Abstract
The rapid expansion of mobile applications featuring personalized recommendation algorithms and infinite scrolling news feeds has raised concerns about their role in shaping societal knowledge acquisition. In digital environments, friction refers to the deliberate or perceived impediments that disrupt seamless content flow, offering users opportunities to reflect on and regulate their engagement. Grounded in the frameworks of algorithm dependence, this study investigates how different types of algorithmic Apps (news, social media, and short video) impact users’ news knowledge, with a focus on the mediating effect of algorithm dependence in this process. Specifically, perceived information narrowing is introduced as an operationalization of user friction that moderates the link between algorithm App usage and dependence. Data was collected via online survey with 354 responded participants. Results revealed that short video Apps decrease users’ news knowledge, social media Apps indirectly reduce news knowledge through algorithm dependence, and news Apps diminish news knowledge only among users perceiving high levels of information narrowing.
These findings suggest the potential for introducing user friction to regulate algorithmic curation and mitigate its negative impact on knowledge gain, especially within algorithmic news Apps. This study contributes to understanding the complex interplay between algorithmic dependence and knowledge gain, highlighting user-centered approaches to enhancing informational diversity in algorithm-driven media.
Keywords
The widespread proliferation of algorithmic recommendation applications (Apps) featuring infinite scrolling news feeds has raised significant concerns among both the public and scholars. From the perspective of techno-social ecology, these applications are no longer simple information tools; instead, they have gradually embedded themselves into users’ daily lives, forming a complex ecosystem (Couldry & Mejias, 2019). Toutiao (Today's Headlines) emerged as one of the earliest and most watched personalized news recommendation Apps with the slogan “what you care about is the headline” in 2012. Characterized by a lack of editorial staff, absence of content production, and no explicit stance or values, its operational core is a set of algorithms built from the code. Following Toutiao's lead, similar algorithmic recommendation news Apps, such as Yidian (One-point Information) and Kuaibao (Daily Express), have surfaced. Short video Apps like Douyin (the Chinese version of TikTok, launched in 2016) utilize a funnel mechanism to push videos, sharing the decentralized recommendation algorithm principle with Toutiao. The Weibo microblog platform is based on recommendation algorithms for social networking. Its recommendations are tailored according to users’ historical data and preferences of similar user groups, promoting content, accounts, or intelligently sorting information. Techno-social ecology theory emphasizes that algorithmic recommendation technology, as an embedded medium, interacts with social structures and individual behaviors, profoundly shaping the ways users’ access and process information. In this ecosystem, news Apps, social media Apps, and short video Apps serve as the primary information channels, each fulfilling distinct functions of information access, social interaction, and entertainment content consumption. For this reason, these three types of algorithmic recommendation applications together form the core channels for information acquisition among users and have thus been selected as the primary media types in this study.
Despite their role as important channels for information and knowledge acquisition in society (Delli Carpini, 2000), algorithm-driven applications have faced increasing scrutiny for the negative effects of their recommendation systems. Terms such as filter bubbles (Pariser, 2011), information cocoons (Sunstein, 2006), and echo chambers (Sunstein, 2002) describe the process by which people selectively obtain information from media, leading to a narrowing of information diversity. The assumption that users can fully engage with information—whether through active seeking or incidental news exposure—and thereby accumulate knowledge is increasingly questioned with the rise of algorithmic recommendation technologies. Users are not truly in a high-choice media environment, as proposed by Van Aelst et al. (2017), because algorithmic recommendations filter out a wealth of information, resulting in users receiving and accumulating limited and biased knowledge. People's Daily's (2018) critique of information cocoons has heightened public awareness of algorithms and sparked intense debate among experts and scholars in China.
Building on the foundation of media dependency theory (Ball-Rokeach & DeFleur, 1976), this study proposes the concept of algorithm dependency. Media dependency theory posits that users’ reliance on specific media is determined by its ability to meet their informational needs, a dependency that becomes especially pronounced in today's complex information environment. In the context of algorithmic recommendations, this dependency evolves into algorithm dependency, wherein users’ reliance on applications is reinforced not only by the medium itself but also by the intensifying effects of personalized recommendations.
These algorithm-driven applications share a distinctive feature: the infinite scroll interface. This design element enables users to continuously scroll through content without any clear indication of an endpoint, making it easier to keep scrolling than to stop. By eliminating natural pause cues for the brain, it becomes more challenging for users to disengage, while algorithmic personalized recommendations further reduce the cognitive effort required for decision-making (Abdollahpouri et al., 2021; Natarajan, 2024). The infinite scrolling interface, combined with algorithms and simplified interaction gestures, has created what is often referred to as a “frictionless” digital environment that minimizes operational barriers, immersing users in content consumption and potentially fostering a dependency on algorithmic recommendations, which in turn influences their news knowledge acquisition and information processing.
