The Power of Interaction: Fan Growth in Livestreaming E-Commerce

Abstract

Fan growth serves as a critical performance indicator for the sustainable development of livestreaming e-commerce (LSE). However, existing research has paid limited attention to this topic. This study investigates the unique interactive advantages of LSE over traditional e-commerce by examining how interactivity drives fan growth through the mediating role of user retention and the moderating role of anchors’ facial attractiveness. To conduct the analysis, real-time data were collected from 1472 livestreaming sessions on Douyin, China’s leading LSE platform, between January and March 2023, using Python-based (3.12.7) web scraping and third-party data sources. This study operationalizes key variables through text sentiment analysis and image recognition techniques. Empirical analyses are performed using ordinary least squares (OLS) regression with robust standard errors, propensity score matching (PSM), and sensitivity analysis to ensure robustness. The results reveal the following: (1) Interactivity has a significant positive effect on fan growth. (2) User retention partially mediates the relationship between interactivity and fan growth. (3) There is a substitution effect between anchors’ facial attractiveness and interactivity in enhancing user retention, highlighting the substitution relationship between anchors’ personal characteristics and livestreaming room attributes. This research advances the understanding of interactivity’s mechanisms in LSE and, notably, is among the first to explore the marketing implications of anchors’ facial attractiveness in this context. The findings offer valuable insights for both academic research and managerial practice in the evolving livestreaming commerce landscape.

1. Introduction

With the rapid advancement of information technology and the widespread adoption of mobile devices, livestreaming e-commerce (LSE) on social networking platforms such as Douyin Live and Sina Live has experienced remarkable growth. According to a public report, the number of enterprises in China’s LSE industry reached 18,700 in 2022, representing a year-on-year increase of 17.61%. During the 2022 Double Eleven shopping festival—China’s equivalent of Black Friday—leading platforms like Douyin and Taobao Live attracted nearly 500 million users and generated sales totaling CNY 181.4 billion. Given the growing popularity of LSE and its key role in the development of e-commerce, many researchers have investigated this phenomenon, with a primary focus on the antecedents of LSE performance, such as customer engagement, sales, conversion rates, gifting behavior, return rates, and impulse buying [1,2,3,4,5,6,7]. Studies have found that anchor characteristics, product types, social contexts, perceived experience, perceived trust, scarcity, technological fluency, and promotional strategies are key factors influencing LSE performance [8,9,10,11,12,13,14,15]. Despite numerous findings, these studies have mostly focused on short-term performance in LSE, neglecting research on long-term performance. In LSE, anchors with larger fan bases wield greater attraction, influence, and premium-generating capabilities, thereby creating substantial business value for brands and livestreaming managers [16,17]. However, despite the growing popularity of LSE, the average fan conversion rate remains below 1.5%, posing challenges for platforms and managers in attracting and retaining fans. Additionally, a concerning trend has emerged where top-tier anchors dominate approximately 80% of livestreaming traffic. This dominance creates a skewed distribution of traffic, increasingly marginalizing smaller anchors and potentially pushing them off the platform [18]. Unfortunately, no comprehensive theory currently explains why fan growth varies significantly across livestreaming rooms.

Interactivity is a crucial feature of LSE. In this context, anchors introduce and promote products to users in real time through the camera. Users can ask personalized questions regarding product quality, after-sales service, logistics, and more, while also engaging with other users through bullet comments and other interactive features [19,20], thereby forming a dynamic, interactive community that transcends physical boundaries. This process greatly enhances users’ sense of presence and product experience, creating favorable conditions for positive behavioral decisions [21,22]. According to the Interactive Ritual Chain Theory (IRCT), as users watch livestreams and interact with both the anchor and other viewers, they form an interactive ritual chain that fosters emotional attachment and sustained engagement in livestreaming activities [23]. However, interactivity may also yield adverse effects, such as user attrition and the disruption of online community relationships [24,25], underscoring the complex impact of interactivity on user behavior and marketing performance. Therefore, it is necessary to empirically investigate the role of interactivity in LSE to design more effective interaction strategies that support its sustainable growth. Meanwhile, with increasing competition across the livestreaming industry, user attention has become highly fragmented. Standing out among a multitude of livestreams and maintaining high user retention have become crucial to LSE’s success. Prior research suggests that when users perceive interactions as informative, entertaining, and practical, they tend to engage more actively, devote more time to in-depth exchanges, and ultimately enter an immersive state that promotes positive behaviors [26,27]. Thus, whether interactivity influences fan growth by enhancing user retention remains an important question worthy of further exploration.

Furthermore, we hypothesize that an anchor’s facial attractiveness may moderate the relationship between interactivity in LSE and user retention. According to Dual Systems Theory (DST), observable user features in online environments function as important heuristic cues that shape consumer decision-making [28]. These cues often complement content-based information, assisting consumers in evaluating the credibility of the source [29]. In the context of LSE, an anchor’s facial attractiveness serves as the first visual cue encountered in a livestreaming room, playing a crucial role in shaping users’ initial impressions of the marketing content. Previous research suggests that individuals with higher facial attractiveness are perceived as having greater social skills compared to their less attractive counterparts and are generally more effective in domains such as product marketing [30]. Therefore, it is plausible that the impact of interactivity on user retention in LSE may vary depending on the anchor’s facial attractiveness. However, existing studies have largely overlooked the potential interaction effects between livestreaming room characteristics (e.g., interactivity) and anchor-specific attributes, thereby limiting our understanding of their combined influence on user behavior. These considerations lead to the following research questions:

RQ1: How does interactivity influence fan growth in LSE?

RQ2: What role does user retention play in mediating the relationship between interactivity and fan growth?

RQ3: Does an anchor’s facial attractiveness moderate the relationship between interactivity and user retention?

The contributions of this study are reflected in the following three aspects: (1) This study reveals the significant impact of interactivity on fan growth and uncovers its underlying mechanisms in the context of LSE, thereby broadening the research scope of LSE performance. Furthermore, it identifies a substitution effect between livestreaming room characteristics and anchor characteristics in influencing user retention. (2) By employing objective data, the empirical analysis validates the applicability of Interactive Ritual Chain Theory (IRCT) and Dual Systems Theory (DST) within the LSE context, enriching the theoretical foundation of the field and offering new directions for future research. (3) This study offers practical implications for the operation and management of LSE, particularly in the areas of anchor selection and training.

