Psychological correlates of problematic short video use risk profiles: A latent profile and network analysis

Abstract

Background

The rapid rise of short video platforms has raised growing concerns about problematic short video use (PSVU). However, most existing research has adopted variable-centered approaches, with limited attention to potential heterogeneity among users with different levels of PSVU. In addition, little is known about how psychological correlates vary across PSVU risk profiles.

Methods

To address this gap, the present study employed latent profile analysis (LPA) and network analysis to identify PSVU risk profiles, characterize their psychological features, and explore associations among relevant psychological factors. A total of 1891 participants completed self-report measures assessing short video use patterns and related psychological variables.

Results

Three PSVU risk profiles were identified: adaptive users, vulnerable users, and problematic users. These profiles differed in short video use patterns and psychological characteristics. Network analyses further indicated differences in network structure across profiles. Motivational factors exhibited relatively high centrality, particularly among higher-risk profiles. The observed associations among emotional distress, personality traits, motivational factors, and behavioral indicators were summarized within an Emotion–Personality–Motivation–Behavior network perspective.

Conclusions

The findings provide an exploratory person-centered perspective on PSVU heterogeneity and suggest that the psychological correlates of PSVU vary across risk profiles. More broadly, the results may help inform future longitudinal research and support the development of more tailored approaches for identifying and addressing PSVU.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s40359-026-05145-6.

Introduction

Problematic short video use

Over the past decade, advances in internet technologies and artificial intelligence have contributed to the rapid development of short video platforms, which have become one of the most widely used forms of social media worldwide. TikTok has reported over 1.5 billion monthly active users and more than 4 billion downloads globally [34]. In China, the number of short video users reached nearly 1.04 billion, accounting for 93.8% of the country’s internet population [20]. Given the substantial amount of time users spend on short video platforms each day, short video use has become an increasingly important aspect of everyday digital life.

This surge in usage has raised growing concerns about problematic short video use (PSVU), generally described as excessive and poorly controlled short video engagement that is associated with negative psychological or functional outcomes [19, 88, 92, 94]. Prior research has reported associations between PSVU and various psychosocial difficulties, including anxiety, depression, and stress [17, 63, 69], sleep disturbances [38, 68], academic difficulties [17, 93], and cognitive functioning problems such as attention and memory difficulties [69, 84, 91, 96]. Given its widespread use and potential negative correlates, more research is needed to better understand the psychological factors linked to PSVU.

The I-PACE model

While various frameworks exist to explain addictive behaviors such as PSVU, the Person–Affect–Cognition–Execution model (I-PACE; [13, 14]) was selected for this study because it offers a more comprehensive and dynamic perspective compared to traditional models. According to the I-PACE model, problematic Internet-related behaviors arise from dynamic interactions among predisposing person-related characteristics, affective and cognitive responses to internal or external stimuli, and executive control processes.

Within this framework, the Person (P) component represents relatively stable predispositions that increase susceptibility to problematic online behaviors. These characteristics do not directly cause addiction, but create a vulnerability that manifests under specific conditions. The Affect (A) and Cognition (C) components represent more proximal affective and cognitive responses. Triggered by internal states or external cues, individuals may experience negative affective states and develop specific motivations. These responses may contribute to urges to engage in online behaviors as a means of regulating emotions or temporarily disengaging from everyday difficulties. The Execution (E) component involves executive functions. This system acts as a regulatory mechanism, allowing individuals to resist impulses and maintain behavioral flexibility.

These four components interact differently across the stages of addictive behavior [13]. In the early stages, individuals with specific predispositions (P) may engage in online behaviors driven by positive affective and cognitive expectancies (A/C) when executive control (E) is momentarily weakened, while in the late stages, affective responses evolve into intense craving and withdrawal symptoms, while cognitive responses become automated habits. Crucially, executive control (E) significantly deteriorates in the presence of addiction-related cues. Consequently, negative reinforcement (avoiding distress) becomes the dominant driver, resulting in rigid, compulsive behaviors that are difficult to control.

Core psychological variables in the network

This study posits that PSVU shares similarities with internet addiction and is associated with multidimensional psychological factors. However, existing research has largely focused on isolated correlates of PSVU while paying less attention to the interrelationships among multiple psychological characteristics. Therefore, the present study employed network analysis to explore patterns of associations among theoretically relevant psychological variables.

In selecting variables for the network analysis, the present study prioritized psychological constructs that most directly corresponded to the core components of the I-PACE framework and that have been consistently implicated in problematic Internet-related behaviors in prior research. At the person level, neuroticism and extraversion were included, as previous network studies have identified these personality traits as relevant correlates of problematic social media use (PSMU; [46]). In addition, self-esteem and self-control were included as two important person-related characteristics within the I-PACE framework. Low self-esteem has been widely associated with problematic internet-related usage [16, 67, 85], whereas self-control is considered a central vulnerability factor linked to problematic online behavior [35, 47].

At the affect and cognition level, depression, anxiety, stress, and loneliness were included because substantial evidence has linked these negative emotional states to PSVU [17, 49, 95] and PSMU [46, 60]. Additionally, escapism and stress-relief motivation were included because our previous research identified these motivations as showing the strongest associations with PSVU [19]. Habitual use was also incorporated, as repetitive and automatic engagement patterns are considered important reinforcement-related processes within the I-PACE framework and have been consistently linked to problematic online behaviors [26, 78].

