Research on the influence of algorithm technology on consumers’ shopping intention in live-streaming bandwagons

Abstract

This study investigates how algorithmic technology influences consumers’ shopping intentions in the context of live-streaming bandwagon consumption. Grounded in the Theory of Planned Behavior (TPB) and the Technology Acceptance Model (TAM), it develops an integrated framework that combines psychological and technological dimensions. Based on data collected through an online questionnaire and analyzed via structural equation modeling, the study finds that algorithmic technology enhances consumers’ perceived ease of use by improving access to live-streaming interfaces and increases perceived enjoyment by strengthening anchor effectiveness. These factors jointly shape consumers’ attitudes toward live shopping. Additionally, algorithmic mechanisms such as data targeting, user labeling, scoring systems, and group-based recommendations significantly affect behavioral intentions through both direct and mediated pathways, including attitudes, perceived behavioral control, and promotional incentives. This research offers a novel empirical contribution by modeling how algorithmic personalization impacts live shopping behaviors, providing theoretical and practical insights for digital commerce applications.

Keywords

algorithmic personalization live-streaming commerce consumer behavior technology acceptance model theory of planned behavior

Introduction

The rapid proliferation of mobile Internet and digital technologies has profoundly reshaped consumer behavior, catalyzing the emergence of new digital marketing formats. Among these, live-streaming commerce—particularly the use of algorithmically driven recommendation systems within live-stream environments—has become one of the most transformative developments in online retail.¹ According to the Statistical Report on the Development of the Internet in China, as of June 2023, the number of Internet users reached 1.079 billion, with instant messaging, online video, and short video platforms maintaining dominant user engagement. In this ecosystem, live-streaming commerce represents a hybrid model that integrates entertainment, interaction, and instant transaction capabilities, and is increasingly powered by sophisticated algorithmic technologies.

Algorithmic personalization plays a pivotal role in shaping user experiences by delivering targeted content, optimizing traffic flow, and constructing user profiles in real time. This has allowed platforms to more effectively match products with user preferences, increase engagement, and stimulate impulsive purchases. For consumers, however, such algorithmic interventions raise questions about behavioral influence, decision autonomy, and the psychological pathways through which algorithmic systems affect purchase intentions.^2,3 Despite growing industry adoption, academic research exploring the behavioral mechanisms behind algorithm-driven live commerce remains limited and fragmented.

Previous studies on live-streaming commerce have primarily focused on its operational models, user engagement strategies, and legal or ethical concerns. Some empirical efforts, such as those by Pingsheng et al.,⁴ have evaluated marketing effectiveness using AISAS theory, while others have discussed trust⁵ and e-commerce legal frameworks.⁶ Parallel research on algorithmic technologies has addressed their social, ethical, and cognitive implications,^7,8 but rarely in the context of commerce, let alone live-stream commerce. Moreover, few empirical studies have integrated algorithmic factors into robust theoretical models of consumer behavior.

To address this gap, this study integrates two widely used behavioral theories—the Technology Acceptance Model (TAM) and the Theory of Planned Behavior (TPB)—to construct a comprehensive analytical framework. TAM provides a foundation for understanding how consumers perceive ease of use and enjoyment in technology-mediated environments, while TPB captures attitudinal, normative, and control factors that predict behavioral intention. By embedding algorithmic constructs such as personalization, scoring, and labeling into this dual-theoretical lens, the study aims to empirically examine how algorithmic technologies influence users’ shopping intentions in live-streaming environments.

Unlike previous studies that focus separately on algorithm design or consumer psychology, this paper is among the first to integrate algorithm-driven constructs—such as datafication, labeling, and group relations—into a TAM-TPB behavioral framework to examine their effect on live-stream shopping intention.

This research contributes to the literature in three significant ways: (1) it models algorithmic influence on consumer behavior within a theoretically grounded framework; (2) it bridges the gap between algorithmic personalization and behavioral intention in live-streaming contexts; and (3) it offers practical implications for the design and governance of algorithm-driven e-commerce platforms. Through empirical investigation using questionnaire data and structural equation modeling, the study advances our understanding of how algorithmic technologies shape consumer decision-making in the era of mobile digital commerce.