The remainder of this paper is structured as follows: Section 2 reviews the relevant literature on LSE, Interactive Ritual Chain Theory (IRCT), and Dual Systems Theory (DST). Section 3 presents the main research hypotheses. Section 4 describes the study design, data collection procedures, and variable measurements. Section 5 provides descriptive statistics and reports the empirical findings. Finally, Section 6 discusses the results, draws conclusions, offers marketing management recommendations, and outlines directions for future research.

2. Literature Review

2.1. Performance Research in LSE

As an innovative online sales model, LSE integrates image transmission technology with social interaction, fundamentally transforming how people shop and engage with one another [6,31]. Unlike traditional e-commerce, LSE enables real-time interactions between anchors and users, as well as among users themselves, offering richer information and more engaging, immersive experiences. This unique feature has attracted a growing number of merchants and consumers to participate in LSE [7,32], thereby prompting increasing interest in its performance metrics. Among these, consumer purchases and reward-giving behaviors are the most frequently studied indicators [2,3,5,17]. Previous studies have identified several critical factors that influence consumer purchases and sales outcomes in LSE, including the variety of products offered [33], the level of social interaction [34], the presence of informational cues [35], and anchor characteristics [36]. From the perspective of dual attachment, Zhang et al. [37] found that both bond-based and identity-based social attachments significantly drive users’ reward-giving behavior. However, due to the real-time nature of LSE, researchers have also turned their attention to its potential downsides, such as compulsive purchasing and order cancellations. Sun and Bao [19] demonstrated that fear of missing out and social influence play a significant role in driving compulsive purchases during livestreaming sessions. Similarly, Wang et al. [38], drawing on emotional contagion theory, identified group emotion as a key factor contributing to consumers’ impulse-driven order cancellations in LSE.

Despite the growing body of research on LSE performance, the critical performance indicator of fan growth has been largely overlooked. In the context of LSE, fan growth typically refers to the process by which ordinary users are converted into loyal followers of a particular brand, product, or anchor through targeted strategies and activities [16]. Zhao et al. [17] noted that anchors with larger fan bases tend to exhibit greater appeal, influence, and monetization potential, thereby enhancing the operational value of both the brand and the livestreaming room. Prior research has also shown that fan growth can serve as a leading indicator of users’ purchase intentions in livestreaming settings [18]. However, few studies have directly investigated fan growth within the LSE context, and its underlying drivers remain insufficiently understood. This study seeks to address this gap by examining the determinants and mechanisms of fan growth through the lens of interactivity, thereby offering theoretical insights to inform the effective operation and strategic management of LSE platforms.

2.2. Interactivity in LSE

Interactivity is a core advantage and defining feature of LSE, a consensus widely recognized in both academic research and industry practice [24]. In LSE, anchors engage with users in real time through product demonstrations, topical discussions, and timely responses, fostering a positive and dynamic environment. Such engagement enhances consumers’ perceived value and significantly influences their behavior [19]. Existing research on interactivity in LSE primarily centers on three key areas: First, scholars have examined how interactivity affects consumers’ purchase intentions and participatory behaviors through various psychological mechanisms and moderating factors. For instance, interactivity has been found to increase perceived usefulness, reduce perceived risk, and strengthen psychological connections with consumers, thereby promoting purchasing behavior [15,34,39,40]. Additionally, interactivity is closely associated with consumers’ impulse-buying tendencies [41]. Second, researchers have investigated the role of interactivity in value co-creation within online communities. Studies indicate that user participation in value co-creation is influenced by interactive behaviors, with the transition from human–computer interaction to human–human interaction largely driven by factors such as community incentives, social enhancement, and mutual co-creation [42]. Third, studies have explored the constituent elements of interactivity across different contexts. Kang et al. [24] conceptualized interactivity as comprising two dimensions: responsiveness and personalization. Responsiveness refers to the speed at which anchors respond to user input, while personalization pertains to the degree to which the information or service aligns with users’ specific needs. Bonner [27] further characterized interactivity through bi-directionality, participation, and joint problem-solving, highlighting the intensity and richness of user engagement. Moreover, the frequency and quality of interaction have been identified as critical indicators of interactive behavior [43].

Although existing studies have examined the value of interactivity in LSE, few have specifically investigated its effect on fan growth. Moreover, prior research has primarily focused on users’ behavioral intentions based on relatively subjective factors, such as psychological perceptions [12,14], while overlooking the mechanisms through which more objective factors—such as user retention—influence actual behavioral outcomes. In particular, limited attention has been given to the complex interplay between livestreaming room interactivity and the personal characteristics of anchors in shaping user behavior. Additionally, most existing studies rely on survey-based data, whereas empirical analyses utilizing real-time behavioral data in the LSE context remain scarce. Addressing these research gaps, this study proposes a conceptual framework of “interactivity–user retention–fan growth” and explores the boundary conditions under which interactivity affects user behavior. This approach not only enhances the current understanding of LSE mechanisms but also contributes to the development of more robust theoretical models and methodological tools in this domain.

2.3. Interactive Ritual Chain Theory

Interactive Ritual Chain Theory (IRCT), first proposed by Collins, posits that interaction is fundamentally ritualistic in nature and comprises four core elements: (1) physical co-presence, (2) boundaries that exclude outsiders, (3) a shared focus of attention, and (4) a common emotional experience and behavioral synchronization. Participants gather within a defined space to engage in emotionally charged symbolic activities, which generate or reinforce collective identity and emotional energy, thereby increasing their motivation to continue interacting with one another [44]. As a context-sensitive theory, IRCT has been widely applied in research on community relations and group dynamics. It elucidates how members of a group or community draw upon emotional energy to foster social interactions and cultivate a sense of belonging and shared identity [45,46,47]. For instance, in virtual environments such as online communities, interaction rituals frequently emerge, eliciting collective enthusiasm, reinforcing group identity, and generating emotional energy that helps sustain social bonds [48]. Recent theoretical advancements have extended IRCT into the realm of LSE. In the LSE context, users interact with anchors and fellow viewers in real time, forming virtual community ties that foster emotional resonance and a sense of belonging, ultimately shaping user behavior. Meng et al. [23] found that as users engage with livestream content and participate in real-time interactions, they form interactive ritual chains that generate emotional energy and foster a sense of cohesion. This emotional engagement, in turn, promotes sustained participation and prolonged viewership in livestreaming sessions.