Finally, at the execution level, cognitive flexibility was included because it represents an important component of executive functioning built upon inhibitory control and working memory processes [24]. Reduced cognitive flexibility may be related to difficulties regulating attention, emotions, and behavioral responses in an adaptive manner [8, 80], and may therefore be related to more rigid patterns of short video use. Additional behavioral variables, including usage patterns (active/passive use), usage and engagement level, algorithm dependency, and other motivational dimensions, were also assessed in this study. However, these variables were used primarily for demographic characterization and were not included in the network models as the present study primarily focused on psychological constructs most theoretically relevant to the I-PACE framework and most consistently associated with PSVU based on existing evidence.

Person-centered approaches

Beyond examining psychological association patterns related to PSVU, relatively limited attention has been paid to potential heterogeneity among individuals with different levels of PSVU. This is important because PSVU may manifest differently across individuals, with potentially distinct psychological characteristics across risk levels [60, 70, 72]. Identifying subgroups characterized by different PSVU risk profiles may therefore provide a more nuanced understanding of how psychological characteristics co-occur across different user groups.

Latent profile analysis (LPA) is a model-based person-centered approach that identifies unobserved subpopulations characterized by distinct patterns across variables [51, 59]. Compared with traditional clustering techniques, LPA provides several methodological advantages, including probabilistic class assignment, formal statistical criteria for model evaluation, and estimates of classification uncertainty, thereby improving the reliability and interpretability of subgroup classification [22, 56]. Moreover, person-centered approaches may reveal qualitatively different psychological configurations that are not observable through overall severity scores alone [6]. Previous studies have successfully applied LPA to problematic internet use [61], PSMU [60], and related addictive online behaviors [4, 43], demonstrating meaningful heterogeneity in psychological characteristics and behavioral risk patterns across latent subgroups. Therefore, LPA provides a useful framework for identifying distinct PSVU profiles before examining subgroup-specific psychological association patterns.

The present study

Therefore, combining LPA with network analysis may offer a more nuanced understanding of heterogeneity in PSVU. Specifically, LPA can identify distinct PSVU subgroups [22, 73], whereas network analysis can be used as an exploratory tool to examine statistical associations among psychological variables within each subgroup [27]. By integrating these approaches, the present study aimed to identify PSVU risk profiles, characterize their psychological features, and explore patterns of associations among theoretically relevant variables. Within this context, the observed associations were interpreted from an exploratory Emotion–Personality–Motivation–Behavior perspective informed by the I-PACE framework to facilitate interpretation of the network patterns rather than to propose or test a causal theoretical model.

Method

Participants

Data were collected from 2246 individuals between April 26 and May 8, 2025, using the online survey platform Wenjuanxing (www.wjx.cn). Data quality control was implemented prior to analysis. Specifically, two attention-check items were embedded within the questionnaire, and participants who failed both items were excluded. Responses with completion times shorter than 400 s, duplicate submissions, and highly invariant response patterns (e.g., selecting the same response option across the majority of items) were also excluded. After excluding incomplete and invalid responses based on these predefined criteria, the final analytic sample consisted of 1891 participants (male-to-female ratio = 0.69; Mage = 22.79 years, SD = 4.62). Participants reported an average of 14.87 years of education (SD = 1.62). The sample primarily consisted of students (n = 1402, 74.14%), most of whom were undergraduate students, with smaller numbers of postgraduate and high school students. Non-student participants (n = 489, 25.86%) mainly included employed or unemployed adults. Participants most commonly used Douyin (Chinese version of TikTok; n = 1457, 77.05%), followed by Kuaishou (Chinese version of Kwai; n = 183, 9.68%), Xiaohongshu (RedNote; n = 134, 7.09%), Bilibili (n = 85, 4.49%), and other short video platforms (n = 32, 1.69%). The study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Shanghai Mental Health Center (Approval No. 2023-63). All participants provided informed consent prior to participation. The survey took approximately 10 min to complete, and participants received 5 Chinese yuan (RMB) as compensation for their participation.

Measures

Participants completed a series of self-report questionnaires assessing short video use patterns and related psychological variables (see Table 1). The core variable, PSVU, was assessed using the self-developed Problematic Short Video Use Scale (PSVUS). The PSVUS consists of 21 items across five dimensions: cognitive interference, usage dependence, physiological damage, control failure, and social alienation. All items are rated on a 5-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree), with higher scores indicating more severe problematic use.

Table 1.

The brief information of the questionnaires

Measurement	Variable	Questionnaire (description)	Items	Cronbach’s α	Reference
SV usage	PSVU	Problematic Short Video Use Scale (PSVUS)	21	0.97	Self-developed
	SV engagement	The extent to which participants used SV in daily lives	6	0.72	Przybylski et al., [62]
	SV usage pattern (active/passive)	Passive and active SV use	7	0.72	Li, [48]
	SV usage motivation	Motives of SV usage	24	0.88	Omar and Dequan, [57]
	Algorithm dependency	Use of recommender systems	6	0.82	Hasan et al., [32]
Personal characteristics	Neuroticism	Chinese Big Five Personality Inventory Brief Version (CBF-PI-B)	8	0.90	Wang et al., [87]
	Extraversion		8	0.87
	Self-control	Brief Self-Control Scale (BSCS)	7	0.84	Morean et al., [54]
	Self-esteem	Rosenberg Self-esteem Scale (RSE)	10	0.87	Rosenberg, [65]
	Cognitive flexibility	Cognitive Flexibility Inventory (CFI)	20	0.92	Dennis and Vander Wal, [25]
Negative emotion	Depression	Depression Anxiety Stress Scale (DASS-21)	7	0.91	Antony et al., [2]
	Anxiety		7	0.87
	Stress		7	0.88
	Loneliness	UCLA Loneliness Scale (ULS-8)	8	0.87	Hays and DiMatteo, [33]