Theoretical basis

Theories of planned behavior

Fishbein and Ajzen (1975) initially proposed the Theory of Reasoned Action (TRA) to explain human behavior based on intention and attitude. Ajzen⁹ later extended this theory by introducing the concept of perceived behavioral control, formally establishing the Theory of Planned Behavior (TPB). The theory reached maturity with Ajzen’s¹⁰ publication and has since become one of the most widely applied social-psychological frameworks for understanding and predicting human behavior.¹¹

In the field of consumer behavior, TPB has proven effective in explaining various purchasing decisions.¹² The model suggests that behavioral intention, which directly predicts actual behavior, is influenced by three components: attitude toward the behavior, subjective norms, and perceived behavioral control. Attitude refers to an individual’s favorable or unfavorable evaluation of performing a specific behavior.⁹ Subjective norms reflect perceived social pressure to engage or not engage in a particular behavior.¹³ Perceived behavioral control represents an individual’s perception of how easy or difficult it is to perform the behavior, accounting for both internal ability and external constraints.^10,14

TPB is particularly useful for understanding consumer decisions in live-streaming commerce, where social influence, individual confidence, and environmental constraints often coexist. In such dynamic and interactive environments, consumers’ purchase intentions are shaped not only by personal evaluation but also by perceived expectations and the degree of control afforded by the technological platform.

Technology acceptance model

The Technology Acceptance Model (TAM) was proposed by Davis¹⁵ and published in 1989. Drawing from expectancy theory and self-efficacy theory, TAM is designed to predict individuals’ acceptance, use, or rejection of new information technologies. Since its introduction, the model has undergone repeated empirical testing and refinement, becoming one of the most influential frameworks in behavioral and technological adoption research. It has been widely applied in fields such as management, psychology, and information systems.¹⁶

TAM posits that users’ behavioral intention to adopt a new technology is primarily influenced by their attitude toward using it, which is in turn determined by two cognitive evaluations: perceived usefulness (PU) and perceived ease of use (PEU). Davis¹⁵ defines perceived usefulness as the degree to which a person believes that using a particular system will enhance their performance. Perceived ease of use refers to the degree to which a person believes that using the system will require minimal effort.¹⁷

In the context of live-streaming commerce, TAM offers critical insights into how users access platform technologies. For instance, algorithmic recommendation systems can enhance perceived usefulness by increasing content relevance, while intuitive interfaces and personalized access pathways may improve perceived ease of use. As live shopping becomes increasingly data-driven, understanding these perceptions is essential to capturing the full behavioral impact of algorithmic technology.

Taken together, TPB and TAM provide a complementary theoretical foundation for analyzing algorithm-driven consumer behavior. While TAM focuses on individual perceptions of technology functionality, TPB emphasizes the social and motivational dimensions of decision-making. Integrating both models allows for a more holistic understanding of how algorithmic personalization influences user attitudes, intentions, and behaviors in live-streaming commerce environments.

Application of TPB and TAM in algorithmic live-streaming commerce

To ground these theoretical frameworks in the specific context of algorithm-driven live-streaming commerce, it is essential to map the core constructs of TPB and TAM to relevant phenomena in real-world platforms. In live commerce, algorithmic technologies such as personalized recommendation, consumer profiling, and traffic allocation systems directly influence user perceptions, social interactions, and behavioral control.

For example, perceived usefulness is shaped by the system’s ability to deliver relevant products through real-time algorithmic recommendation and ranking, which improves consumers’ decision efficiency. Perceived ease of use is enhanced by the system’s automatic content curation and one-click purchasing pathways, reducing cognitive and operational effort during live-streaming sessions.