Nevertheless, most of the existing research on IRCT remains at a theoretical and qualitative exploratory stage, with limited empirical studies validating its explanatory power. Within the context of LSE, users collectively form a virtual community with a sense of physical co-presence, naturally creating barriers that exclude non-participating outsiders. Additionally, anchors, products, and other livestream features serve as shared focal points of attention, establishing the conditions for interaction and fostering shared emotional experiences and synchronized behaviors. Building on these contextual features, this study investigates the effect of interactivity on fan growth through the lens of IRCT, thereby extending the empirical application of IRCT in the LSE domain.

2.4. Dual Systems Theory of Decision-Making

Dual Systems Theory (DST) delineates two cognitive systems engaged in human judgment and decision-making processes. The intuitive–heuristic system operates based on intuition, processes information rapidly, provides automatic responses, and typically considers only partial information during evaluation. This system is particularly vulnerable to various cognitive biases [49]. In contrast, the rational–analytical system functions through deliberate effort, processes information systematically, and employs algorithmic reasoning to solve problems. It operates more slowly, requires greater cognitive resources, and can effectively reduce decision-making biases [50]. While the heuristic system tends to rely on contextual cues and stereotypes—focusing more on outcomes than on the reasoning process—the analytical system emphasizes logical rules and maintains awareness of both reasoning processes and results [51].

As DST continues to evolve, it has been widely applied in consumer behavior research [52]. Recent studies have identified online user characteristics as important heuristic cues influencing consumer decision-making [28]. In the context of LSE, an anchor’s physical appearance—particularly their facial attractiveness—serves as one of the most salient and intuitive online features. It provides an immediate visual impression and acts as a powerful heuristic cue for viewers. The effect of facial attractiveness on consumer decision-making is well documented in fields such as behavioral economics [53,54]. However, in the relatively nascent domain of LSE research, it remains unclear whether anchors’ facial attractiveness, as a heuristic cue, significantly affects user behavior. In particular, the potential interaction between this easily perceived cue and livestreaming room characteristics has yet to be empirically explored. To address this gap, the present study incorporates anchors’ facial attractiveness into a moderation model, thereby enriching the theoretical framework of marketing management within the LSE context.

3. Research Hypothesis

3.1. Interactivity and Fan Growth

In this study, we argue that interactivity in LSE plays a critical role in driving fan growth. Prior research grounded in Interactive Ritual Chain Theory (IRCT) suggests that ritualized interactions frequently emerge in virtual communities, evoking collective euphoria, group identity, and emotional energy, thereby contributing to the stability and maintenance of social relationships [48]. In the context of LSE, the livestreaming room functions as a virtual community in which users engage in real-time interactions with both the anchor and other users. Such interactions foster quasi-social connections that generate emotional energy and cultivate a sense of belonging, which, in turn, influences user behavior. Meng et al. [23] found that when users actively engage with a livestream and interact with the anchor and other participants, they form chains of interactive rituals that produce emotional energy and a shared sense of togetherness. This emotional dynamic encourages users to repeatedly participate and maintain long-term viewing behavior. Moreover, real-time interaction contributes to a perceived sense of “virtual physical presence,” or spatial co-presence, within the livestreaming environment. This perceived presence enhances users’ feelings of transparency, engagement, trust, and enjoyment—factors that have been shown to increase user satisfaction and platform loyalty [15,34,40]. Hu and Chaudhry [55] further demonstrated that interactions between anchors and users often resemble quasi-social relationships, wherein both parties develop a sense of intimacy and view each other as “real friends.” This perceived intimacy helps reduce users’ uncertainty and mitigates psychological barriers [42], ultimately increasing the likelihood that users will become loyal fans.

In LSE, the frequency of interaction reflects the intensity of engagement between anchors and users. Higher interaction frequency can enhance users’ perceived role identity and foster stronger community ties [40]. In addition, frequent interactions allow users to obtain more accurate and detailed information about products and services, thereby elevating their perception of the value provided during the livestream [42]. Prior studies have shown that regular contact between sales representatives and customers enhances customer satisfaction and commercial performance [56]. Therefore, frequent interactions in LSE may contribute to the development of interpersonal emotional bonds, foster a sense of belonging, and thus promote fan growth. Furthermore, the quality of interaction captures the effectiveness of anchor-led engagements within the livestreaming room. High-quality interactions are more likely to induce immersive engagement and deepen the relationship between users and anchors [31,40]. Based on these insights, this study proposes the following hypotheses:

H1a. 

Interaction frequency significantly and positively affects fan growth.

H1b. 

Interaction quality significantly and positively affects fan growth.

3.2. The Mediating Role of User Retention

IRCT posits that an interaction ritual involves individuals jointly engaging in emotionally driven symbolic activities, which, in turn, generate or reinforce collective identity and emotional energy. This emotional energy strengthens individuals’ willingness to continue interacting, thereby sustaining quasi-social relationships and fostering positive emotional experiences characterized by enjoyment and pleasure [23]. Prior research further suggests that when interactions are perceived as informative, entertaining, and compelling, individuals are more inclined to engage actively, devote more time to communication, and ultimately enter an immersive state [27,57]. In support of this, Pelet et al. [58] demonstrated that immersive experiences significantly enhance visit frequency and browsing duration on websites. Building on these findings, this study argues that higher interaction frequency and superior interaction quality within a livestreaming room increase users’ willingness to remain engaged, deepen their interaction with the anchor, and focus more intently on the promoted products or services. Furthermore, prior research has confirmed that user retention plays a significant positive role in shaping consumers’ purchase intentions and behaviors across diverse digital contexts [59]. Based on this theoretical and empirical foundation, this study proposes the following hypotheses:

H2a. 

User retention mediates the relationship between interaction frequency and fan growth.

H2b. 

User retention mediates the relationship between interaction quality and fan growth.

3.3. The Moderating Role of Anchors’ Facial Attractiveness

Beauty influences perceptions and shapes judgments and decisions across a wide range of domains [53]. In traditional marketing, consumers often rely on observable online cues—such as avatars—as indicators of source credibility. These visual cues, alongside textual information, attract attention and help shape online impressions [60]. Attractive individuals are frequently perceived as possessing socially desirable personality traits and higher competence. This “beauty premium” has been widely documented across various fields, including the labor market, advertising, marketing, and online socialization [53,54]. In the context of LSE, an anchor’s facial attractiveness functions as an intuitive and salient heuristic cue that can dominate users’ information processing and significantly influence their viewing behavior. Anchors with higher facial attractiveness generally achieve higher levels of user retention in livestreaming sessions.