SV Short video

The scale was developed through a multi-stage validation procedure, including item generation based on theoretical and empirical literature, as well as exploratory and confirmatory factor analyses conducted in separate samples. The final PSVUS demonstrated good reliability (Cronbach’s α = 0.937, McDonald’s ω = 0.944, test-retest reliability = 0.855) and satisfactory construct validity (χ²/df = 2.276, CFI = 0.962, IFI = 0.962, RMSEA = 0.047). Criterion validity was evaluated using the Problematic Mobile Social Media Usage Assessment Questionnaire (PM-SAQ), a 20-item instrument widely used in Chinese research on PSMU [39]. The PM-SAQ has demonstrated strong psychometric properties, with a Cronbach’s alpha of 0.91 and a test–retest reliability coefficient of 0.87. Correlations between PSVUS and PM-SAQ dimensions were positive and statistically significant, supporting the criterion validity of the PSVUS. In the present sample, the PSVUS showed good internal consistency (Cronbach’s α = 0.97). Detailed information regarding the development and psychometric validation of the PSVUS is provided in Appendix S1, while descriptions of the remaining measures are presented in Appendix S2.

Statistical analysis

LPA was conducted using Mplus 8.3 [55] to identify latent PSVU profiles. LPA estimates the probability of each participant belonging to a latent subgroup and determines the optimal number of profiles based on model fit indices [77]. The analysis began with a one-class solution, with the number of classes increased sequentially to identify the best-fitting model. Model selection was assessed by multiple fit indices, including the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), sample-size adjusted BIC (aBIC), Lo–Mendell–Rubin adjusted likelihood ratio test (LMRT), bootstrapped likelihood ratio test (BLRT), entropy, the Relative Fit Index for Small Classes (RFSC), and the Classification Maximum Likelihood Posterior Probability (CMP). Better model fit is generally reflected by lower AIC, BIC, and aBIC values, higher entropy values, significant LMRT and BLRT tests, and RFSC values exceeding 5% [56, 73].

Network analysis was conducted in R 4.5.1 [64] using the packages qgraph, igraph, mgm, bootnet, and NetworkComparisonTest. A Gaussian Graphical Model (GGM) was estimated using the Extended Bayesian Information Criterion (EBIC) for model selection [29]. In the network, variables were represented as nodes, and edges represented regularized partial correlations between variables and should not be interpreted as evidence of causal relationships. Communities (i.e., clusters of strongly connected nodes) were interpreted descriptively as groups of closely related variables rather than as evidence of underlying causal processes [9]. Node centrality indices were calculated, including strength and expected influence (EI), which are generally considered the most interpretable centrality measures in psychological networks. Closeness and betweenness were reported descriptively due to their relatively lower stability in psychological network research [28]. Predictability of each node was assessed using R² values. The stability of centrality estimates was evaluated using nonparametric bootstrapping with 1000 resamples to generate 95% confidence intervals. The correlation stability coefficient (CS-coefficient) was also computed to assess the robustness of centrality metrics, with values above 0.25 considered acceptable and above 0.50 considered good [29]. Finally, network comparison tests (NCTs) were used to examine differences in global network structure and specific edge weights across profiles [10, 81].

To facilitate interpretation of practical significance, effect sizes were reported for profile comparisons. For one-way analyses of variance (ANOVAs), eta squared (η²) was calculated, with values of approximately 0.01, 0.06, and 0.14 interpreted as small, medium, and large effects, respectively. For mixed ANOVAs, partial eta squared (η2 p) was reported and interpreted using the same conventional benchmarks [21]. Multiple comparisons were adjusted using the Holm–Bonferroni method.

Results

Latent profile analysis

LPA was conducted using participants’ item-level responses on the PSVU scale as observed indicators. Model fit indices for solutions with varying numbers of latent profiles are presented in Table 2. As shown in Fig. 1, improvements in model fit became smaller after the three-profile solution, suggesting limited incremental benefit from additional profiles. The selection of the three-profile solution was based on a combination of statistical fit indices (AIC, BIC, aBIC, LMRT, BLRT, and RFSC), classification quality (entropy and posterior probabilities), parsimony, and interpretability, consistent with recommended practices [56, 73]. The three-profile model demonstrated high classification accuracy, with an entropy value of 0.951 and high average posterior probabilities (0.985, 0.970, and 0.981), suggesting good separation among the identified profiles. For subsequent network and network comparison analyses, participants were assigned to latent profiles based on their highest posterior probability (i.e., most likely class membership). Thus, participants were assigned to the profile for which they had the highest posterior probability. The high entropy value and posterior probabilities suggested relatively low classification uncertainty.

Table 2.