The term “friction” originally referred to physical resistance, and it has gained analytical traction in media and platform studies as a way to understand the tension between seamless design and user agency (Popiel & Vasudevan, 2024; Tomalin, 2023). In the context of online interactions, “friction” is used to refer to any unnecessary retardation of a process or activity (e.g., a financial transaction, the uploading of a picture and an unsolicited pop-up window), and such things can frustrate and annoy users (Tomalin, 2023). Natale and Treré (2024) note that friction is not only inevitable but also constructive, as it allows users to “stand firm” and avoid being fully absorbed by media. Drawing on this literature, user friction in our framework arises from user awareness and can serve as a self-regulatory resource.
Through this theoretical framework, the study aims to reveal how “frictionless” algorithmic recommendation Apps foster usage dependency, thereby shaping users’ processes of news knowledge acquisition. It should be clarified that news knowledge in this study refers to an individual's awareness of current social issues that have been prominently reported in mainstream media. It is measured by evaluating participants’ accuracy in identifying true or false statements about popular news topics.
Literature review
Algorithm dependence: an extension of Media system dependency theory
Media system dependence (MSD) theory, introduced by Ball-Rokeach and DeFleur in 1976, provides a foundational framework for understanding the nuanced relationship between individuals and media, emphasizing how audiences develop dependencies on media to fulfill specific needs or achieve goals (Ball-Rokeach & DeFleur, 1976; Wright, 1986). This theory explains the dynamics of dependency in the context of both traditional and new media. Dependence on certain media or technology has been defined as the psychological and behavioral reliance that individuals develop in their daily lives, often manifesting as excessive use or addiction (Huang et al., 2024; Zhang et al., 2025). In the MSD framework, media usage has emerged as a significant predictor of media dependence. Empirical research has confirmed a positive correlation between media usage and dependence, across various media platforms and demographic groups (Aldamen, 2023; Kim & Jung, 2017; Loges & Ball-Rokeach, 1993; Lowrey, 2004; Ng, 2022; Xie et al., 2023).
In the context of algorithmic media, dependency patterns evolve further, giving rise to what can be termed algorithmic dependence. Schaetz et al. (2025) define algorithm dependency as the degree to which individuals rely on algorithms in platformized media use to meet their information needs, even when they are aware of associated risks such as loss of control over personal data. This conceptualization emphasizes both the structural nature of dependency and the inherent tension between the convenience of algorithmic curation and users’ concerns about its consequences. Algorithmic feeds with infinite scrolling and personalized recommendations have been likened to a “pleasure machine”, deliberately designed to progressively satisfy user needs in order to manipulate users, identify vulnerabilities, cultivate compulsive habits, and exploit user attention (Bhargava & Velasquez, 2021; Helm & Matzner, 2023; Matzner, 2024). This argument has been discussed in the context of psychological hedonism (Gal, 2017; Reviglio & Agosti, 2020). Neuroscientific research has revealed that TikTok's personalized video recommendations in contrast to non-personalized ones activate brain regions associated with self-referential processing and reward systems. This finding underpins how recommended algorithms are able to keep the user's attention to suggested contents (Su et al., 2021). Guess et al.’ (2023) study also revealed that moving users out of algorithmic feeds to the reverse-chronologically-ordered feeds substantially decreased the time they spent on the platforms and their activity. Similarly, Rixen et al. (2023) identified external and internal loops within algorithmic feeds: frequent App openings (external loop) and continuous scrolling (internal loop), where users enter for a specific purpose but end up browsing unconsciously.
Based on this understanding, we propose the following hypotheses to investigate the relationship between the use of different types of algorithmic Apps and algorithmic dependence: H1a: Use of algorithmic news Apps is positively related to algorithm dependence. H1b: Use of algorithmic social media Apps is positively related to algorithm dependence. H1c: Use of algorithmic short video Apps is positively related to algorithm dependence.
Impact of different types of algorithmic apps on news knowledge
Algorithmic news apps
Algorithmic news Apps, which tend to provide users with full news reports and articles, similar with traditional forms of news consumption like newspapers, television news, and online news sites, have been found to increase knowledge, especially in political settings (Chaffee & Kanihan, 1997; Dimitrova et al., 2014). Studies have shown a positive relationship between the usage frequency of online websites that feature full-length news articles and factual knowledge (Andersen et al., 2016; Kaufhold et al., 2010). Although the Beam (2014) study showed a direct negative effect of personalized news recommender system use on knowledge, an indirect positive effect mediated by elaboration also exists when only recommended headlines are displayed. Eveland's (2001) cognitive mediation model highlights the role of elaboration in facilitating learning from news, providing a theoretical basis for understanding how different modes of news presentation may influence knowledge acquisition. In digital news environments, Feezell (2018) provided empirical evidence in digital news environments, finding that algorithmically curated exposure can enhance users’ issue salience, particularly among those with lower political interest. Algorithmic news Apps offer comprehensive news coverage rather than just headlines or snippets. So they may encourage elaborative processing and retention of information compared to more fragmented exposure formats, though direct evidence on knowledge acquisition remains mixed.
