e s

is a set of dummy variables representing photo sources. Among the photo source categories, ‘staff photojournalist’ was used as the baseline group.

p u b l i s h e r_d u m m i e s

and

s e c t i o n_d u m m i e s

were included in the model as control variables.

p u b l i s h e r_d u m m i e s

represents dummy variables for publishers.

S e c t i o n_d u m m i e s

is a set of dummy variables representing the sections considered in this study—Politics, Economy, Society, and World—with Politics used as the baseline group.

When y represented whether a person was included in a photo, (1 if present, 0 otherwise), a binary logistic regression method was employed. A linear regression method was used for the remaining photo characteristics. All regression models were estimated using heteroscedasticity-robust standard errors.

Results

The number of photos by photo providers (sources)

To address the first research question (RQ1), this study examined the extent to which photographs from external sources were used in online news content. As detailed in the Methods section (Table 2), the largest proportion of published images originated from wire services, accounting for 7,540 photographs (59.4%), whereas citizen-generated photographs were the least represented, totalling only 50 images (0.4%). Photographs from social media and news sources comprised 13.9% (1,765) and 7.1% (907), respectively. In comparison, staff members produced 19.1% (2,429) of the photographs. In other words, only two out of 10 published photographs were created by staff photojournalists, whereas the remaining eight were provided by external sources.

Topics of photos

Next, we examined whether visual quality in terms of news value and visual features differed according to the type of source (RQ2a, b). First, regarding news value, this study examined the topics of photographs. The topic modelling analysis resulted in 61 topics, as reported in Appendix 1. The results of topic modeling did not directly indicate what the topics were; rather, they showed which keywords were most strongly associated with each topic. To examine whether the topics of photos varied with photo source, a Chi-square test of independence was conducted. In addition, because the contingency table contains cells with values less than or equal to five, the results of the Chi-square test may be unreliable; therefore, those cells were removed. The Chi-square test resulted in $χ^{2} (174, N = 5, 346)$ = 2264.39 (p < 0.05). These results indicate a statistically significant association between photo sources and topics. Furthermore, a Cramér’s V of 0.29 was obtained, indicating a moderate level of association between the two variables. The percentage of each topic covered by photo sources is shown in Figure 1.

Figure 1.

Percentage of each topic covered by photo sources.

The figure indicates that the most frequent topics of photos provided by wire services, staff photojournalists, and news sources are associated with Topic 0, which depicts a person or people delivering an address or speech. Another scene frequently covered by the photos from these sources was people sitting around a desk or table, represented by Topic 1. Another frequently appearing topic in photos taken by staff journalists is Topic 4, which represents people gathering for a protest. One of the frequently appearing topics in photos provided by news sources is buildings, which is Topic 13. The main topics of the photos provided by the wire services were similar to those of the photos taken by staff journalists. Many of the photos obtained from social media were in the form of collages. In other words, instead of a single photo, collages consisting of two or more photos were often acquired from social media. Photos acquired from social media frequently featured women. The main topics of the photos provided by the citizens were cars parked on roads and fire scenes.

Similarity between a photo and the text of the article that embeds the photo

To examine whether the similarity between a photo and the text of the article containing the photo varied according to the photo source, a linear regression analysis was conducted. The estimation results of equation (1), with the dependent variable representing the similarity between a photo and the text of the article that embeds the photo, are presented in Table 5. According to the table, when the source of the photos was a wire service, news source, or citizen, the similarity between the photos and article content was lower compared to photos taken by staff journalists.

Table 5.

The regression results of equation (1), DV: Similarity between a photo and the text.

Variables	Coefficient	SE	t	p	95% CI
Wire service	−0.840	0.106	−7.91	<0.001	−1.048	−0.631
News source	−1.817	0.150	−12.10	<0.001	−2.112	−1.523
Social media	0.063	0.134	0.47	0.641	−0.201	0.326
Citizen	−1.306	0.434	−3.01	0.003	−2.157	−0.455
$R^{2} = 0.363, F (11, 7639) = 388.78, p < 0.001$

N = 7,651, Basis value = Newspaper staff photographer.