Fit indices for the latent profile models

Classes	AIC	BIC	aBIC	pLMR	pBLR	RFSC(%)	Entropy	CMP(%)
1	133833.874	134066.758	133933.324	-	-	-	-
2	113596.121	113950.992	113747.664	< 0.001	< 0.001	40.719	0.970	0.407/0.593
3	108010.762	108487.620	108214.398	< 0.001	< 0.001	23.109	0.951	0.231/0.353/0.416
4	105892.714	106491.559	106148.444	< 0.001	< 0.001	17.292	0.931	0.224/0.301/0.173/0.301
5	105013.881	105734.713	105321.703	0.477	< 0.001	9.572	0.920	0.170/0.202/0.244/0.096/0.289

Fig. 1.

Scree plot of change trend of aBIC

Profile characteristics

Descriptive statistics are presented in Table 3. On average, participants had used short video platforms for 5.31 years (SD = 1.86), with a mean daily usage time of 3.25 h (SD = 1.59). Regarding key behavioral and psychological variables, PSVU was positively correlated with depression, anxiety, stress, loneliness, neuroticism, escapism, habitual use, and stress-relief motivation, and negatively correlated with self-control, extraversion, cognitive flexibility, and self-esteem (all p < 0.001; see Appendix S3).

Table 3.

Profile characteristics

Dimension	Variable	Total (n = 1891)	Adaptive (n = 437)	Vulnerable (n = 668)	Problematic (n = 786)	p	η²
Demographic	Age	22.78 (4.64)	22.92 (5.11)	22.72 (4.64)	22.77 (4.32)	0.772	< 0.001
	Gender (Male/Female)	776/1115	216/221	245/423	315/471	< 0.001	0.010
SV usage	Years of usage	5.31 (1.86)	5.24 (1.86)	5.26 (1.85)	5.39 (1.77)	0.282	0.001
	Daily usage (h)	3.25 (1.59)	2.65 (1.53)	3.14 (1.54)	3.63 (1.59)	< 0.001	0.057
	SV Engagement	35.00 (8.87)	30.56 (9.21)	34.33 (8.66)	38.03 (7.62)	< 0.001	0.108
	Algorithm dependency	32.99 (5.40)	31.59 (6.59)	31.76 (5.27)	34.80 (4.10)	< 0.001	0.081
	SV passive use	12.02 (1.95)	11.83 (2.11)	11.83 (1.98)	12.29 (1.81)	< 0.001	0.013
	SV active use	11.95 (3.50)	12.33 (3.85)	11.32 (3.24)	12.27 (3.42)	< 0.001	0.018
	PSVU	65.64 (21.13)	35.49 (7.64)	61.46 (7.58)	85.94 (7.77)	< 0.001	0.868
Motivation	Information acquisition	12.28 (1.80)	12.61 (1.68)	12.10 (1.77)	12.87 (1.87)	< 0.001	0.011
	Novelty seeking	12.13 (1.89)	11.97 (2.20)	11.69 (1.87)	12.60 (1.60)	< 0.001	0.046
	Entertainment	12.81 (1.66)	12.45 (1.89)	12.55 (1.73)	13.23 (1.35)	< 0.001	0.045
	Stress relief	12.20 (2.14)	11.55 (2.60)	11.76 (2.14)	12.94 (1.56)	< 0.001	0.086
	Social interaction	14.24 (6.24)	14.96 (6.68)	13.29 (5.76)	14.65 (6.29)	< 0.001	0.013
	Self-expression	9.56 (3.63)	9.89 (3.70)	8.70 (3.56)	10.11 (3.52)	< 0.001	0.032
	Escapism	9.80 (3.28)	7.62 (3.30)	8.99 (2.89)	11.71 (2.43)	< 0.001	0.264
	Habitual use	11.93 (2.32)	10.68 (2.92)	11.57 (2.17)	12.93 (1.52)	< 0.001	0.151
Emotion	Depression	4.99 (4.78)	2.11 (3.04)	4.30 (4.08)	7.18 (5.10)	< 0.001	0.178
	Anxiety	5.76 (4.46)	2.93 (3.17)	5.42 (3.87)	7.62 (4.63)	< 0.001	0.168
	Stress	6.95 (4.70)	3.69 (3.40)	6.43 (4.09)	9.22 (4.61)	< 0.001	0.212
	Loneliness	16.56 (5.15)	12.71 (3.96)	16.19 (4.49)	19.02 (4.85)	< 0.001	0.226
Personality/trait	Cognitive flexibility	72.18 (6.05)	80.94 (11.26)	72.42 (10.79)	67.10 (12.00)	< 0.001	0.180
	Self-control	21.51 (5.78)	27.33 (4.51)	21.89 (4.61)	17.94 (4.39)	< 0.001	0.395
	Neuroticism	26.13 (8.73)	18.27 (7.53)	25.80 (7.10)	30.78 (7.30)	< 0.001	0.306
	Extraversion	30.44 (8.23)	35.46 (7.87)	29.95 (7.36)	28.07 (7.93)	< 0.001	0.121
	Self-esteem	29.53 (5.27)	32.78 (4.06)	29.71 (4.75)	27.57 (5.36)	< 0.001	0.146

Means and standard deviations (in parentheses) of demographic characteristics; Gender was coded as 0 = male and 1 = female