Algorithmic social media apps
By contrast, algorithmic social media apps often provide snack news, requiring users to click on links to access full articles. This mode of news consumption has not shown the same positive effects on political knowledge as seen with full-length articles (Barberá et al., 2015; Dimitrova et al., 2014; Shehata & Stromback, 2018). Evidence suggests that while social media can increase incidental exposure to news, leading to some positive effects on information recall and recognition, these effects are contingent on users engaging with full-length stories (Lee & Kim, 2017). Su et al. (2024) found that the News-Finds-Me perception did not directly influence climate change knowledge, but had an indirect effect through users’ evaluations of algorithmic news content. Only social media use for news rather than general social media use was found to be positively related to political knowledge, and news elaboration is a key mediator (Park & Kaye, 2019).
However, such conditions are not typical of everyday social media use. Social media platforms are predominantly used for entertainment and relationship-oriented purposes (Bright, 2016), which encourages low-effort, fragmented, and passive exposure to news content rather than sustained attention to full-length articles. Furthermore, social media's partial control environment, characterized by personalized content that may reinforce confirmation biases (Kim et al., 2013), promotes selective exposure to like-minded information (Pariser, 2011), and limits exposure to diverse viewpoints, has been linked to lower levels of political knowledge (Cacciatore et al., 2018; Shehata & Strömbäck, 2018; Stroud, 2011). As was criticized, social media is structured to encourage “agnotology”, the cultural creation of ignorance (John & Nissenbaum, 2019).
Overall, although social media may provide opportunities for incidental exposure, the predominance of low-elaboration use patterns and selective exposure suggests that these negative mechanisms are more likely to dominate on average, thereby hindering substantive political learning.
Algorithmic short video apps
Algorithmic short video apps present a unique case for examining the impact on users’ news knowledge, especially when compared to algorithmic news and social media apps. Short video apps, known for their heavily customized and entertainment-oriented content, align closely with the concept of snack news, which provides a brief overview of news topics through headlines, short teasers, and images (Schäfer et al., 2017). The format of short videos stands in stark contrast to the more comprehensive news coverage provided by algorithmic news apps, and the mixed approach of algorithmic social media apps that may offer both in-depth articles and snackable news. Engagement with short video Apps typically involves a low level of elaboration, as the content is designed for fast consumption, aiming to give users a superficial overview of current events rather than fostering a deep understanding (Dimitrova et al., 2014).
Research indicates that reliance on snack news, prevalent on social media and short video platforms, is either unrelated or potentially harmful to the recall of political facts (Cacciatore et al., 2018; Dimitrova et al., 2014; Shehata & Strömbäck, 2018). While these findings suggest variability across contexts, the structural characteristics of short video platforms make negative effects more likely to emerge in practice. First, the rapid and continuous flow of short video content leaves little room for elaborative processing, which is essential for knowledge acquisition. Second, the fragmented and decontextualized presentation of information hinders the integration of discrete pieces of information into coherent knowledge structures. Third, repeated exposure to similar topics through algorithmic recommendation may create an illusion of knowledge, whereby users feel informed without actually improving factual understanding (Schäfer, 2020). In sum, these features suggest that short video Apps are more likely to impede, rather than support, the acquisition and retention of news knowledge.
Given these considerations, the following hypotheses are proposed: H2a: The use of algorithmic news Apps is positively related to news knowledge. H2b: The use of algorithmic social media Apps is negatively related to news knowledge. H2c: Use of algorithmic short video Apps is negatively related to news knowledge.
Algorithm dependence and news knowledge
Early research based on newspaper and television media has shown from different perspectives how media dependency influences information acquisition, processing, and social behaviors. Gaziano (1990) finds that lower-income and less educated groups rely more on traditional media for news, which may limit their knowledge levels and social participation. Moy et al. (2005) emphasize that media dependence is also linked to trust and civic engagement, noting that highly dependent individuals may be more active in public affairs due to high trust in the media, but this could also weaken their critical thinking. These studies suggest that the relationship between media dependence and knowledge acquisition is quite complex. It is also underexplored the conditions under which media dependence may have positive or negative effects.
With the shift to algorithm-driven media, dependency patterns evolve, and algorithmic applications highlight the complex interplay between users’ reliance on these systems and the breadth of their news knowledge. While such systems may prioritize engagement over informational value, potentially leading to a filter bubble effect, which is defined as a cluster of information that has been filtered and selected based on predictions of who a user is and what they will do next (Eg et al., 2023). In the filter bubbles, users are exposed to the narrowing and fragmentation of news sources and perspectives, thereby negatively impacting their overall news knowledge (Geiß et al., 2021; Sunstein, 2007; Swart, 2021; Thorson & Wells, 2016).