*Due to space constraints, the coefficient results of the control variables (i.e., publisher and section variables) were omitted.

Dominant emotions

To examine the relationship between photo sources and dominant emotions embedded in photo faces, a Chi-square test was conducted:

χ^{2} (8, N = 4, 447)

= 237.09 (p < 0.05). The result indicates a statistically significant relationship between the types of photo source and emotion categories. In other words, the distribution of emotions is not independent of the photo source, suggesting that different image sources are associated with varying emotional expressions. We also obtained a Cramér’s V of 0.16, which indicates a small to moderate level of association between the types of photo source and emotion categories. The contingency tables of the photo sources and dominant emotion variables are presented in Table 6. The numbers in Table 6 indicate that photos acquired from social media and provided by news sources tended to show human faces with positive emotions, whereas photos provided by wire services were likely to show negative emotions.

Table 6.

The contingency table of photo sources and dominant emotions.

Emotion	Citizen	Social media	Staff photographer	News sources	Wire services	Total
Negative	0	97	114	10	198	419
Neutral	1	490	1073	310	921	2795
Positive	5	403	292	185	348	1233
Total	6	990	1479	505	1467	4447

Presence of humans

The estimation results of equation (1) with the dependent variable representing ‘whether a person(s) is present in a photo’ are reported in Table 7. According to Table 7, a person(s) was more likely to be present in a photo provided by wire services or acquired from social media than in one taken by staff photographers, whereas the opposite was true for photos supplied by news sources or citizens.

Table 7.

Regression results of equation (1), DV: Whether a person(s) is present in a photo.

Variable	Coefficient	O/R	SE	Z	p	95% CI
Wire service	0.263	1.301	0.100	2.640	0.008	0.068	0.459
News source	−0.438	0.645	0.118	−3.700	<0.001	−0.670	−0.206
Social media	0.387	1.473	0.117	3.300	0.001	0.157	0.617
Citizen	−0.994	0.370	0.343	−2.900	0.004	−1.666	−0.322
$χ^{2} (11) = 444.64, p < 0.001$

N = 6,674.

Saturation and value

The estimation results of equation (1) with the dependent variables of ‘saturation’ and ‘value’ are reported in Table 8. The left side of the table shows the results for the ‘saturation’ variable, while the right side shows the results for the ‘value’ variable. According to the results, the saturation did not differ greatly across with photo sources; however, photos provided by wire services or news sources tended to have higher saturation, while those from social media had lower saturation compared to those taken by staff journalists. The results on the right-hand side indicate that photos acquired from social media or news sources are likely to have a higher value than photos taken by staff journalists, indicating that photos by newspaper staff are darker than photos from social media or news sources.

Table 8.

Regression results of equation (1), DV: Saturation (left) and value (right).

Variables	DV: Saturation			DV: Value
Variables	Coef.	SE	p	Coef.	SE	p
Wire service	1.803	0.972	0.064	−1.297	1.031	0.208
News source	0.460	1.346	0.733	6.762	1.482	<0.001
Social media	−1.856	1.167	0.112	4.172	1.228	0.001
Citizen	−8.493	4.387	0.053	1.284	4.169	0.758
	$F (11, 7551) = 28.03,$ $p < 0.001$			$F (11, 7551) = 20.27,$ $p < 0.001$

N = 7,563.

Discussion

In a digital media environment marked by a decreasing number of staff photojournalists and an increasing reliance on external sources, this study analysed how this shift affects the quality of visual journalism and the gatekeeping role of professional photojournalists, particularly within online news media. Drawing on previous studies (Bednarek and Caple, 2012, 2017; Caple and Bednarek, 2015; Greenwood and Thomas, 2015; Klein-Avraham and Reich, 2016; Lough and McIntyre, 2019; Lough; Mortensen, 2023; Mortensen et al., 2023), this study examined the quality of news photographs and investigated how the characteristics varied according to the photo source. Five photo sources were considered: 1) staff photojournalists, 2) wire services, 3) social media, 4) news sources, and 5) citizens.