From the TPB perspective, attitude may be influenced by consumers’ enjoyment and trust in the algorithmically tailored content and hosts. Subjective norms are affected by social cues embedded in the platform, such as live comment sections, influencer popularity, and peer engagement indicators. Perceived behavioral control is modulated by algorithmic prompts (e.g., limited-time offers and scarcity signals) that may either enhance or limit users’ perceived autonomy in decision-making.

Furthermore, algorithmic features such as user labeling, scoring systems, and group-based segmentation also intervene in these perceptions by reinforcing identity constructs, highlighting hierarchical status cues, and differentiating consumer experiences across user types. By embedding these dynamic algorithmic mechanisms into the TPB-TAM framework, this study advances a context-specific behavioral model tailored to the digital architecture of live-streaming commerce.

Comparison with alternative models and algorithmic foundations

In addition to the Technology Acceptance Model (TAM) and the Theory of Planned Behavior (TPB), other theoretical models have been widely adopted to explain user behavior in digital environments. The Unified Theory of Acceptance and Use of Technology (UTAUT), for example, integrates constructs such as performance expectancy and social influence from eight earlier models, and has demonstrated strong explanatory power, particularly in organizational and institutional contexts. Likewise, the Stimulus-Organism-Response (S-O-R) model emphasizes how external stimuli affect emotional and cognitive responses, and has been applied to understand affective mechanisms in consumer behavior. While these models are insightful, they may not fully capture the interplay between technological cognition and social influence in user-driven, algorithm-mediated commerce environments.

Moreover, to better understand the technical basis behind algorithmic personalization in live-streaming commerce, it is important to recognize the contribution of neural network–based recommendation systems. Recent studies have shown that recurrent neural networks (RNNs), including complex-valued and Clifford-valued variants, can effectively capture temporal dependencies and non-linear patterns in user behavior, enabling more accurate content matching and personalized traffic distribution. For example, Rajchakit et al.¹⁸ and Boonsatit et al.¹⁹ demonstrated how stability-enhanced neural network architectures improve decision robustness under time-delay and feedback conditions. Similarly, Wang et al.²⁰ and Zhang et al.²¹ explored event-triggered control and fixed-time synchronization to optimize recommendation delivery. In addition, Ma et al.²² applied deep learning to develop adaptive, personalized recommendation systems for social media environments.

These algorithmic advances underscore the necessity of integrating behavioral models with algorithmic mechanisms. The TAM-TPB framework adopted in this study allows for such integration, by connecting cognitive perceptions (e.g., perceived usefulness and ease of use) and social-psychological variables (e.g., attitudes, norms, and control) to data-driven, neural network–based personalization processes, thus providing a comprehensive foundation for modeling user behavior in live-streaming commerce.

Hypothesis modeling and scale design

Research hypothesis and model formulation

In this study, three context-specific constructs are introduced to reflect how algorithmic live-streaming platforms influence user perception:

Datafication refers to the process by which user behaviors are translated into algorithmically processed information, such as labels, scores, and groupings.²³

Anchor effect is defined as the emotional and cognitive influence of livestream hosts on viewers, stemming from their charisma, expertise, and interactive engagement.

Scene building describes the deliberate creation of visual and auditory elements in the livestream environment (e.g., lighting, props, and layout) to enhance immersion and emotional impact.

Each of these constructs is reflected in the questionnaire through dedicated measurement items, and is treated as a latent variable in the proposed model.

Based on the concept of perceived ease of use (PEU) in the Technology Acceptance Model (TAM), this study further operationalizes PEU into two contextual dimensions: (1) ease of access to the live room, and (2) ease of operation during commodity purchase. “Ease of access to the live room” is defined as the user’s perceived effort required to enter a relevant and engaging live-streaming session, typically influenced by algorithmic recommendations, push notifications, and content curation. As algorithmic technology primarily functions in the background—analyzing user behaviors and recommending content—it significantly enhances this access process but does not directly influence the transactional steps of commodity purchase. Therefore, “ease of access” is considered a valid subdimension of PEU in the context of algorithm-driven live shopping platforms, the relevant concepts are shown in Table 1 below.

Table 1.

Definitions and theoretical sources of key constructs.