Moreover, once users develop algorithmic dependency, characterized by internal and external loops, they are prone to entering a state of mindless scrolling (Baym et al., 2020; de Segovia Vicente et al., 2024). As the technological processes of liking, commenting, and retweeting have been mechanistically integrated into buttons (Burgess & Baym, 2020), these actions have become increasingly habitual, blending voluntary and involuntary behaviors, both conscious and unconscious (Karppi, 2018). Passive scanning of incidentally encountered information is defined as first-level incidental news exposure, while intentional processing of incidentally encountered content appraised as relevant is defined as second-level incidental news exposure (Matthes et al., 2020). It was assumed that second-level incidental news exposure has stronger effects on knowledge acquisition than first-level incidental news exposure. However, within algorithm-driven information streams dominated by snack news, cognitive processing frequently remains at the lower first-level, making it difficult for users to transition to deeper processing when encountering more substantial content. This results in missed opportunities for meaningful engagement and hinders effective knowledge acquisition. Research has further revealed that user engagement with personalized news is low, with many participants admitting to superficial and incomplete reading of news stories (Oeldorf-Hirsch & Srinivasan, 2022).
Based on this body of research, the following hypothesis is proposed: H3: Algorithm dependence is negatively related to news knowledge.
Perceived information narrowing as user friction: moderating the link between algorithmic app use and dependence
Researchers generally agree that transferring control from platforms to users is a key strategy for mitigating the potentially problematic consequences of algorithms. It has been argued that operators have a responsibility to provide design solutions that more effectively raise user awareness and autonomy. Concepts such as algorithmic experience, which advocates transparent labelling of personalized content, and algorithmic sovereignty, which promotes user control over algorithmic processes, highlight the need for user empowerment in navigating algorithm-driven environments (Alvarado & Waern, 2018; Reviglio & Agosti, 2020). Mattis et al. (2024) advocate for personalizing diversity nudges within algorithmic recommender systems to enhance the diversity of users’ news consumption.
Shifting from platform responsibility to user perspectives, studies have indicated the positive influence of users’ algorithm awareness and proactive control on news feed curation. For example, algorithmic awareness has been found to lead to more active engagement with Facebook, bolstered overall feelings of control on the site (Eslami et al., 2015). Furthermore, users who actively personalize their feeds tend to consume more diverse content (Merten, 2021), and awareness of algorithmic curation's impact on content diversity influences trust and reliance on these platforms (Shin, 2021; Wölker & Powell, 2021). Rather than advocating for the complete abandonment of algorithmic Apps to avoid the negative impacts, researchers emphasize terms like mindful or conscious scrolling, encouraging individuals to observe and learn from their experiences, allowing them to create personal guidelines for their technology use (Baym et al., 2020; Rauch, 2018). With this awareness, individuals can create opportunities to break habits and change behaviors (Levy, 2016), thus, potentially improving their cognitive elaboration levels and gaining knowledge.
In the context of infinite-scrolling algorithmic feeds, we operationalize user friction through perceived information narrowing. This choice is theoretically grounded: perceived information narrowing captures users’ recognition that algorithmic personalization has begun to limit the diversity of their news exposure. This awareness can function as a cognitive “speed bump” in otherwise frictionless scrolling, prompting users to reconsider their reliance on the platform or seek more varied content. Perceived information narrowing highlights the consequences of algorithmic personalization and offers a more nuanced reflection of mindful engagement with algorithmic Apps. More importantly, perceived information narrowing could serve as a rapid feedback mechanism from users to the algorithm, aiding in the adjustment of content diversity within the news feed.
Thus, this study examines perceived information narrowing as a form of user-empowered friction that could potentially mitigate excessive dependence on algorithmic Apps. The following research questions are proposed: RQ: How does perceived information narrowing moderate the relationship between (1) algorithmic news App use, (2) algorithmic social media App use, (3) algorithmic short video App use and algorithm dependence?
Current study
This study seeks to comparatively analyze the influence of different categories of algorithmic Apps—specifically news, social media, and short video Apps—on the public's acquisition of news knowledge. Grounded in the hedonistic critique of infinite scrolling within algorithmic feeds, which posits that such designs exploit users’ attention and foster dependency, this research investigates the mediating role of algorithm dependence in the relationship between App usage and knowledge acquisition. Furthermore, the study introduces the concept of user friction as a mechanism in the development of algorithm dependence, with perceived information narrowing functioning as a moderating factor. This framework highlights the importance of user mindfulness during App engagement as a proactive strategy for regulating algorithm-curated content. Theoretically, this study aims to extend algorithm dependence theory by revealing how user awareness of content narrowing can serve as a self-regulatory mechanism. Practically, it aims to introduce user friction to the algorithm platforms as a potential mechanism to adjust content diversity. Figure 1 illustrates the conceptual framework of this study.