Figure 2 presents the mean item scores across the three latent profiles. Based on their PSVU score patterns, the three profiles were descriptively labeled as “adaptive users” (AU, 23.11%), “vulnerable users” (VU, 35.33%), and “problematic users” (PU, 41.57%) to facilitate interpretation. These profiles showed significant differences in short video usage patterns and psychological characteristics (see Table 3). In terms of usage patterns (Fig. 3), the PU profile reported the longest daily short video use (3.63 h), followed by the VU (3.14 h) and AU profiles (2.65 h), with all pairwise differences statistically significant (p < 0.001). The PU profile also showed higher levels of algorithm-related engagement and platform involvement compared to the other profiles. In addition, the PU profile reported relatively higher levels of passive use behaviors compared to the AU and VU profiles, whereas both the AU and PU profiles reported higher levels of active use behaviors than the VU profile (p < 0.001). A two-way mixed ANOVA was conducted to examine differences in motivation levels across PSVU profiles and motivation types, with PSVU profile (AU, VU, PU) as the between-subject factor and the eight motivation categories (information acquisition, novelty seeking, entertainment, stress relief, social interaction, self-expression, escapism, and habitual use) as the within-subject factor. The analysis revealed a significant interaction effect between PSVU profile and motivation type (F = 47.227, p < 0.01, η2 p = 0.048), indicating that the pattern of motivation levels differed across the three profiles. Follow-up pairwise comparisons showed that the PU profile reported higher levels of novelty seeking, entertainment, stress-relief, escapism, and habitual use motivations compared to the other profiles (p < 0.01). Regarding psychological characteristics, the PU profile reported higher levels of depression, anxiety, stress, loneliness, and neuroticism, as well as lower levels of cognitive flexibility, self-control, extraversion, and self-esteem compared to the other profiles, with all pairwise differences statistically significant (all pairwise differences p < 0.001; see Appendix S4).

Fig. 2.

Mean item scores of PSVUS for the 3-latent profile. PSVUS dimensions: CI = Cognitive Interference, UD = Usage Dependence, PD = Physiological Damage, CF = Control Failure, SA = Social Alienation

Fig. 3.

Short video usage characteristics of the PSVU profiles. Panel a shows years of short video use; panel b shows daily usage duration; panel c shows engagement; panel d shows algorithm dependency; panel e shows passive use; panel f shows active use; and panel g shows short video use motivations. IA = information acquisition, NS = novelty seeking, EN = entertainment, SR = stress relief, SI = social interaction, SE = self-expression, ES = escapism, and HA = habitual use. Error bars represent standard deviations. ^*p < 0.05, ^**p < 0.01, ^***p < 0.001; ns = non-significant

Network analysis

Network estimation

Based on the I-PACE framework and prior empirical research [3, 17, 19, 45, 46, 74], theoretically relevant psychological variables were selected for exploratory network analysis within each PSVU profile. Figure 4 displays the estimated networks for the adaptive, vulnerable, and problematic profiles. Each network included 13 nodes, which could be descriptively grouped into three domains (A = emotional variables; B = personality traits; C = motivational factors) for ease of presentation. Across profiles, different patterns of associations between PSVU and other variables were observed. In the AU network, the strongest edges connected PSVU with self-control (r = -0.163), cognitive flexibility (r = -0.102), and neuroticism (r = 0.093). In the VU network, PSVU showed relatively stronger associations with escapism (r = 0.241), self-control (r = -0.145), habitual use (r = 0.105), and stress (r = 0.066). In the PU network, the largest edge weights were observed between PSVU and escapism (r = 0.221), habitual use (r = 0.218), stress-relief motivation (r = 0.178), and self-control (r = -0.151). Other edge weights across the three networks were relatively small. Detailed edge weights and network statistics are provided in Appendix S5.

Fig. 4.

Estimated psychological networks across the three PSVU profiles: (a) adaptive users, (b) vulnerable users, and (c) problematic users. Nodes represent psychological variables and edges represent regularized partial correlations between variables after controlling for all other nodes in the network. Blue edges indicate positive associations and red edges indicate negative associations, with thicker edges representing stronger associations. Node positions are determined algorithmically for visualization purposes and should not be interpreted as reflecting the quantitative strength or distance of associations

Network centrality

Figure 5 presents node centrality indices (strength, betweenness, closeness, and expected influence) across the three PSVU profiles. These indices were interpreted descriptively to identify relatively well-connected nodes within each network. In the AU network, neuroticism showed the highest strength (2.11) and relatively high predictability (R² = 0.72). In the VU network, depression showed the highest strength (1.35), while stress showed the highest expected influence (1.70). In the PU network, stress showed the highest strength (1.50) and expected influence (1.63). Descriptively, the PSVU node showed higher strength and expected influence in the VU and PU networks than in the AU network. In addition, expected influence for neuroticism was higher in PU compared to AU (p = 0.010) and VU (p = 0.040). Detailed statistics on node centrality and between-profile comparisons are provided in Appendices S6 and S7.

Fig. 5.

Centrality plots for the networks depicting the strength, betweenness, closeness, and expected influence of the three PSVU profiles

Network comparison

Network comparison tests indicated differences in overall network structure and global strength across the three PSVU profiles. Pairwise comparisons indicated significant differences in overall network structure (p < 0.01). In addition, the PU network showed higher global strength than the AU and VU networks. More specifically, the PU profile showed stronger associations between PSVU and several motivational variables than the AU profile, although some of these differences were modest in magnitude. Compared to AU, PU showed larger edge weights between PSVU and escapism (mean difference = 0.184, p = 0.010), habitual use (mean difference = 0.181, p = 0.010), and stress-relief motivation (mean difference = 0.199, p = 0.040). Compared to VU, PU also showed stronger associations between PSVU and habitual use (mean difference = 0.113, p = 0.040) and stress-relief motivation (mean difference = 0.178, p = 0.010).