However, a substitution effect may exist between an anchor’s facial attractiveness and livestream interactivity. According to Dual Systems Theory (DST), people form judgments about social competence, age, health, and service quality based on facial cues, which contribute to more favorable first impressions [61]. For instance, a visually appealing personal photo can facilitate positive evaluations and increase customer attachment to the service provider [62]. Consequently, higher facial attractiveness may enhance perceived value and extend user retention time. Conversely, research has also acknowledged the “dark side” of attractiveness stereotypes [53], particularly in LSE environments, where interaction and social engagement are vital. When an anchor’s facial attractiveness is lower, users may interpret frequent and high-quality interactions as signals of competence and professionalism. These interactions can then compensate for lower attractiveness, foster immersive experiences, and more strongly influence user retention. Based on these insights, this study hypothesizes the following:

H3a. 

In a livestreaming room where the anchor’s facial attractiveness is higher, the promotion effect of interaction frequency on user retention will be significantly reduced.

H3b. 

In a livestreaming room where the anchor’s facial attractiveness is higher, the promotion effect of interaction quality on user retention will be significantly reduced.

Figure 1 shows the research model of this paper.

4. Research Design

4.1. Data Collection Platform

The data for this study were collected from Douyin, the largest and fastest-growing LSE platform in Mainland China (the domestic counterpart of TikTok) [38]. As of 2024, Douyin hosts over 700 million daily active users and offers a wide array of products, including fresh produce, apparel and accessories, daily necessities, and electronic goods. Due to its widespread popularity, Douyin’s LSE ecosystem has been the focus of extensive research in authoritative academic literature [8,16,34]. The platform’s highly interactive livestreaming rooms facilitate real-time communication between anchors and users, creating a dynamic environment that is well suited for the objectives of this study. Moreover, Douyin’s social features support the random sampling of livestreaming sessions through the “Live Square,” thereby enhancing the external validity of the research. The platform also provides an official data interface, allowing researchers to retrieve data programmatically using a Python-based API.

4.2. Data Collection Process and Processing

Data Collection: Data collection was conducted through a combination of crawler programs and manual efforts by research assistants. A total of twelve research assistants were recruited for this study. Each assistant was instructed to enter a randomly selected livestreaming room via the Douyin Live Square once every hour between 9:00 a.m. and 9:00 p.m. daily, from January to March 2023, excluding lunch breaks and the Chinese New Year period to avoid holiday-related anomalies. During each session, the assistants recorded a video that captured a complete product presentation delivered by the anchor. Simultaneously, they used a designated Python-based program to retrieve livestreaming room links and collect multiple types of data, including bullet comments, product information, anchor attributes, and user statistics. After completing each session, the research assistants downloaded and compiled additional structured data from a third-party analytics platform (www.huitun.com), which provided key performance metrics such as livestream sales, average user engagement time, broadcast duration, and viewer counts.

Data Processing: We processed unstructured video and text data. Specifically, video processing was carried out using OpenCV (4.12.0) (Open Source Computer Vision Library: an open-source computer vision and machine learning software library; it provides a wealth of image processing and computer vision algorithms that can be applied to a variety of fields, such as machine vision, autonomous driving, medical image processing, and so on) to capture keyframes (specific frames extracted from the video at equal time intervals or frame intervals that represent key information or change points in the video content), which were then uploaded to the Face++ (Face++ is an open AI platform that provides developers with AI capabilities such as face recognition, portrait processing, human body recognition, text recognition, and image recognition) image processing interface to analyze facial keypoints and assess the attractiveness of the anchor. Additionally, bullet comments’ text was analyzed for sentiment using the Baidu Intelligent Platform API. The raw dataset initially contained 1,020,5902 bullet comments from 1891 livestreaming rooms. However, after excluding records with missing data on host facial recognition, the final empirical dataset used for this study comprised 1472 livestreaming rooms. The complete data flow is shown in Figure 2.

4.3. Variable Measurement

Explanatory Variables: Following the framework of Hawkins et al. [43], we categorize interactivity into two dimensions: interaction frequency and interaction quality. Interaction frequency refers to the intensity of interactions between the anchor and users, as well as among the users themselves, within a specific timeframe in the livestreaming room. We use the number of bullet comments captured per minute by the crawler program as a proxy measure; a higher frequency of bullet comments per unit of time indicates increased interaction frequency. Interaction quality mainly refers to the quality and depth of user interactions within the livestreaming room. Compared to interaction frequency, interaction quality places more emphasis on the substance and effectiveness of user interactions. We measure interaction quality through sentiment analysis of bullet comments’ text. Specifically, we use the sentiment analysis API from Baidu’s open AI platform, which can assess the emotional tendency of text containing subjective information, effectively identifying the user emotions in each bullet comment and categorizing them as positive, neutral, or negative. We calculate the ratio of positive comments to total comments in the livestreaming room as a proxy measure of interaction quality.

Explained Variable: Fan growth is measured by the number of new fans gained after each livestream, reflecting the livestream’s appeal to new users. In addition, in the robustness test, we measure fan growth through the fan conversion rate in the livestreaming room.

Mediating Variable: User retention refers to the average duration of each user’s stay in the livestreaming room, serving as a key indicator of the content’s and merchandise’s attractiveness. It is calculated by dividing the total duration of user visits by the total number of visitors in the livestreaming room.

Moderating Variable: The anchor’s facial attractiveness is assessed using the Face++ facial attribute detection API. Face++ is a reputable provider of facial recognition and body detection services and has been used in multiple studies for tasks such as facial attractiveness, age classification, and emotion recognition [53]. We use the captured keyframes of the anchor’s face as inputs. Face++ provides two attractiveness scores for each image, ranging from 0 to 100, representing perspectives from both female and male users. We calculate the average of these scores to determine the anchor’s facial attractiveness, following the methodology of Ma et al. [53] and Peng et al. [61]

Control Variables: Based on existing research, other factors that may influence audience behavior are included in the model [63,64,65]. These factors include anchor characteristics (reputation, gender, age, speed), product characteristics (commodity, price), and livestreaming room characteristics (duration, type, recommendation, likes). To mitigate heteroskedasticity and reduce the impact of outliers, all continuous variables are log-transformed. The measurements of the control variables are shown in

Table 1. Variable descriptions.

Variable	Description	References
Reputation	The anchor’s electronic word-of-mouth score (0–5)	[63,64,65]
Gender	The gender of the anchor (1 for female and 0 for male)
Age	The age of the anchor
Speed	The number of Chinese words spoken by the anchor every minute
Commodity	The number of products sold in the LSE room
Price	The average price of products sold in the LSE room
Duration	The LSE section duration (minutes)
Type	With rooms hosted by independent anchors marked as 1 and those hosted by brand-affiliated anchors marked as 0
Recommendation	Percentage of viewers entering the LSE room through the recommendations of Douyin video (0–1)
Likes	The number of likes in the LSE room

.