Construct	Sub-variable	Definition	Model source
Perceived ease of use (PEU)	Ease of access to live room	The degree to which users feel it is easy to reach relevant live sessions via algorithmic recommendations and notifications	TAM
Perceived ease of use (PEU)	Ease of operation	The degree to which users perceive the product purchasing process in the live room to be effortless	TAM

Based on this reasoning, Hypothesis 1 is proposed.

Algorithmic technology can positively influence users’ attitudes towards live-streaming commerce by enhancing the perceived ease of access to the live room.

The theory of planned behavior suggests that perceived behavioral control influences consumers’ willingness to buy, which in turn influences behavior, so basic hypothesis 2 is proposed.

Algorithmic techniques can positively influence users’ perceived behavioral control.

Lee and Olafsson²⁴ found that promotion has a significant positive effect on purchase intention when they studied the relationship between promotion and purchase intention, and Li et al.²⁵ found that the perception of promotion is the highest in consumers’ purchase intention, and the effect is significant. Based on this, promotional discount’ is added into the model as an exogenous latent variable affecting behavioral intention. Therefore, Hypothesis 3 is proposed.

Algorithm technology can positively influence the effect of promotional discounts on users’ intention to shop on live platforms.

Lan²³ noted that algorithms function as intermediaries, transforming individual presence into data-driven identities. A core outcome of this process is labeling, where users, content, and products are tagged based on behavioral data to enable personalized recommendations. Algorithms also incorporate rating systems—via peer or institutional evaluations—to shape social reputation and visibility. Additionally, by assigning tags and clustering users with similar traits, algorithms foster group identity, and community formation.

Based on these observations, this study categorizes the four mechanisms into two higher-order constructs.

(1) Algorithmic Profiling, which includes datatization and labeling, reflects how user data is extracted and operationalized to form dynamic profiles for personalized engagement.

(2) Algorithmic Social Feedback, which includes scoring systems and group relationships, captures how social mechanisms are embedded into algorithms to influence users’ perceived social value and behavioral alignment.

Accordingly, the following hypotheses are proposed.

Algorithmic technology can positively influence users’ live shopping intention through datatization.

Algorithmic technology can positively influence users’ live shopping intention through labeling.

Algorithmic technology can positively influence users’ live shopping intention through scoring systems.

Algorithmic technology can positively influence users’ live shopping intention through constructing group relationships.

Moon and Kim²⁶ identified perceived pleasantness as a key factor influencing user attitudes within the Technology Acceptance Model, while Ahuja and Khazanchi²⁷ emphasized its central role in technology acceptance. Building on this, the present study incorporates perceived pleasantness as a determinant of consumer attitudes toward live shopping. It is conceptualized as an emotional response evoked through the anchor effect and scene building. The anchor effect refers to the positive influence exerted by the anchor’s charisma, expertise, and audience engagement. Scene building involves creating an immersive live-streaming environment—through layout, lighting, and props—that aligns with the content theme and enhances user experience. At the same time, this paper assumes that there is an interactive relationship between the user’s attitude towards live shopping, algorithmic technology and anchor effect, so it proposes hypotheses 8, 9, and 10.

Algorithmic technology can positively affect the anchor effect in the process of users’ live shopping.

The anchor effect and scene building can positively affect the perceived infectiousness, thus positively affecting the user’s attitude towards live shopping.

H10

Algorithmic technology can positively affect the anchor effect in the process of users’ live shopping through the positive effect on attitude.

To enhance conceptual cohesion and theoretical transparency, the 10 hypotheses proposed in this study are categorized into three major groups based on their underlying frameworks. Specifically, H1 and H9 are grounded in the Technology Acceptance Model (TAM), focusing on perceived ease of use and pleasantness. H2, H3, and H10 are based on the Theory of Planned Behavior (TPB), addressing perceived behavioral control, social influence, and attitude mediation. The remaining hypotheses—H4 through H8—are classified under algorithmic social mechanisms, which reflect the socio-technical impact of algorithmic systems on consumer behavior. This classification helps clarify the narrative structure and theoretical alignment of the model.