This study had several important findings. First, the results indicate that most lead photographs published online by major newspapers in South Korea were provided by external sources, including wire services, news sources, and citizens (RQ1). Additionally, the results of the Chi-square test and topic modelling analyses demonstrated that the topic of a photo differed depending on the type of photo source used (RQ2a). For example, staff journalists often capture political and social protests, while news sources typically provide images related to companies, often featuring their buildings. Notably, the current study also found that the several visual features varied depending on the type of photo source used (RQ2b). For example, photos acquired from social media and news sources tend to present human faces with positive emotions, whereas photos provided by wire services are likely to present negative emotions.

The findings of this study have important implications for visual journalism in the digital media environment. First, the results revealed that the work of staff photojournalists has been increasingly displaced by outsourced content. Only two out of 10 lead photographs published online were produced by staff photojournalists; the rest were outsourced. This contrasts with previous research showing that outsourced photographs accounted for a small proportion of published images (Greenwood and Thomas, 2015; Lough, 2018; Mast and Hanegreefs, 2015). The differing results of the current study may reflect the rapid transformations in visual journalism over the past decade.

Consistent with the findings of previous studies (Mortensen and Gade, 2018; Nilsson, 2017), the quality of visual journalism differed between professional staff and non-professionals. For example, regarding accurate representation, photos taken by staff journalists better represent article content than do photos acquired from other sources, such as wire services, news sources, or citizens. The findings of previous studies (e.g., Mortensen and Gade, 2018) suggest that an accurate representation of the corresponding article content in news photographs is a critical attribute of professional photojournalism. In addition, the current study showed that staff journalists frequently photographed scenes of political and social protests; Mortensen and Gade (2018) similarly demonstrated that professional journalists’ photographs are often more dynamic, representing conflict and tension, with a focus on capturing moments that involve multiple individuals, whereas non-professionals’ images tend to be static.

Regarding citizen journalism, consistent with the findings of previous studies (Brennen and Brennen, 2015), we found that citizen-generated photographs were the least represented, totalling only 50 images (0.4%). Photos provided by citizens more often captured streets, disasters, and fires, likely because such disasters are unpredictable, and the photos come from people who are in the area at the time of the event. Therefore, although some scholars have proposed that, in a digital technological environment where anyone can easily take photographs, visual journalism can function more effectively through the participation of citizens present at the scene of crises or conflicts (Allan, 2014; Allan and Peters, 2015), such citizen participation remains limited in Korea.

The findings reveal that the gatekeeping role of professional staff photojournalists is changing. As several scholars posit, visual journalists’ gatekeeping role has not diminished but rather transformed. It makes the professional responsibilities of verification and curation (Bakker, 2014) even more essential, as ‘part of that curation is determining the validity of images that emerge from news situations, and that is a huge challenge when dealing with amateur sources’ (Schwalbe et al., 2015: p. 476). However, extensive reliance on externally sourced photographs—including those provided by news sources—rather than images produced by staff photojournalists suggests a substantial weakening of journalism’s gatekeeping function, particularly when gatekeeping is conceptualized as the role of individual photojournalists or news gatherers (Bass, 1969).

Notably, the current study found that the emotions expressed through facial expressions varied depending on the type of photo source used: Photos acquired from social media and news sources tend to present human faces with positive emotions, whereas photos provided by wire services are likely to present negative emotions. This suggests that news sources provide photos that are favourable to them. From this perspective, the findings may reveal how the gatekeeping role of photojournalists is threatened, given that photojournalists’ legitimacy is based on their objective and impartial judgment of what to capture, and their goal of focusing on accurate representations of reality (Brennen, 2010; Morten and Gade, 2018; Wheeler and Gleason, 1995). For example, politicians’ photos are often provided by the political party or institution to which a politician belongs, and, in such cases, positive-emotion photos may be provided to generate a favourable attitude toward the politician.