4.4. Model Setting

To test the research hypotheses, the following models are specified in this study:

$${Fan growth}_i={\beta }_0+{\beta }_1{interaction frequency}_i+\gamma {Control}_i+{\epsilon }_i$$

(1)

$${Fan growth}_i=a_0+a_1{interaction quality}_i+\gamma {Control}_i+{\epsilon }_i$$

(2)

where ${Fan growth}_i$ is the dependent variable, while ${interaction frequency}_i$ and ${interaction quality}_i$ are the independent variables. Considering the cross-sectional nature of the data and potential issues of heteroskedasticity, we adopt the ordinary least squares (OLS) regression model with robust standard errors. OLS is a widely used method for estimating linear relationships between variables, and the use of robust standard errors helps correct potential heteroskedasticity that may bias the standard errors, leading to more reliable inference [8,38]. Given our large sample size and clearly defined independent variables, OLS with robust standard errors provides an appropriate baseline estimation strategy for evaluating the relationships among interactivity, user retention, and fan growth.

To account for potential endogeneity and omitted-variable bias, particularly concerning selection into high- or low-interactivity groups, PSM is employed as a robustness check. A binary treatment indicator is constructed, assigning treatment status to livestreaming rooms with above-mean interactivity (frequency and quality). The propensity scores are estimated using a logit model based on pre-treatment covariates (e.g., duration, price, commodity, type, recommendation, likes). Matching is performed using 1-to-1 nearest-neighbor matching without replacement to ensure high match quality. Meanwhile, sensitivity analysis is employed in this study to address concerns related to potential endogeneity, particularly from omitted-variable bias. In LSE contexts, unobserved factors—such as anchor charisma, real-time marketing efforts, or backstage operational support—might simultaneously influence both interactivity and fan growth, leading to biased estimates. This method allows us to quantify how strong an unobserved confounder would need to be, relative to the observed covariates, in order to overturn our results [66].

5. Empirical Analysis

5.1. Descriptive Statistics and Correlation Analysis

Table 2. Descriptive statistics.

Variable	Obs	Mean	Std. Dev.	Min	Max
Fan growth	1472	6.82	1.964	0	12.145
Interaction quality	1472	0.666	0.158	0	1
Interaction frequency	1472	3.508	1.863	0.082	10.375
Duration	1472	5.399	0.73	3.086	8.371
Reputation	1472	4.773	0.233	3.02	5
Commodity	1472	3.69	0.9	0.693	6.648
Price	1472	4.09	1.063	0.262	8.993
Likes	1472	10.575	2.199	2.996	18.858
Recommendation	1472	0.083	0.081	0	1
Type	1472	0.454	0.498	0	1
Gender	1472	0.81	0.393	0	1
Age	1472	26.168	7.135	3	64
Speed	1472	5.634	0.4	0.742	6.679
Facial attractiveness	1472	79.838	7.935	42.838	97.399
Retention	1472	4.294	0.754	2.197	7.051

and

Table 3. Correlation analysis.

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)	(14)	(15)
(1) Fan growth	1.000
(2) Interaction quality	0.308	1.000
(3) Interaction frequency	0.656	0.142	1.000
(4) Duration	0.326	0.034	−0.107	1.000
(5) Reputation	0.146	0.000	0.154	0.102	1.000
(6) Commodity	0.208	−0.034	0.192	0.143	0.166	1.000
(7) Price	0.037	0.027	0.050	0.153	0.058	−0.007	1.000
(8) Likes	0.704	0.096	0.724	0.206	0.200	0.233	0.051	1.000
(9) Recommendation	−0.115	−0.041	−0.081	−0.069	0.048	−0.049	0.016	−0.108	1.000
(10) Type	0.050	−0.138	0.197	−0.197	0.098	0.078	−0.017	0.197	0.120	1.000
(11) Gender	−0.092	0.013	−0.127	0.033	−0.015	−0.046	−0.019	−0.185	0.038	−0.076	1.000
(12) Age	0.046	−0.090	0.037	−0.049	0.010	0.059	−0.033	0.124	−0.008	0.074	−0.072	1.000
(13) Speed	0.130	0.071	0.021	0.101	−0.017	0.004	0.017	−0.001	−0.024	−0.067	0.031	−0.049	1.000
(14) Facial attractiveness	−0.163	−0.555	−0.088	0.016	0.026	0.074	0.070	−0.071	0.017	0.109	0.059	−0.030	−0.040	1.000
(15) Retention	0.304	0.968	0.143	0.035	−0.009	−0.055	0.017	0.101	−0.037	−0.149	0.013	−0.094	0.071	−0.586	1.000

present the results of the descriptive statistics and correlation analysis, respectively. The data in Table 2 illustrate significant variability in fan attraction capabilities across different livestreaming rooms, with fan growth ranging from 0 to 12.145 and a standard deviation of 6.82. The maximum values for interaction frequency and interaction quality are 3.508 and 1, respectively, while their minimum values are as low as 0.082 and 0, respectively, highlighting the diverse nature of interactions in livestreaming. The descriptive statistics for the other variables are consistent with findings from existing studies.

5.2. Regression Analysis

5.2.1. Baseline Regression Results

The regression results for the main effects are displayed in

Table 4. Baseline regression results.

	(1)	(2)	(3)	(4)
	Model 1 (Fan)	Model 2 (Fan)	Model 3 (Retention)	Model 4 (Retention)
Interaction frequency	0.475 ***		0.164 ***
	(15.4895)		(3.9667)
Interaction quality		0.235 ***		0.963 ***
		(11.8856)		(79.7437)
Duration	0.304 ***	0.172 ***	0.012	−0.001
	(17.1543)	(10.2905)	(0.3943)	(−0.0819)
Reputation	−0.012	−0.002	−0.014	−0.008
	(−0.7109)	(−0.1145)	(−0.4822)	(−1.1502)
Commodity	0.005	0.042 **	−0.078 ***	−0.024 ***
	(0.2905)	(2.3109)	(−2.8600)	(−3.8869)
Price	−0.050 ***	−0.029 *	−0.001	−0.010
	(−3.3124)	(−1.7431)	(−0.0430)	(−1.5986)
Likes	0.307 ***	0.640 ***	0.049	0.022 ***
	(9.7489)	(38.0040)	(1.0733)	(3.0192)
Recommendation	−0.015	−0.020	−0.000	0.006
	(−0.7334)	(−1.0423)	(−0.0175)	(0.7909)
Type	−0.036 **	−0.003	−0.170 ***	−0.018 **
	(−2.1159)	(−0.1661)	(−6.3281)	(−2.5725)
Gender	0.010	0.017	0.018	0.001
	(0.6848)	(1.0718)	(0.6876)	(0.1076)
Age	0.012	−0.001	−0.084 ***	−0.007
	(0.6981)	(−0.0490)	(−3.3940)	(−0.9845)
Speed	0.087 ***	0.095 ***	0.050 **	0.002
	(3.3307)	(3.5690)	(2.1166)	(0.2785)
N	1472	1472	1472	1472
Adj. R2	0.6325	0.5960	0.0630	0.9381

Standardized beta coefficients; t statistics in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01.