According to the theory of planned behavior, the theoretical overview of the technology acceptance model and the current state of research on live bandwagon, algorithmic technology and the combination of the two, the theoretical model constructed in this paper is shown in Figure 1.

Figure 1.

Conceptual framework integrating algorithmic technology, TAM, and TPB constructs.

Scale design

To ensure content validity, most measurement items were adapted from previously validated scales in the existing literature. For example, items for perceived ease of use (PE) and perceived pleasure (PP) were adapted from Moon and Kim²⁶ and Ahuja and Khazanchi²⁷; perceived behavioral control (PBC) items were based on Ajzen¹⁰; and promotional discount (PR) items followed the work of Lee and Olafsson.²⁴ Constructs related to algorithmic mechanisms (e.g., labeling, datatization, scoring, and group relationship) were developed based on theoretical insights from Lan.²³ These newly developed items were reviewed by three domain experts and revised accordingly. A small-scale pretest (N = 30) was conducted to assess clarity and relevance, ensuring the quality of the final scale.

Based on the constructed research model of influencing factors of consumer shopping behavior, drawing on the mature scales of many researchers, we get the scale about the influence of algorithmic technology on the factors affecting consumers’ purchase intention in the process of live streaming with goods. The questionnaire includes statistics on consumers’ personal information such as gender, age stage, average monthly disposable amount, education level, identity, etc., of which the identity questions set logical questions based on whether the respondent is an anchor or a user and the time he/she has become an anchor, and the screening questions are set based on whether the respondent has ever had a shopping experience in the live broadcast of e-commerce. The questionnaire also included the measurement of the core variables of the study-perceived ease of use (PE), perceived pleasure (PP), perceived behavioral control (PBC), promotional discounts (PR), datamining (DA), labeling (LA), scoring system (SC), and group relations (GR) using a 5-degree Likert scale.

Data analysis

Among the respondents, 72% were live-stream viewers and 28% were anchors. All participants completed the same version of the questionnaire, as the constructs measured (e.g., perceived ease of use, algorithmic influence, and shopping intention) were applicable to both roles. The responses were analyzed in aggregate. A preliminary comparison of key variables (e.g., attitude and intention) across the two subgroups showed no statistically significant differences, supporting the validity of pooled analysis.

Therefore, the generalizability of the findings should be interpreted with caution. To assess the reliability and validity of the measurement scale, the collected responses were analyzed using SPSS 26.0, including sample characterization, descriptive analysis, reliability and validity tests, correlation analysis, and regression-based hypothesis testing to validate the proposed theoretical model, the data description is shown in the following Figure 2.

Figure 2.

Basic characteristics of the sample (N = 268).

Sample characterization

This study is based on 268 valid questionnaire responses analyzed using SPSS 26.0. The sample demonstrates demographic balance and diversity. Gender distribution is relatively even, minimizing bias (44.4% male, 55.6% female). Respondents span various age groups, with a concentration among youth aged 18–35, particularly 18–25, highlighting the dominance of younger users in live-stream commerce. Income distribution shows that 43.7% of participants have a monthly disposable income above 4000 yuan, indicating a consumer base with substantial spending capacity. Educational attainment is generally high, with the majority holding college or bachelor’s degrees. In terms of role, 72% identified as live-stream platform users, while 28% had experience as anchors, reflecting both consumer and practitioner perspectives. Platform preference is led by TikTok (Jitterbug), with a usage rate of 64.9%, indicating its market dominance in the live-streaming sector.

Descriptive analysis

This paper descriptively analyses the 268 valid questionnaires obtained from the formal research. From the data in Table 2, we know that from the above analysis, the maximum value of the score for each question item is 5, and the minimum value is 1, which indicates that the questionnaire was selected from a wide range of participating objects, with obvious differences, and that there are large differences in the focus of respondents on the questionnaire’s variable items; the mean value of each measurement question item (The average value of each measurement item (AVG) is greater than 3, which indicates that the respondents agree with the formulation of the items, and the participants are generally more adaptive to the questionnaire, and the study is valid.