This supports prior research (e.g., Greenwood and Thomson, 2020), which suggests that how a subject is portrayed in a photo is strongly influenced by photographers’ personal views or the political orientation of their organization. Ferrucci et al. (2020) emphasized this distinction, stating, ‘Professional photojournalists are truly independent. There’s a lot of people working in corporate communication these days as photographers on behalf of that organization, but the work they do isn’t necessarily photojournalism in the traditional sense’. (p. 379). This observation highlights independence as a defining characteristic of professional photojournalism.

This study had several limitations. First, data were acquired from photos and news articles posted on Naver, an online news aggregator in South Korea. Thus, the results of this study may not be generalizable to other countries where the methods of reporting or obtaining photos of news articles differ from those of South Korea. Furthermore, although the criteria for evaluating the quality of visual journalism vary widely, this study focused on elements that could be analysed using computational methods. For example, regarding visual features, each photo’s dynamic characteristics, (active or static; candid or posed) (Greenwood and Thomas, 2015; Lough and Mortensen, 2023), lens distance (or eye contact), and photographic angle (Greenwood and Thomas, 2015; Mortensen and Gade, 2018) were not measured.

While this approach necessarily limited the scope of the assessment, it also enabled the analysis of a large number of photographs, which represents a notable strength of the study. In particular, this study employed a multimodal deep learning model to examine aspects overlooked in previous research, such as the relationship between text and images, and topic modeling analyses of images. Future research would benefit from employing more advanced computational techniques to evaluate photographs across a broader set of quality criteria. In addition, as this study used only photos related to news articles that were published within a particular time period, in many cases, related to important events and issues in 2024.

Despite its limitations, this study contributes to journalism research by elucidating the extent to which photojournalists’ gatekeeping roles have weakened in the digital news environment and how various changes have affected the quality of visual journalism.

Footnotes

ORCID iDs

Na Yeon Lee

Sang Yup Lee

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the Yonsei University Humanities and Social Sciences Field Creative Research Fund of 2025-22-0512.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Notes

Author biographies

Na Yeon Lee (Ph.D., University of Texas at Austin) is a Professor in the Department of Communication at Yonsei University in South Korea. Her research interests include journalism, political communication, and media psychology. Her work has been published in Journalism, Journalism & Mass Communication Quarterly, Information, Communication & Society, Journalism Practice, International Journal of Public Opinion Research, Journal of Broadcasting and Electronic Media, Journal of Health Communication, Computers in Human Behavior, The Social Science Journal, and Asian Journal of Communication, among others.

Whan Choi is a master student at the Department of Digital Analytics, Yonsei University in South Korea. His research interests include data science, computational social science, and time series.

GwangGyu Park is an undergraduate student at the Department of Communication, Yonsei University in South Korea. His research interests include computational social science, computer vision, and multimodal modeling.

Sang Yup Lee is a Professor in the Department of Communication at Yonsei University in South Korea. He received his Ph.D. and Master’s degree from the Media Information Department at Michigan State University. He studied Computer Science for his Bachelor’s degree at Yonsei University. His research interests include text mining, large language models, computational social science, and data science.

Appendix

Appendix 1. Topic modeling results

Topic_0	Topic_1	Topic_2	Topic_3	Topic_4	Topic_5	Topic_6	Topic_7	Topic_8	Topic_9
Podium	Sitting	table	Photograph	Signs	Hair	Car	Korean	Scene	Interviewed
Microphone	Desk	Seated	Collage	People	Woman	Road	Sign	Fire	Crowd
Speaking	table	Group	Two	Gathered	Long	Parked	Writing	Destruction	People
Addressing	Shaking	Men	Each	Crowd	Her	Vehicle	Building	Damaged	Cameras
Audience	Hands	Meeting	Showing	Holding	Dark	Street	Text	Debris	Suit