. Models 1 and 2 reveal that the coefficients for interaction frequency and interaction quality are 0.475 and 0.235, respectively, both of which are significantly positive. This indicates that both interaction frequency and interaction quality positively influence fan growth in the livestreaming room. In other words, higher interaction frequency and better interaction quality lead to greater fan growth, thus confirming H1a and H1b. In Models 3 and 4, the coefficients for interaction frequency and quality on user retention are 0.164 and 0.963, respectively, which are also significantly positive. This demonstrates that both higher interaction frequency and better interaction quality positively affect user retention in the livestreaming room, indicating that users tend to stay longer.

5.2.2. Regression Results of Mediating Effect and Moderating Effect

The regression results for the mediating and moderating effects are presented in

Table 5. Results of mediating effect and moderating effect.

	(1)	(2)	(3)	(4)
	Model 5	Model 6	Model 7	Model 8
Interaction frequency	0.443 ***		2.737 ***
	(14.7774)		(22.1521)
Interaction quality		0.224 **		1.854 ***
		(2.0118)		(15.2527)
Retention	0.198 ***	0.087 *
	(11.1876)	(1.7951)
Facial attractiveness			0.208 ***	0.366 ***
			(8.3003)	(6.4285)
Interaction frequency * Facial attractiveness			−2.668 ***
			(−22.1925)
Interaction quality * Facial attractiveness				−0.778 ***
				(−7.5363)
Duration	0.302 ***	0.172 ***	−0.003	0.001
	(17.7651)	(10.2874)	(−0.1186)	(0.2178)
Reputation	−0.010	−0.002	−0.006	−0.006
	(−0.5563)	(−0.1152)	(−0.2580)	(−0.8981)
Commodity	0.020	0.042 **	−0.054 **	−0.020 ***
	(1.2569)	(2.3129)	(−2.3555)	(−3.4140)
Price	−0.049 ***	−0.029 *	0.013	−0.004
	(−3.3851)	(−1.7473)	(0.5583)	(−0.6947)
Likes	0.297 ***	0.640 ***	0.050	0.017 **
	(9.7369)	(37.8592)	(1.3169)	(2.4243)
Recommendation	−0.015	−0.020	−0.020	0.004
	(−0.7993)	(−1.0401)	(−0.8076)	(0.5450)
Type	−0.003	−0.003	−0.108 ***	−0.013 *
	(−0.1740)	(−0.1672)	(−4.6977)	(−1.9165)
Gender	0.007	0.017	0.017	0.004
	(0.4608)	(1.0718)	(0.7949)	(0.6972)
Age	0.028 *	−0.001	−0.047 **	−0.014 **
	(1.7508)	(−0.0497)	(−2.2622)	(−2.0129)
Speed	0.077 ***	0.095 ***	0.031 *	0.003
	(3.1795)	(3.5680)	(1.6527)	(0.4896)
N	1472	1472	1472	1472
Adj. R2	0.6689	0.5957	0.3281	0.9441

Standardized beta coefficients; t statistics in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01.

. Models 5 and 6 indicate that even after including the mediating variable (user retention), the coefficients of interaction frequency and quality on fan growth remain significantly positive, although they are reduced from 0.475 to 0.443 and from 0.235 to 0.224, respectively. The coefficients of user retention on fan growth are also significantly positive, confirming that user retention partially mediates the relationship between livestream interactions and fan growth, thus verifying H2a and H2b.

Models 7 and 8 explore the moderating effects of the anchor’s facial attractiveness. The interaction terms of interaction frequency and quality with the anchor’s facial attractiveness yield coefficients of −2.668 and −0.778, respectively, both of which are significantly negative. This suggests that the anchor’s facial attractiveness negatively moderates the positive impacts of interaction frequency and quality on fan growth. Specifically, livestreaming rooms with less attractive anchors experience a more pronounced positive effect from interaction frequency and quality on fan growth compared to those with more attractive anchors. This result implies that the effects of the anchor’s facial attractiveness and interactivity are interrelated and compensatory, thus verifying H3a and H3b.

5.3. Robustness Test

We tested the robustness of the results using several methods. First, we replaced the regression model. Since fan growth is truncated count data, we used the Tobit model for the main effect regression. The results are shown in Model 9 and Model 10 in

Table 6. Robustness test using model replacement and dependent variable substitution.

	(1)	(2)	(3)	(4)
	Model 9	Model 10	Model 11	Model 12
Interaction frequency	0.475 ***		0.112 **
	(18.8965)		(2.2884)
Interaction quality		0.235 ***		0.521 ***
		(13.8617)		(26.3221)
Duration	0.304 ***	0.172 ***	0.007	−0.007
	(16.3457)	(9.6139)	(0.2363)	(−0.2799)
Reputation	−0.012	−0.002	−0.073 **	−0.069 **
	(−0.7492)	(−0.1368)	(−2.1952)	(−2.4280)
Commodity	0.005	0.042 **	−0.135 ***	−0.104 **
	(0.3054)	(2.4283)	(−2.6767)	(−2.0923)
Price	−0.050 ***	−0.029 *	0.005	0.001
	(−3.1000)	(−1.7388)	(0.1837)	(0.0531)
Likes	0.307 ***	0.640 ***	−0.028	−0.024
	(11.6585)	(34.3125)	(−0.5611)	(−1.0565)
Recommendation	−0.015	−0.020	−0.000	0.003
	(−0.9204)	(−1.1785)	(−0.0016)	(0.1650)
Type	−0.036 **	−0.003	−0.155 ***	−0.073 ***
	(−2.1646)	(−0.1647)	(−6.7551)	(−3.8477)
Gender	0.010	0.017	0.043 **	0.034 **
	(0.6387)	(0.9825)	(2.0774)	(2.2521)
Age	0.012	−0.001	−0.058 ***	−0.017
	(0.7226)	(−0.0523)	(−2.6829)	(−0.9799)
Speed	0.087 ***	0.095 ***	0.036	0.011
	(5.4779)	(5.7190)	(1.6298)	(0.5606)
N	1472	1472	1472	1472
Adj. R2			0.0584	0.3131

Standardized beta coefficients; t statistics in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01.