Table 2.

Sample characterization (N = 268).

Measurement items	AVG	Standard error	Median	Mode	Standard deviation	Variance	MIN	MAX
PE1	4.07	0.057	4	5	0.933	0.871	1	5
PE2	3.94	0.065	4	5	1.058	1.12	1	5
PP1	4.24	0.096	4	5	0.836	0.698	2	5
PP2	4.08	0.057	4	5	0.933	0.87	1	5
PP3	3.96	0.06	4	5	0.99	0.98	1	5
PP4	4.02	0.056	4	5	0.91	0.827	1	5
PBC1	4.04	0.056	4	5	0.909	0.826	1	5
PBC2	3.88	0.058	4	5	0.955	0.912	1	5
PR1	3.98	0.056	4	4	0.91	0.827	1	5
PR2	3.97	0.059	4	5	0.973	0.946	1	5
DA1	3.98	0.06	4	5	0.977	0.955	1	5
DA2	4.1	0.057	4	5	0.927	0.859	1	5
DA3	4.05	0.054	4	4	0.885	0.784	1	5
LA1	4.11	0.116	4	5	1.008	1.015	1	5
LA2	3.99	0.115	4	5	0.993	0.986	1	5
SC1	4.07	0.053	4	4	0.876	0.767	1	5
SC2	4.03	0.057	4	5	0.934	0.872	1	5
SC3	4.1	0.052	4	5	0.857	0.735	2	5
GR1	4.13	0.108	4	5	0.935	0.874	2	5
GR2	3.82	0.065	4	4	1.058	1.119	1	5
GR3	3.95	0.056	4	4	0.925	0.855	1	5
GR4	3.9	0.06	4	4	0.984	0.967	1	5

Reliability and validity analysis

Reliability analysis aims to assess the internal consistency of the sample and is commonly measured by Cronbach’s alpha coefficient. The Alpha coefficient of this questionnaire is 0.966, which is high reliability. Validity tests, on the other hand, focus on the validity of the questionnaire structure and content. The items of this questionnaire were borrowed from the mature scales in the published research results of high-quality academic journals to ensure the content validity. As shown in Table 3, the reliability of the questionnaire items meets the standard, which provides a reliable basis for the subsequent analysis. The KMO and Bartlett’s sphericity test is shown in Table 4, the KMO value of the questionnaire is 0.902, which is much larger than the standard of 0.7, and the p-value is 0.000, which is able to do further factor analysis.

Table 3.

Scale and its reliability and factor loading coefficient results.