Topic_10	Topic_11	Topic_12	Topic_13	Topic_14	Topic_15	Topic_16	Topic_17	Topic_18	Topic_19
Cityscape	Military	Up	Building	Airplane	Posing	Trump	Gesturing	Walking	Posing
Buildings	Soldiers	Close	Story	Airport	Moment	Donald	Hand	Down	Selfie
Rise	Uniforms	Hair	Sky	Aircraft	Tender	President	Hands	Past	Young
Urban	Camouflage	Portrait	Multi	Terminal	Family	Former	Raised	Umbrella	Woman
Sprawling	Soldier	Short	Set	Descending	Bride	Biden	Suit	Street	Photograph

Topic_20	Topic_21	Topic_22	Topic_23	Topic_24	Topic_25	Topic_26	Topic_27	Topic_28	Topic_29
Hospital	Plastic	Police	Serene	Screenshot	Soccer	Stock	Industrial	Room	Ship
Medical	Food	Officers	Landscape	Video	Player	Market	Complex	Living	Boat
Bed	Filled	Bridge	Water	Korean	Field	Displaying	Situated	Bedroom	Ocean
Corridor	Bowl	Restrained	Beach	Featuring	Players	Screen	Aerial	Modern	Water
Protective	Dishes	Scene	Scene	Text	Jersey	Data	Site	Minimalist	Docked

Topic_30	Topic_31	Topic_32	Topic_33	Topic_34	Topic_35	Topic_36	Topic_37	Topic_38	Topic_39
Medal	Memorial	Parked	Panda	Wall	Store	Desks	Baseball	Hand	Phone
Proudly	Somber	Kia	Giant	Plates	Grocery	Chairs	Player	Card	Smartphone
Olympic	Funeral	Car	Bamboo	Framed	Shopping	Rows	Action	Holding	Screen
Medals	Service	Showroom	Bear	Front	Stocked	Room	Bat	Reaching	Holding
Ring	Poignant	Suv	Enclosure	Artwork	Products	Filled	Swinging	Object	Cell

Topic_40	Topic_41	Topic_42	Topic_43	Topic_44	Topic_45	Topic_46	Topic_47	Topic_48	Topic_49
Bustling	Surface	Train	Conversation	Trees	Kitchen	Warehouse	Lying	Stage	Holding
Market	View	Tunnel	Engaged	Building	Cooking	Personnel	Their	Sashes	Paper
Diverse	Object	Subway	Suits	Standing	Food	Boxes	Bed	Arms	Suit
Array	Metal	Station	Men	Background	Preparation	Military	Person	Individuals	Device
Display	Close	Platform	Two	Like	Commercial	Facility	Deceased	Flowers	Piece

Topic_50	Topic_51	Topic_52	Topic_53	Topic_54	Topic_55	Topic_56	Topic_57	Topic_58	Topic_59
Poster	Stacks	Bottle	Stadium	Blue	Gazing	Golf	Sky	Crown	Dog
Large	Bills	Drinking	Arena	Logo	Window	Course	Nighttime	Sash	Terrier
Chart	Money	Savoring	People	Front	Out	Golfer	Streaks	Pageant	Jack
Podium	Bitcoin	From	Crowd	Text	Scenic	Clubs	Dark	Catherine	Russell
Suit	Stack	Wine	Seats	Backdrop	Cityscape	Lush	Streaking	Beauty	Dogs

Topic_60
Hallway
Blurry
Pixelated
Security
Highlighting

References

Allan

(2013) Citizen Witnessing: Revisiting Journalism in Times of Crisis. Polity press.