Several between-profile differences also emerged among emotional, personality, and motivational variables. For example, depression and stress showed stronger associations with neuroticism in VU and PU compared to AU, and the association between neuroticism and escapism was also stronger in PU. In addition, PSVU showed stronger associations with motivational variables and self-control in the PU profile. These findings were interpreted as descriptive patterns of co-occurrence among variables rather than evidence of causal pathways. Detailed NCT results are provided in Appendix S8.

Network accuracy and stability

Bootstrapped confidence intervals for edge weights were relatively narrow, suggesting acceptable estimation precision (Appendix S9). The CS-coefficients for strength centrality were 0.75 across networks, indicating good stability according to recommended guidelines (Appendix S10).

Discussion

The present study contributes to the growing literature on PSVU primarily by identifying heterogeneous user profiles using a person-centered approach. In addition, the network analyses provide complementary evidence regarding how emotional, personality, and motivational factors are associated with PSVU within these profiles.

Profiles of PSVU

Using latent profile analysis, this study identified three user groups: adaptive users, vulnerable users, and problematic users. Adaptive users generally demonstrated relatively low levels of PSVU and more favorable psychological functioning, whereas vulnerable users showed intermediate levels of PSVU and psychological risk. In contrast, problematic users exhibited the highest levels of PSVU, emotional distress, and maladaptive motivational tendencies, together with lower levels of self-regulatory resources such as self-control, self-esteem, and cognitive flexibility.

Although the identified profiles appeared to differ primarily in the severity of PSVU and psychological vulnerability rather than entirely distinct symptom configurations, they nevertheless demonstrated meaningful variation in behavioral and psychological characteristics. In particular, the largest differences were observed between problematic users and the other two groups, suggesting that problematic users may represent a subgroup characterized by elevated psychological distress and reduced regulatory capacity. These findings further suggest that PSVU is accompanied by meaningful variation in psychological functioning and short video engagement patterns across users. Problematic users appeared more likely to engage with short video platforms in an intensive and passive manner, relying more heavily on algorithm-driven content feeds and using short videos for emotional coping purposes such as escapism and stress relief. Such engagement patterns may co-occur with lower intentional media regulation and greater psychological discomfort [17, 23].

The present findings are consistent with previous research suggesting that PSVU may vary along a continuum of psychological vulnerability and behavioral severity rather than reflecting entirely discrete categories [11, 50]. From a theoretical perspective, the identification of these profiles contributes to the understanding of heterogeneity in PSVU by demonstrating that differences across users may involve both graded severity and meaningful variation in psychological characteristics. Notably, the identification of a vulnerable-user profile may be particularly informative. Compared with adaptive users, vulnerable users exhibited elevated emotional distress and lower self-regulatory resources, yet their PSVU levels remained substantially below those of problematic users. These findings suggest that psychological vulnerability and PSVU severity may be partially dissociable across user profiles. Instead, some individuals may experience elevated psychological risk before reaching more severe levels of problematic use. From a person-centered perspective, the identification of an intermediate vulnerable-user profile further highlights the graded nature of PSVU and the potential importance of psychological vulnerability prior to more severe problematic use. At the same time, because the identified profiles largely reflected differences in overall PSVU severity, they may be best understood as representing different levels of risk along a continuum of PSVU rather than clearly differentiated user types. Consequently, conclusions regarding qualitatively distinct forms of PSVU remain limited. Future research may benefit from incorporating broader behavioral and psychological indicators into person-centered analyses to examine whether more differentiated profile structures emerge.

Motivational factors of PSVU

Network analysis indicated that motivational variables, particularly escapism, habitual use, and stress relief, showed relatively strong associations with PSVU, especially among vulnerable and problematic users. Although some between-profile differences in edge strength were numerically modest, the overall pattern consistently suggested stronger associations between motivational factors and PSVU among higher-risk users than among adaptive users. These findings suggest that motivational processes may be more closely linked to PSVU than several other psychological characteristics across profiles.

These findings are consistent with previous research identifying motivational processes as important correlates of problematic internet behaviors [5, 50, 86] and as proximal factors associated with problematic engagement [97]. According to Uses and Gratifications (U&G) theory [41, 53, 76], individuals actively use media to satisfy psychological and social needs. The immediacy, personalization, and accessibility of short video platforms may facilitate repeated engagement and strengthen motivationally driven patterns of use [18, 66, 79].

Among the motivational factors examined, escapism, stress relief, and habitual use showed the most prominent associations with PSVU [52, 58, 75, 78, 97]. Consistent with our previous findings [19], escapism demonstrated relatively strong associations with PSVU among vulnerable and problematic users, whereas stress relief showed stronger associations within the problematic group. These findings are also consistent with the compensatory model of problematic internet use [40], which proposes that online activities may be used to cope with psychological distress or unmet needs. However, because the present networks were estimated from cross-sectional data, these associations should not be interpreted as evidence of causal motivational mechanisms.