, and they remain consistent with the previous findings. Second, we replaced the dependent variable with the fan conversion rate, which is calculated as the ratio of the number of new fans acquired during the LSE session to the total number of viewers. The regression results are shown in Model 11 and Model 12 in Table 6, and there were no significant changes in the results. Finally, to eliminate the impact of extreme values, we applied a 1% trimming to all continuous variables. The results are shown in

Table 7. Robustness test using winsorized continuous variables.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	Model 13	Model 14	Model 15	Model 16	Model 17	Model 18	Model 19	Model 20
Interaction frequency	0.478 ***		0.163 ***		0.445 ***		2.772 ***
	(15.3898)		(3.9509)		(14.6690)		(23.0799)
Interaction quality		0.233 ***		0.966 ***		0.204 ***		1.729 ***
		(12.5223)		(93.7488)		(2.7964)		(16.1169)
Retention					0.198 ***	0.030
					(11.6543)	(0.4120)
Facial attractiveness							0.206 ***	0.313 ***
							(8.2904)	(6.1992)
Interaction frequency * Facial attractiveness							−2.701 ***
							(−23.0974)
Interaction quality * Facial attractiveness								−0.663 ***
								(−7.4236)
Control	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
N	1472	1472	1472	1472	1472	1472	1472	1472
Adj. R2	0.6305	0.5915	0.0632	0.9438	0.6670	0.5913	0.3333	0.9481

Standardized beta coefficients; t statistics in parentheses; *** p < 0.01.

, and all findings are consistent with the previous ones, indicating that the research conclusions are robust.

5.4. Endogeneity Test

5.4.1. Propensity Score Matching (PSM)

Considering the potential endogeneity issues arising from selection bias, we used propensity score matching (PSM) to simulate a randomized experiment, allowing for a more accurate test of the causal effects of interaction frequency and interaction quality on fan growth. Based on the average interaction frequency and interaction quality in the sample, we divided the livestream rooms into a high-interaction-frequency and high-interaction-quality group (treatment group) and a low-interaction-frequency and low-interaction-quality group (control group). We also used all of the control variables in the model as matching variables and applied 1:1 caliper matching to calculate propensity scores. The distribution of characteristics before and after matching for the treatment and control groups is shown in Figure 3. Model 21 and Model 22 in

Table 8. PSM.

	(1)	(2)
	Model 21	Model 22
Interaction frequency	0.378 ***
	(9.0666)
Interaction quality		0.232 ***
		(8.4640)
Duration	0.361 ***	0.188 ***
	(8.5946)	(7.8809)
Reputation	−0.004	−0.012
	(−0.1250)	(−0.3947)
Commodity	0.025	0.069 ***
	(0.7656)	(2.8228)
Price	−0.036	−0.027
	(−1.1739)	(−1.1372)
Likes	0.225 ***	0.622 ***
	(4.2699)	(26.5425)
Recommendation	−0.065	−0.040
	(−1.5820)	(−1.4341)
Type	−0.030	0.037
	(−0.7975)	(1.5435)
Gender	0.005	0.011
	(0.1589)	(0.5437)
Age	−0.023	0.004
	(−0.5902)	(0.1639)
Speed	0.091 *	0.140 ***
	(1.8076)	(3.3272)
N	455	757
Adj. R2	0.4436	0.6064

Standardized beta coefficients; t statistics in parentheses; * p < 0.1, *** p < 0.01.

present the regression results for fan growth after matching. The results indicate that the results of this study are not affected by selection bias.

5.4.2. Sensitivity Analysis

We considered potential endogeneity issues that may arise from omitted variables and conducted a sensitivity analysis. This methodology attempts to demonstrate that the direction of the study’s results is not affected by omitted variables when the strength of these omitted variables is less than some multiple of the core control variables [66]. We used the duration of the livestreaming as a comparison variable. As shown in Figure 4, the direction of the coefficients (a) and the T-value for interaction frequency (b) did not change when the omitted variable explained less than or equal to three times the strength of the comparison variable. The results for the interaction quality also remained consistent. This suggests that omitted variables are unlikely to affect this study’s results.

6. Discussion

6.1. Key Findings

This study found that both interaction frequency and interaction quality in LSE significantly affect fan growth. This finding supports previous research, which demonstrated that perceived interactivity stimulates users’ intentions to engage in positive behaviors in online communities [67]. Additionally, these results align with IRCT, which suggests that real-time interactions in livestreaming rooms create a virtual physical presence [23]. Through repeated interaction with anchors and other users, an interactive ritual chain is formed, fostering a sense of belonging and emotional solidarity. This emotional connection enhances users’ perceived transparency, trust, and engagement, further promoting fan conversion [31,56].

Secondly, we discovered that user retention mediates the effect of interactivity on fan growth. This suggests that improving interactivity alone is not sufficient; sustaining viewer engagement over time is the key channel through which interactivity translates into fan acquisition. This aligns with prior studies showing that interactivity enhances emotional energy and collective identity in online communities, leading to longer user retention [68]. In LSE, this extended retention time increases exposure to the content and atmosphere of the livestreaming room, thereby increasing the probability of fan conversion. This mediating effect highlights the importance of designing not only interactive but also engaging and emotionally resonant livestreaming experiences to foster long-term audience attachment.

Thirdly, the results show that the anchor’s facial attractiveness negatively moderates the effect of interactivity on user retention, indicating a substitution effect. While previous studies suggested that facial attractiveness as a heuristic cue could reinforce other persuasive signals (e.g., verbal content), our findings suggest that, in the context of LSE, attractive facial features may reduce the marginal benefits of interactive behavior. One possible explanation for this is that highly attractive anchors may trigger passive consumption patterns, where users are drawn to appearances but less motivated to engage or remain. Conversely, when facial attractiveness is low, interactive behavior becomes a more salient indicator of anchor competence, increasing perceived authenticity and emotional connection. Additionally, our findings may reflect the complexity and saturation of multimodal cues in livestreaming settings. When too many signals (e.g., beauty, interaction) are present simultaneously, cognitive overload or distraction may occur [38], leading to diminishing returns for interaction strategies.