Research conceptual framework.
Methods
Data collection
In this study, snowball sampling was used, and questionnaires were distributed via the online survey platform wjx.cn in October 2019. In total, 354 participants responded to the survey. Participation in this online survey was voluntary and the survey responses were anonymous. The participants comprised 33.6% males and 66.4% females, 40.1% aged between 18 and 25 years, 44.8% aged between 26 and 40 years, 10.5% aged between 41 and 50 years, and 4.6% aged above 51 years. The majority had college or undergraduate education (55.1%), followed by postgraduate (39.5%), and high school or less (5.4%). Their daily time spent on Sina Microblog was 2.87 (SD = 1.38), time spent on TikTok was 2.19 (SD = 1.33), and time spent on TouTiao was 2.39 (SD = 1.30) on a 6-point scale shown in Table 1.
Characteristics of the Participants.
Measurements
Use of algorithmic apps
The use of three types of Apps featuring algorithmic recommendations was measured: news Apps (e.g., Toutiao and Yidian), social media Apps (e.g., Sina Weibo), and short video Apps (e.g., TikTok and Kuaishou). Participants were asked to report their time spent on these three Apps per day (1 = never, 2 = less than 30 min, 3 = 30–60 min, 4 = 1–2 h, 5 = 2–3 h, 6 = more than 3 h).
News knowledge
Referring to Beam (2014) as well as Lee and Kim (2016), news knowledge was measured by five true or false statements concerning social issues prominently reported in mainstream media at the time of survey implementation. Participants were asked to rate whether each statement was true, false, or uncertain. The statements included “Trump and Kim Jong-un met in Vietnam in February,” “Bingbing Fan was fined more than 800 million yuan for tax evasion,” and “Betel nut was a first-class carcinogen.” The number of correct answers was recorded to generate a score from 0 to 5, with higher scores indicating greater knowledge of the most popular news online at the time.
Algorithm dependence
The four-item scale measuring dependence on Apps with algorithmic recommendations was adapted from the scale for Facebook persistence and overuse (Orosz et al., 2016). This scale contains two items on behavioral and affective persistence (i.e., “I often search for Internet connections to visit these algorithmic Apps,” “I feel bad if I don’t check these algorithmic Apps daily”) and two items on excessive use (i.e., “I spent time on these algorithmic Apps at the expense of my obligations,” and “I used these algorithmic Apps instead of sleeping”). Participants evaluated these items on a five-point Likert scale (1 = completely disagree, 3 = neutral, 5 = completely agree). Cronbach's alpha was .79.
Perceived information narrowing
Four items were adapted from the algorithmic media content awareness scale (Zarouali et al., 2021) and the perception of filter bubbles scale (Klug & Strang, 2019), measuring participants’ perceived information narrowing on the selected algorithmic Apps. The first two items focused on awareness of customization of the recommended content (“I feel that the recommended content is tailored for me”, “The recommended content matches my interests”) and the third and fourth items focused on the perception of filter bubbles (“The recommended content is becoming more and more similar”, “The recommended information is similar in themes”). Cronbach's alpha was .65.
Results
Correlation analysis
As displayed in Table 2, Pearson's correlation analysis showed that news knowledge was negatively related to the use of algorithmic social media Apps (r = –.13, p < .05), use of algorithmic short video Apps (r = –.15, p < .01), and algorithm dependence (r = –.16, p < .01), while it was not significantly correlated with the use of algorithmic news Apps (r = –.01, p = .800) or perceived information narrowing (r = –.02, p = .670). Algorithm dependence was positively correlated with the use of algorithmic social media Apps (r = .40, p < .001), use of algorithmic short video Apps (r = .23, p < .001), and perceived information narrowing (r = .17, p < .01), but was not significantly correlated with the use of algorithmic news Apps (r = .10, p = .073).
Correlations Among The Key Variables.
Note. * p < .05, ** p < .01, *** p < .001.
The use of algorithmic Apps was measured as single-item indicators (daily time spent), which are observed variables rather than latent constructs. As shown in Table 2, the correlations among the three usage variables ranged from r = .05 to .50, indicating that they capture distinct types of App use. News knowledge was measured as an objective knowledge score (0–5), which is not a latent construct and therefore was not included in the discriminant validity assessment. Discriminant validity was therefore assessed only for the multi-item latent constructs (algorithm dependence and perceived information narrowing), using the Fornell-Larcker criterion. The square roots of the average variance extracted (AVE) for algorithm dependence (.71) and perceived information narrowing (.65) were both greater than the correlation between the two constructs (r = .09), supporting discriminant validity.