Habitual use also emerged as a prominent correlate of PSVU. Habit reflects increasing automaticity in behavior, whereby repeated actions become less dependent on conscious decision-making and more strongly guided by contextual cues [30, 90]. In the present study, habitual use showed relatively strong associations with PSVU among vulnerable and problematic users, suggesting that automatic engagement patterns may be more characteristic of higher-risk users. At the same time, self-control consistently showed negative associations with PSVU across profiles. Network analyses further indicated that self-control was connected to both motivational variables and PSVU, particularly among problematic users. This finding is consistent with previous studies linking self-regulatory capacity to problematic digital media use and highlights the potential importance of self-control in understanding individual differences in PSVU-related risk [15, 44].

It should also be noted that stress and stress-relief motivation are conceptually related constructs. However, examination of the edge-weight matrices revealed no direct regularized association between these variables in any of the three networks (see Appendix S5), suggesting limited empirical redundancy within the present models. More broadly, some motivational constructs included in the network, particularly escapism, stress relief, habitual use, and PSVU, may share partial conceptual overlap and may partly contribute to some of the stronger associations observed between motivational variables and PSVU. Although GGMs estimate unique associations after controlling for shared variance among variables, such overlap may still influence edge weights among conceptually related nodes. Therefore, these associations should be interpreted with appropriate caution.

Taken together, these findings suggest that motivational factors may distinguish higher-risk users from adaptive users more clearly than some demographic or usage characteristics. This observation further supports the latent profile solution by indicating that the identified profiles differ not only in PSVU severity but also in the psychological motivations underlying short video engagement.

Emotion–personality–motivation–behavior network associated with PSVU

Because the present networks were estimated using cross-sectional GGMs, the observed edges should be interpreted as undirected partial correlations reflecting statistical associations among variables rather than directional or causal pathways [29]. Importantly, the structure and strength of these psychological networks varied across user profiles, suggesting that psychological variables may exhibit different patterns of organization across PSVU risk levels.

The network analyses revealed distinct patterns of associations among emotional, personality, motivational, and behavioral factors across the three PSVU profiles. Consistent with the I-PACE [13, 14] framework, emotional distress variables, personality characteristics, and usage motivations formed interconnected clusters within all networks. In the adaptive-user network, PSVU was primarily connected with self-control, cognitive flexibility, and neuroticism. This pattern is consistent with previous research suggesting that self-regulatory capacities and emotional stability may play important roles in maintaining balanced social media use [13, 46]. Individuals with stronger self-control and greater cognitive flexibility may be better able to regulate their short video consumption and adaptively respond to online stimuli.

In contrast, motivational variables demonstrated stronger associations with PSVU in the vulnerable and problematic-user networks. In particular, escapism, habitual use, and stress-relief motivation were more strongly connected with PSVU among problematic users. Previous studies have similarly reported that individuals may engage with digital media platforms to temporarily disengage from negative emotions or everyday stressors [12, 45, 89]. The present findings extend this literature by indicating that these motivations occupy a more central position within the broader psychological network among individuals with higher levels of PSVU. Consistent with previous research on behavioral addictions and problematic internet use [7, 13], self-control showed negative associations with PSVU and remained connected to motivational variables in the vulnerable and problematic networks. This pattern suggests that regulatory resources may be relevant to understanding why similar motivations are linked to different levels of problematic engagement across users.

The network comparison analyses further indicated that PSVU showed stronger connections with motivational variables in the problematic-user profile than in the adaptive-user profile. However, the practical significance of these differences should be interpreted cautiously. Although several edge-weight differences reached statistical significance, some numerical differences were relatively modest. Therefore, these findings should be viewed as evidence of differences in network organization rather than evidence of large practical differences in the strength of specific associations. Centrality analyses highlighted depression, stress, and neuroticism as relatively well-connected nodes across the three networks, particularly in the vulnerable and problematic-user networks. Previous studies have consistently linked neuroticism, stress, and emotional distress to problematic social media use and PSVU ([3, 17, 36, 37]. Nevertheless, centrality indices identify variables that are strongly connected within the observed network structure and should not be interpreted as evidence of causal influence.

Overall, the network findings suggest that emotional, personality, and motivational factors are differentially connected with PSVU across user profiles. Rather than supporting a single explanatory mechanism, the results indicate that the relative importance of these factors may vary according to the severity of PSVU. These findings complement the latent profile results by illustrating how psychological variables are interconnected within different user groups, thereby providing a more nuanced understanding of PSVU heterogeneity. Additionally, some of the psychological correlates of PSVU identified in the present study are consistent with findings from the substance use disorder literature, such as emotional distress, neuroticism, maladaptive coping-related motivations, and reduced self-regulatory capacities, which have been repeatedly linked to increased vulnerability to substance use disorders [42, 71, 83]. Although PSVU does not involve the ingestion of psychoactive substances, the present findings provide tentative support for the view that certain psychological vulnerabilities may be shared across behavioral and substance-related addictions. Nevertheless, these similarities should be interpreted cautiously, as the current findings do not imply that PSVU is equivalent to substance use disorders, and important distinctions between behavioral and substance-related addictions remain to be clarified.