Allan

(2014) Witnessing in crisis: photo-Reportage of terror attacks in Boston and London. Media, War & Conflict 7(2): 133–151. https://doi.org/10.1177/1750635214531110

Allan

Peters

(2015) Visual truths of citizen reportage: four research problematics. Information, Communication & Society 18(11): 1348–1361. https://doi.org/10.1080/1369118x.2015.1061576

Bakker

(2014) Mr. Gates returns: curation, community management and other new roles for journalists. Journalism Studies 15(5): 596–606. https://doi.org/10.1080/1461670x.2014.901783

Bass

(1969) Refining the “gatekeeper” concept: a UN radio case study. Journalism Quarterly 46(1): 69–72. https://doi.org/10.1177/107769906904600110

Bednarek

Caple

(2012) ‘Value added’: language, image and news values. Discourse, Context & Media 1(2-3): 103–113. https://doi.org/10.1016/j.dcm.2012.05.006

Bednarek

Caple

(2017) The Discourse of News Values: How News Organizations Create Newsworthiness. Oxford University Press.

Bissell

(2000) A return to ‘mr. Gates’: photography and objectivity. Newspaper Research Journal 21(3): 81–93. https://doi.org/10.1177/073953290002100307

Brennen

(2010) Photojournalism: historical dimensions to contemporary debates. In: Allen

(ed) The Routledge Companion to News and Journalism. Routledge, pp. 71–82.

10.

Brennen

(2015) Taking our pictures: citizen photojournalism in traditional US news media. Journalism Practice 9(4): 520–535. https://doi.org/10.1080/17512786.2015.1030138

11.

Caple

Bednarek

(2015) Rethinking news values: what a discursive approach can tell Us about the construction of news discourse and news photography. Journalism 17(4): 435–455. https://doi.org/10.1177/1464884914568078

12.

Carlson

Robinson

Lewis

(2021) Digital press criticism: the symbolic dimensions of Donald Trump’s assault on US journalists as the “enemy of the people”. Digital Journalism 9(6): 737–754. https://doi.org/10.1080/21670811.2020.1836981

13.

Donsbach

(2004) Psychology of news decisions: factors behind journalists’ professional behavior. Journalism 5(2): 131–157. https://doi.org/10.1177/146488490452002

14.

Ferrucci

Taylor

Alaimo

(2020) On the boundaries: professional photojournalists navigating identity in an age of technological democratization. Digital Journalism 8(3): 367–385. https://doi.org/10.1080/21670811.2020.1732830

15.

Fry

(2013) Boston explosions a reminder of how breaking news reporting is changing. Poynterorg 16: April.

16.

Galtung

Ruge

(1965) The structure of foreign news: the presentation of the Congo, Cuba and Cyprus crises in four Norwegian newspapers. Journal of Peace Research 2(1): 64–90.

17.

Gans

(1979) The messages behind the news. Columbia Journalism Review 17(5): 40–46.

18.

Greenwood

Thomas

(2015) Locating the journalism in citizen photojournalism: the use and content of citizen-generated imagery. Digital Journalism 3(4): 615–633. https://doi.org/10.1080/21670811.2015.1034528

19.

Greenwood

Thomson

(2020) Framing the migration: a study of news photographs showing people fleeing war and persecution. International Communication Gazette 82(2): 140–163. https://doi.org/10.1177/1748048519833515

20.

Grootendorst

(2022) BERTopic: neural topic modeling with a class-based TF-IDF procedure. arXiv preprint arXiv:2203.05794.

21.

Harcup

O’Neill

(2001) What is news? Galtung and ruge revisited. Journalism Studies 2(2): 261–280. https://doi.org/10.1080/14616700118449

22.

Harrison

(2010) News media. In: Albertazzi

Cobley

(eds) The Media: An Introduction. 3rd edition. Longman, pp. 246–257.

23.

Kim

Smith

(2005) Sixty years of showing the world to America: pulitzer prize-winning photographs, 1942-2002. Gazette 67(4): 307–323. https://doi.org/10.1177/0016549205054280

24.

Kind

(2017) Imaginative vividness. Journal of the American Philosophical Association 3(1): 32–50. https://doi.org/10.1017/apa.2017.10

25.

Klein-Avraham

Reich

(2016) Out of the frame: a longitudinal perspective on digitization and professional photojournalism. New Media & Society 18(3): 429–446. https://doi.org/10.1177/1461444814545289

26.

Lewin

(1947) Group decision and social change. Readings in Social Psychology 3(1): 197–211.