Moderated mediation analysis on news knowledge
To examine the moderated mediation effects for the three types of algorithmic Apps on news knowledge, PROCESS Model 7 (Hayes, 2018) was used for data analysis. This model allows for testing a mediation effect with a moderating variable on the direct path between the independent variable and the mediator. Specifically, algorithmic App use was defined as the independent variable, algorithm dependence was defined as the mediator, news knowledge was defined as the dependent variable, and perceived information narrowing was included as a moderator between algorithmic App use and algorithm dependence. Results for each App category (news, social media, and short video) are displayed in Figure 2.

Relationships among use of algorithmic apps, algorithm dependence, news knowledge, and perceived information narrowing.
In the first model, news knowledge was predicted with use of algorithmic news Apps as the independent variable, algorithm dependence as the mediator, and perceived information narrowing as the moderator. The results showed that the direct link between news App use and news knowledge was nonsignificant (b = .002, p = .962). H1a was not supported. The relationship between use of news Apps and algorithm dependence is moderated by perceived information narrowing moderated this link (b = .15, p < .01). Specifically, post-hoc analysis showed that when the level of perceived information narrowing was high, the effect of the use of algorithmic news Apps on algorithm dependence was significant and positive (b = .395, p < .01), but when the level of perceived information narrowing was medium (b = .095, p = .469) or low (b = –.085, p = .583), the effect of use of algorithmic news Apps on algorithm dependence was nonsignificant. The interaction between the use of algorithmic news Apps and perceived information narrowing on algorithm dependence is shown in Figure 3. Thus, H2a was partially supported and RQ1 was responded positively. Algorithm dependence is negatively correlated with news knowledge (b = –.064, p < .01), supporting H3. In summary, there was a nonsignificant direct effect of algorithmic news App use on news knowledge (b = .002, p = .962), but a negative indirect effect on news knowledge mediated by algorithm dependence and moderated by perceived information narrowing (b = –.010, BootCI [–.019, –.002]), which was significant only among those with high levels of perceived information narrowing (b = –.025, BootCI [–.054, –.004]).

Relationship between use of algorithmic news apps and algorithm dependence at different levels of perceived information narrowing.
In the second model, news knowledge was predicted with use of algorithmic social media Apps as the independent variable, algorithm dependence as the mediator, and perceived information narrowing as the moderator. The results showed that there was no direct effect of use of social media Apps on news knowledge (b = –.070, p = .178). H1b was not supported. The use of social media Apps was positively related to algorithmic dependence (b = 1.668, p < .05), algorithmic dependence was negatively related to news knowledge (b = –.052, p < .05), and perceived information narrowing had a nonsignificant moderating effect (b = –.058, p = .240). H2b and H3 were supported and RQ2 was responded negatively in this model. In sum, there was a nonsignificant direct effect of algorithmic social media App use on news knowledge (b = –.070, p = .178), whereas there was a negative indirect effect through algorithm dependence, which was not moderated by perceived information narrowing (b = .003, BootCI [–.004, .010]).
In the third model, news knowledge was predicted with use of algorithmic short video Apps as the independent variable, algorithm dependence as the mediator, and perceived information narrowing as the moderator. The results showed that the direct effect of the use of short video Apps on news knowledge was significant and negative (b = –.109, p < .05), whereas the indirect effect of algorithm dependence (b = –.036, p = .956; b = –.054, p < .05) and the moderating effect of perceived information narrowing (b = –.040, p = .409) were not significant. H1c and H3 were supported, H2c was not supported, and RQ3 was responded negatively in this model. In sum, there was a negative direct effect of the use of algorithmic short video Apps on news knowledge (b = –.109, p < .05) but a nonsignificant indirect effect; thus, there was no significant moderated mediation (b = –.002, BootCI [–.010, .005]).
Discussion
This study investigated the effects of three distinct types of media platforms that utilize algorithmic recommendation systems on public news knowledge. Overall, this investigation presents a rather concerning portrayal of the public's acquisition of news knowledge within the context of the algorithmic apps under study.
First, it was found that algorithmic apps focused on short videos directly diminish users’ news knowledge. This effect is likely related to the primary entertainment-focused nature of these platforms. From a techno-social ecology perspective, the integration of algorithm-driven, short-form content into everyday routines contributes to a fragmented, entertainment-oriented media ecosystem that prioritizes engagement over informational depth. The consumption of snack news through these Apps tends to give users the illusion of being informed without actually providing them with in-depth factual knowledge (Schäfer, 2020).
Second, this study found that algorithmic social media apps indirectly reduce news knowledge by increasing user dependence on algorithms. This dependence facilitates the filter bubble effect, thus limiting the exposure to diverse perspectives and knowledge. Similar to short video apps, the repetition of topics filled with snack news on algorithmic social media feeds may increase familiarity with certain news topics but does not necessarily translate into a well-rounded understanding, serving merely as a heuristic for judging knowledge (Metcalfe et al., 1993).