6.2. Theoretical Contributions

This study offers several theoretical contributions: (1) Broadening the scope of LSE performance research: Previous studies have primarily focused on short-term performance metrics such as livestreaming sales, conversion rates, and gift-giving [69,70,71] but have overlooked fan growth as a long-term performance indicator. Our research demonstrates that interactivity has a significant impact on fan growth, thereby broadening the scope of LSE performance research and providing a new perspective on the long-term sustainability of LSE. (2) Revealing the mechanism of interactivity’s impact on fan growth in LSE: Although previous research has explored the impact of interactivity on social performance, these studies have primarily focused on short-term user behaviors such as immediate engagement and interaction frequency [34], and the underlying mechanisms have not been fully clarified. Our findings show that interactivity promotes fan growth by enhancing user retention, providing a new theoretical framework for studying the relationship between interactivity and LSE performance. (3) Incorporating the anchor’s facial attractiveness into the LSE research framework: Existing research on anchors’ facial features mostly focuses on aspects such as emotional fluctuations [72,73], neglecting facial attractiveness as an important digital marketing cue. This study is one of the first to explore the moderating role of anchors’ facial attractiveness in the process of interactivity influencing fan growth, clarifying the substitution effect between interactivity and anchor characteristics (such as facial attractiveness) on user retention, and providing a new theoretical perspective for the study of multiple signal interactions in LSE. (4) Validating the applicability of IRCT and DST in the LSE context: Existing research on IRCT and DST has mostly focused on traditional business or other social fields [74,75]. This study establishes the relationships between interactivity, user retention, anchors’ facial attractiveness, and fan growth, validating the applicability of IRCT and DST in the emerging context of LSE, enriching the theoretical foundation of LSE, and providing new insights for future research.

6.3. Practical Implications

The findings of this study have significant practical implications for livestreaming platforms, marketing managers, and anchors, especially in terms of enhancing fan growth, user retention, and livestreaming room marketing performance. The specific implications are as follows: (1) The role of interactivity in promoting fan growth: The results indicate that both the frequency and quality of livestreaming interactivity have a significant positive impact on fan growth. Therefore, marketing managers and anchors should fully recognize the key role that interactivity plays in driving fan growth. To maximize the effectiveness of interactivity, it is essential to provide anchors with systematic training in interaction skills, helping them improve the quality of their interactions, including how to better engage with the audience, answer questions, and guide discussions. Furthermore, anchors are encouraged to provide professional explanations and thematic guidance during livestreams, creating a positive and harmonious interactive atmosphere that fosters more active bullet comments, enhancing users’ sense of participation and belonging. (2) User retention as a mediating factor: This study found that user retention plays a mediating role in the impact of interactivity on fan growth. Therefore, livestreaming platforms can promote fan growth by improving user retention. To enhance user retention, platforms should innovate interactive methods, such as incorporating gamification elements, reward-based Q&A, interactive challenges, etc., to increase audience engagement and stickiness. Additionally, platforms should optimize the interactive experience by setting up more interactive sessions, such as raffles and Q&A segments, to improve the frequency and quality of audience participation during the livestream. This will help extend user retention time, further promoting fan growth. (3) The substitution effect between the anchor’s personal characteristics and interactivity: This study confirms the potential substitution effect between an anchor’s facial attractiveness and interactivity on user retention. Specifically, anchors with higher facial attractiveness can compensate for lower interactivity by leveraging the “facial attractiveness premium,” which helps mitigate the decline in user retention caused by lower interactivity. Based on this, we recommend that when selecting anchors, especially for newly established livestreaming rooms with fewer bullet comments and lower comment activity, facial attractiveness and other personal characteristics should be included as selection criteria. By choosing attractive anchors, platforms can compensate for retention issues caused by insufficient interactivity. Additionally, for livestreaming rooms with lower interactivity, selecting anchors with higher facial attractiveness can enhance user retention, while supplementing other interactive strategies to improve the room’s marketing performance.

6.4. Limitations and Prospects

This study has several limitations: (1) The empirical data were collected exclusively from Douyin, China’s leading LSE platform. Due to differences in culture, economic systems, and platform functionality, our findings may not be directly generalizable to international platforms such as TikTok, YouTube Live, or Amazon Live. Future studies should consider cross-country or cross-platform comparisons to provide more comprehensive and generalizable insights. (2) Our analysis did not incorporate user-level demographic information (e.g., age, gender, income, occupation) due to data access restrictions and ethical concerns related to user privacy. While this omission is consistent with practices in prior research [61,76], individual user characteristics may influence fan acquisition dynamics. Future research could enrich the model by integrating user-level attributes through ethically approved and privacy-compliant data collection strategies. (3) This study focused primarily on visual (e.g., facial attractiveness) and textual (e.g., bullet comments) signals in the livestreaming environment. However, other important multimodal cues, such as tone of voice, background music, and color schemes, were not examined. Future research could adopt a more holistic approach by incorporating these non-verbal and ambient factors to better capture the full spectrum of user engagement mechanisms. (4) The data collection period excluded the Chinese New Year holiday to avoid confounding results with abnormal seasonal patterns. However, this also limits the scope of the analysis, as holidays often represent peak user activity and unique behavioral patterns. Future studies should consider including holiday data to examine how temporal contexts affect interactivity and fan growth. (5) This study involved the use of large-scale behavioral and visual data, including facial recognition techniques to assess anchors’ attractiveness. While data were collected from publicly accessible sources and no personally identifiable information was used, we acknowledge the growing importance of aligning with data ethics and platform policies. Future research should further address issues of informed consent and data governance in the context of large-scale social media analytics.

7. Conclusions

Increasing the number of fans in LSE through marketing has become a crucial topic in the field of management. This study explored the determinants of fan growth in LSE from the perspectives of interactivity and user retention, while also considering the complex interaction effect between anchor characteristics and interactivity on fan growth. Using a real-time, objective dataset from Douyin, we validated the theoretical model. Empirical analyses showed that interactivity significantly and positively affects fan growth by enhancing user retention. Additionally, we found that interactivity and an anchor’s facial attractiveness have a substitution effect on fan growth, suggesting a substitution relationship between anchor characteristics and livestreaming room characteristics. These findings offer both practical and theoretical insights for LSE operations managers and future researchers.