27.

Lewis

Liu

Goyal

, et al. (2019). Bart: denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension. arXiv preprint arXiv:1910.13461.

28.

(2025) Research and analysis of facial recognition based on FaceNet, DeepFace, and OpenFace. ITM Web of Conferences 70: 03009. EDP Sciences. https://doi.org/10.1051/itmconf/20257003009

29.

Lough

(2018) Intermedia visual agenda setting: comparing wire service top photo distribution to what makes the front page. The Agenda Setting Journal 2(1): 25–40. https://doi.org/10.1075/asj.17002.lou

30.

Lough

McIntyre

(2019) Visualizing the solution: an analysis of the images that accompany solutions-oriented news stories. Journalism 20(4): 583–599. https://doi.org/10.1177/1464884918770553

31.

Lough

Mortensen

(2023) Routine and individual-level influences on newspaper front-page images: wire photographs, staff photojournalism, race and gender. Journalism Practice 17(9): 2015–2034. https://doi.org/10.1080/17512786.2022.2033635

32.

Mast

Hanegreefs

(2015) When news media turn to citizen-generated images of war: transparency and graphicness in the visual coverage of the Syrian conflict. Digital Journalism 3(4): 594–614. https://doi.org/10.1080/21670811.2015.1034527

33.

Meta (2024) Llama 3.2: Revolutionizing Edge AI and Vision with Open, Customizable Models. Meta. Retrieved January, 10 from. https://ai.meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/

34.

Mortensen

Gade

(2018) Does photojournalism matter? News image content and presentation in the middletown (NY) times herald-record before and after layoffs of the photojournalism staff. Journalism & Mass Communication Quarterly 95(4): 990–1010. https://doi.org/10.1177/1077699018760771

35.

Mortensen

McDermott

Ejaz

(2023) Measuring photo credibility in journalistic contexts: scale development and application to staff and stock photography. Journalism Practice 17(6): 1158–1177. https://doi.org/10.1080/17512786.2021.1976073

36.

Newman

Fletcher

Schulz

, et al. (2021) Digital News Report 2021. Reuters Institute for the Study of Journalism.

37.

Nilsson

(2017) A faster kind of photojournalism? image-Selection processes in a Swedish newsroom. Nordicom Review 38(Supplement 2): 41–56. https://doi.org/10.1515/nor-2017-0413

38.

Schwalbe

Silcock

Candello

(2015) Gatecheckers at the visual news stream: a new model for classic gatekeeping theory. Journalism Practice 9(4): 465–483. https://doi.org/10.1080/17512786.2015.1030133

39.

Serengil

(2024) Deepface. Retrieved December 2, 2024 from. https://github.com/serengil/deepface

40.

Shoemaker

Vos

(2009) Gatekeeping Theory. Routledge.

41.

Shoemaker

Vos

Reese

(2009) Journalists as gatekeepers. In: The Handbook of Journalism Studies. Routledge, pp. 93–107.

42.

Singer

(2006) Stepping back from the gate: online newspaper editors and the co-production of content in campaign 2004. Journalism & Mass Communication Quarterly 83(2): 265–280. https://doi.org/10.1177/107769900608300203

43.

Thomson

(2018) Freelance photojournalists and photo editors: learning and adapting in a (mostly faceless) virtual world. Journalism Studies 19(6): 803–823. https://doi.org/10.1080/1461670x.2016.1215851

44.

Ultralytics (2024) Ultralytics YOLO11. Ultralytics, Retrieved January 10, 2025 from. https://docs.ultralytics.com/ko/models/yolo11/

45.

Wheeler

Gleason

(1995) Photography or photofiction: an ethical protocol for the digital age. Visual Communication Quarterly 2(1): 8–12. https://doi.org/10.1080/15551393.1995.10387515

46.

White

(1950) The “gate keeper”: a case study in the selection of news. Journalism Quarterly 27(4): 383–390. https://doi.org/10.1177/107769905002700403