Third, the relationship between the use of algorithmic news Apps and news knowledge is the most complex among the three categories studied. It was observed that, when users perceived a high degree of information narrowing, the use of algorithmic news apps indirectly diminished news knowledge through increased algorithmic dependence. This finding is consistent with the techno-social ecology perspective, which posits that digital media environments embedded in everyday life can subtly shape patterns of information exposure and knowledge acquisition, reinforcing certain viewpoints while filtering out others. By contrast, in cases of low-to-moderate perceived information narrowing, the use of algorithmic news Apps does not significantly impact news knowledge.
Algorithm dependence was found to be an important underlying mechanism that explains why the use of algorithmic Apps can potentially decrease news knowledge. Literature suggests that algorithmic experience encompasses three dimensions: cognitive, affective, and behavioral (Swart, 2021). Most existing literature explores the relationship between algorithms and knowledge from the perspective of cognitive processing, such as elaboration curation (Eveland, 2001; Park & Kaye, 2019). However, the algorithm-elaboration-knowledge is not the only path between algorithmic feeds and knowledge. Given that algorithmic apps are fundamentally designed to promote hedonism and continuously exploit user attention, this study prioritizes the affective and behavioral dimensions of use, specifically the persistence and overuse of these apps. From an algorithm dependency perspective, this affective and behavioral engagement fosters a reliance that undermines deep cognitive processing, as users become more accustomed to passive content consumption. By examining the mediating role of algorithm dependence in the relationship between algorithmic app use and knowledge acquisition, the findings confirm that affective and behavioral engagement with these Apps is detrimental to knowledge gain, contrasting with the positive effects of cognitive elaboration.
The most intriguing and significant finding is the friction effect of user perception of information narrowing in safeguarding against algorithm dependence and diminished knowledge gain. It was found to be a key factor in reversing the generally negative impact of algorithmic Apps on knowledge gain. From a techno-social ecology viewpoint, this suggests that users’ awareness of content narrowing can serve as a self-regulating mechanism within the broader media ecosystem, helping to counterbalance the algorithm's natural tendency toward homogeneity. Finding of the study showed that when using algorithmic news Apps, users who perceive a high degree of information narrowing are more likely to become behaviorally and affectively reliant on these apps, leading to overuse and subsequently reduced knowledge acquisition. Conversely, users who perceive medium or low levels of information narrowing do not develop such dependence, thereby avoiding any negative impact on knowledge gain.
Findings of this study have some practical implications. First, results suggest that even Apps designed primarily for news dissemination fail to effectively enhance news knowledge. Unlike traditional news websites, algorithmic news apps follow algorithmic logic, which offers an infinite and personalized stream of headlines. Users must sift through numerous tailored headlines to find news that interests them, which can result in fragmented information intake and cognitive overload (Van Aelst et al., 2017). Moreover, readers still need to click on titles that interest them and read full articles for news elaboration and engagement to facilitate news knowledge gain. This step is paramount for transitioning from a superficial encounter with news to deeper understanding and retention of information. If only headlines are read, the information obtained remains fragmented, creating an illusion of knowledge. Second, findings indicate that perceived information narrowing can function as a form of user friction, helping to regulate the extent of algorithmic customization and ultimately promoting knowledge acquisition. Additionally, perceived information narrowing may act as an early indicator of the filter-bubble effect, emphasizing the importance of empowering users to engage in proactive personalization. Current algorithmic Apps provide personalization settings that function much like a simple “on/off switch”. Turning off these settings or quit the Apps entirely can lead to considerable inconvenience and cost, often leading many users to keep personalization features activated to avoid these drawbacks. However, this study suggests that a more nuanced approach to personalization could be more effective and meaningful. A “faucet valve” type of switch to the content streams would be ideal, where users have the ability to finely adjust the degree of personalized content they receive. Algorithmic Apps and platforms should proactively provide users with periodic feedback alerts that enable them to report or assess their perceived degree of information narrowing during App use. This feedback can assist in adjusting the breadth of content delivered by the algorithms. When users report a high level of narrowing, algorithms should then expand content diversity to counteract this effect. This finding aligns with researchers’ advocacy for promoting news diversity through personalized nudges in algorithmic recommender systems, suggesting that such nudges can gradually increase the diversity of users’ news consumption, helping them develop new reading habits and explore novel interests (Mattis et al., 2024). This finding also supports the concept of algorithmic sovereignty which advocates users’ right to decide how, and to what extent, algorithms control online life (Reviglio & Agosti, 2020). While existing practices of customization on online platforms, services, and content are often performed without the user's explicit choice, future efforts of algorithmic Apps should aim to actively offer explicit user friction choices within the infinite scrolling of algorithmic feeds (Lee et al., 2019; Merten, 2021; Zuiderveen Borgesius et al., 2016).