Limitations and future directions

Several limitations should be acknowledged. First, although the network analyses provide insights into associations among psychological variables across PSVU profiles, the cross-sectional design precludes causal inference and limits conclusions regarding the temporal stability of the identified profiles. It remains unclear whether these profiles represent relatively stable user characteristics or more dynamic states that fluctuate over time. Longitudinal studies are needed to examine both transitions between PSVU profiles and changes in psychological network structures over time. In addition, network analysis is designed to characterize association patterns rather than the relative contribution of individual variables to PSVU. Future studies may complement network approaches with predictive methods to further examine the relative contribution of specific psychological variables to PSVU.

Second, all measures relied on self-report data and were collected within a single survey session, which may be subject to social desirability, recall bias, and common-method variance. As a result, some observed associations may have been inflated, particularly among conceptually related emotional variables. Furthermore, the stronger associations observed in the vulnerable and problematic-user networks may partly reflect broader negative affectivity or shared distress variance rather than PSVU-specific mechanisms. Similarly, higher symptom severity in the problematic-user profile may have amplified associations among psychological variables, contributing to stronger network connectivity. In addition, cognitive flexibility was assessed using a self-report measure, which may not fully reflect actual cognitive performance, particularly among individuals with limited awareness of cognitive changes. Future studies should incorporate multimethod assessments, including objective behavioral measures of cognitive flexibility, and longitudinal designs to evaluate these alternative explanations.

Third, the sample consisted primarily of young adults from China, with a relatively high proportion of students, which may limit the generalizability of the findings to other age groups, occupational backgrounds, and cultural contexts. In addition, data were collected over a two-week period in late April–May, which may have coincided with mid-semester academic pressures and could have modestly influenced reported levels of stress and PSVU. Future research should examine more diverse populations and conduct cross-cultural comparisons to determine whether the identified profile and network patterns can be replicated across different sociocultural settings.

Fourth, the predictability (R²) of the PSVU node was relatively modest across the three networks, suggesting that the included psychological variables explained only part of the variance in PSVU. Although this may partly reflect the regularization characteristics of GGMs [29, 31], it also indicates that additional psychological, behavioral, and contextual factors may contribute to PSVU. Future studies could incorporate a broader range of variables to provide a more comprehensive understanding of PSVU-related psychological networks.

Fifth, although the PSVUS demonstrated good psychometric properties, it remains a self-developed instrument and was primarily validated against a Chinese PSMU measure. Future studies should further evaluate its cross-cultural applicability and examine its convergent validity with internationally established measures, such as the Bergen Social Media Addiction Scale [1] and the Social Media Disorder Scale [82]. The PSVUS also did not assess whether participants experienced their short video use as ego-syntonic or ego-dystonic, despite potential differences in whether users perceive their behavior as acceptable or as distressing and difficult to control. Future studies should assess users’ subjective appraisal, perceived impairment, and motivation to change. The present study did not assess participants’ substance use or other potentially addictive behaviors, such as problematic gaming or gambling, and therefore could not directly examine whether the observed psychological correlates were shared with, or specific to, PSVU.

Finally, as with all person-centered approaches, latent profile solutions depend partly on model selection criteria, and alternative classifications may also be plausible. Furthermore, the identified profiles primarily reflected differences in PSVU severity, which may limit conclusions regarding qualitatively distinct PSVU subtypes. Replication studies are therefore needed to confirm the robustness of the identified profiles.

Conclusion

This study identified three PSVU risk profiles and revealed their associated psychological patterns. The profiles differed primarily in PSVU severity, while also showing variation in the network organization of emotional, personality, and motivational factors connected with PSVU. Specifically, the pattern of associations among these variables varied across profiles, suggesting that the psychological correlates of PSVU may differ according to risk level. These findings extend the literature by illustrating both graded differences in PSVU risk and profile-specific patterns of psychological associations. More broadly, the results suggest that prevention and intervention efforts may benefit from considering PSVU severity alongside the psychological factors most strongly associated with PSVU within each profile.

Authors’ contributions

Chen Yuzhou: Conceptualization, Methodology, Software, Investigation, Formal analysis, Writing - Original Draft, Writing - Review & Editing, Visualization. Li Jiayu: Investigation, Methodology, Resources, Validation. Shan haidi: Methodology, Data Curation, Writing - Review & Editing. Zhong Na: Supervision, Project administration, Funding acquisition. Zhao Min: Supervision, Project administration, Funding acquisition.

Funding

This work was supported by STI2030-major projects (2022ZD0211100, 2021ZD0202105); National Nature Science Foundation (82171485, 82130041, 82171483, 82201650, 82201649); Shanghai Municipal Health Commission Talent Project (GWVI-11.2-XD26, 2022YQ048); Shanghai Clinical Research Center for Mental Health (19MC1911100); Shanghai Rising-star Cultivation Program (22YF1439200); Shanghai Mental Health Center Hospital Project (2022zd01); Shanghai Mental Health Center Flying Talent Project (2022-FX-01); Shanghai Shenkang Hospital Development Center (SHDC2020CR3045B); Science Foundation of Hebei Normal University (No. L2026B42), and Science Research Project of Hebei Education Department (No. QN2026828).

Data availability

The data supporting the findings of this study are available at https://doi.org/10.17605/OSF.IO/5MC7A.

Declarations

Ethics approval and consent to participate

The study was carried out following the Declaration of Helsinki and was approved by the Ethics Committee of Shanghai Mental Health Center (Approval No. 2023-63). All participants provided written informed consent prior to participating in the survey.

Competing interests

The authors declare no competing interests.