This study had certain limitations. First, there could be inherent selection bias in our sample. The data collection method, online snowball sampling, yielded a participant pool skewed towards younger individuals with higher education levels, and caution must be exercised when extrapolating these results to broader populations. Older adults may exhibit lower algorithmic dependence due to more established news consumption habits and potentially lower social media frequency, yet they might also be more vulnerable to algorithmic curation if they possess less digital literacy to recognize personalization mechanisms. For less educated populations, individuals may rely more heavily on algorithmic Apps as primary news sources, potentially leading to higher algorithm dependence, but they may also engage in less elaborative processing, which could amplify the negative effects of algorithm dependence on knowledge acquisition. Future research should directly test these moderating effects across diverse demographic groups. Second, the cross-sectional design of the study does not rule out reversed causal mechanisms as explanations of the findings (e.g., participants with little knowledge might be more inclined to use short video apps). A longitudinal design would have been a better choice. Third, some of the measurement of the key variables employed in this study could be improved. While some studies suggest a negligible correlation between time spent on social media and dependence, engagement with social media activities—such as sharing, liking, and commenting—has been positively linked to dependence (Kim & Jung, 2017). Our research utilized time/frequency as metrics for algorithmic app usage, potentially overlooking the depth of user engagement. Future investigations could benefit from measuring algorithmic App use with detailed engagement into distinct categories, such as politics, entertainment, and sports, to ascertain their differential impacts on algorithm dependence. In addition, the measurement of algorithmic social media App use focused exclusively on Sina Weibo, which may not fully capture the diversity of social media platforms users engage with. Future research should consider including a broader range of platforms to enhance generalizability. Moreover, exploratory factor analysis revealed that the four items of perceived information narrowing loaded onto two distinct dimensions—perceived personalization and perceived content homogeneity—suggesting that the scale may capture related but not identical facets of information narrowing. The relatively low internal consistency (α = .65) further reflects this multidimensionality, and future research should consider developing separate subscales to more precisely measure each dimension. Fourth, the timing of the study, which was conducted at the end of 2019, coincided with a period of intense scrutiny and debate over China's algorithm recommendation technology. Since then, major regulatory changes have been introduced, including the Personal Information Protection Law (2021) and the Internet Information Service Algorithm Recommendation Management Provisions (2022), which require platforms to provide users with options to disable personalized recommendations and increase algorithmic transparency. These changes may reduce algorithm dependence by enhancing user awareness and control. Meanwhile, major algorithmic platforms have introduced features such as usage reminders, content diversity labels, and manual refresh options that introduce “user friction” and may mitigate algorithm dependence. Consequently, the levels of algorithm dependence observed in our 2019 data may be higher than what would be observed in the current environment. Our findings are best interpreted as capturing the “early stage” of algorithmic media consumption before widespread regulatory and platform interventions. Future research should examine how these regulatory and platform changes have reshaped the relationship between algorithmic App use, algorithm dependence, and knowledge acquisition.
Conclusion
This study scrutinized the impact of news, social media, and short video algorithmic Apps on news knowledge, revealing that such systems might impede rather than enhance comprehensive news knowledge through different underlying mechanisms. Algorithmic dependence played a key mediating role. Our findings contribute to the understanding of algorithm dependence by illustrating its mediating role in the relationship between algorithmic App use and knowledge acquisition. The study highlights how reliance on algorithmic curation can fragment users’ informational environment, leading to cognitive shortcuts that limit comprehensive knowledge gain. Moreover, the moderating effect of perceived information narrowing suggests the potential for introducing a user friction mechanism to regulate algorithmic curation and mitigate the filter bubble effect, particularly within algorithmic news Apps. This psychological cue may prompt users to seek a broader range of information, acting as a self-regulatory prompt. To support this process, platforms could offer personalized settings that provide feedback on content diversity or suggestions for broadening information exposure. Within algorithmic news applications, such features could allow users to adjust the flow of personalized content, actively counteracting filter bubble effects and increasing news diversity.
Footnotes
Ethical statements
This research is funded by National Social Science Fund of China (Grant Number: 22CWX017). Approval for the online survey procedures of this study was obtained from the academic committee (act as ethics committee) of School of Humanities, Shanghai University of Finance and Economics on 1st, September, 2019 (Approval number: 2019-9-1). All procedures involving human participants in the research were performed in accordance with the ethical standards of the institutional research committee and with the 1964 Declaration of Helsinki and its later amendments or comparable standards. The initial page of the online survey provided participants with a succinct overview of the survey's objectives and the organizing entity. Additionally, it included a confidentiality statement assuring that their responses would be kept confidential and solely utilized for academic research purposes. Informed consent was secured from participants upon the completion of the online survey.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the National Social Science Fund of China, (grant number 22CWX017).
Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Availability of data and materials
The data underlying this article will be shared on reasonable request to the corresponding author