Research variables		Measurement item	Factor loading coefficient	Cronbach’s a coefficient
Perceived ease of use	—	PE1: Algorithmic recommendations make it easier for users to find live rooms that meet their needs	0.743	0.805
Perceived ease of use	—	PE2: The ease of access to live rooms caused by algorithmic recommendations makes users feel that live shopping is a good way to shop	0.782	0.805
Perceptual pleasure	Anchor effect	PP1: The live broadcast is subject to an algorithmic push stream allowing the anchor to mobilize selling and pay more attention to personal image	0.666	—
	Anchor effect	PP2: The ability of live-streaming anchors to carry goods and their persona is contagious to consumers’ emotions during the shopping process	0.762	0.866
	—	PP3: Live room scene building is contagious to the user’s emotions during the shopping process	0.721
	Attitude	PP4: The anchor’s strong ability to bring goods, delightful personal image, and infectious scene building can mobilize the user’s willingness to shop on live shopping platforms	0.749
Perceptual-behavioral control	—	PBC1: Algorithmic technology allows users to more easily find their favorite items and complete their purchase	0.746	0.829
Perceptual-behavioral control	—	PBC2: Users can more easily interact with the hosts in the algorithmically recommended live streams	0.663	0.829
Promotional discount	—	PR1: With the algorithm’s recommendation, users can easily learn which live rooms on the platform are doing promotional discount activities	0.820	0.844
Promotional discount	—	PR2: When the algorithm learns that a live room is doing promotional activities, users will enter the live room to view, select, and purchase goods	0.787	0.844
Datafication	Algorithm technology	DA1: Users open the platform’s permission to use personal information before using the live-streaming platform	0.868	0.818
Datafication		DA2: Users believe that their personal information, such as browsing traces, shopping experience, and product information, has been collected by the platform	0.777
Shopping intentions - datamining		DA3: Users believe that the e-commerce products recommended by the live-streaming platform based on their personal data meet their needs	0.789
Labeling		LA1: Anchor in the background will be through the algorithm of the live shopping platform users or live audience, commodities with various types of tags	0.720	0.725
Shopping intentions - labeling		LA2: The anchor believes that the algorithm is accurate classification of goods and consumers, which enables different user groups to buy goods that match their needs	0.854	0.725
Marking scheme		SC1: Most live stores have comprehensive scores on product experience, service experience, logistics experience, etc.	0.706	0.839
Marking scheme		SC2: Users will look at the comprehensive rating of the live shop before shopping	0.732
Shopping intentions - scoring system		SC3: The level of the shop’s comprehensive score will affect users’ shopping choices to a certain extent	0.770
Community relation		GR1: Anchor will interact with the audience in the live room who sends pop-ups	0.752	—
Community relation		GR2: Users will interact with the rest of the audience and the anchor on the pop-up screen when watching the bandwagon live broadcasting room	0.740	0.866
Shopping intentions - community relationships		GR3: The words of other viewers in the live stream will affect users’ willingness to buy in the live stream	0.763
Shopping intentions - community relationships		GR4: The interaction between the anchor and the pop-up screen in the live broadcast room will affect the users’ willingness to buy in the live broadcast room	0.775

Table 4.

Results of KMO and Bartlett’s test of sphericity.

KMO sample suitability quantity		0.902
Bartlett’s test of sphericity	Approximate chi-square	1491.481
	Degree of freedom	231
	Saliency	0.000

Although this study primarily relied on exploratory factor analysis (EFA) to assess construct validity—given the inclusion of newly developed items—confirmatory factor analysis (CFA) will be conducted in future research to further validate the latent structure of the theoretical model. We acknowledge this as a limitation and encourage subsequent studies to perform CFA using larger or independent samples to verify the robustness of the measurement framework.

Correlation analysis

In empirical testing, correlation analysis is usually used to determine the specific relationship between the questionnaire items. In this paper, to analyze the relationship between the variables of the questionnaire on the impact of algorithmic technology on consumers’ purchase intention in the process of live-streaming bandwagon, correlation analysis is carried out, and the results of the study are shown in the table.

(1) Correlation analysis between the independent variable-algorithmic technology and the dependent variables-perceived ease of use, perceived behavioral control, promotional discounts, and anchor effect.

As can be seen from Figure 3, there is a significant positive correlation between perceived ease of use, perceived behavioral control, promotional discounts, anchor effect, and the independent variable-algorithmic technology, and the link is strong, with correlation coefficients of 0.539, 0.609, 0.580, and 0.890, respectively, preliminary validation of the hypotheses 1, 2, 3, and 8. In order to make the correlation graph more convenient and intuitive, this study uses PEU, PBC, PD, AE, and AT for perceived ease of use, perceptual-behavioral control, promotional discount, anchor effect, and algorithm technology.

(2) Correlation analysis between the independent variable (datatization, labeling, scoring system, and group relationship) and the dependent variable (shopping intention).

Figure 3.

Results of correlation analysis.

As can be seen from Figure 4, there is a significant positive correlation between datatization, tagging, scoring system, group relationship and shopping intention, and the correlation coefficients are 0.731, 0.967, 0.707, and 0.443, respectively. Hypotheses 4, 5, 6, and 7 are preliminarily verified. In order to make the correlation graph more convenient and intuitive, this study uses RS and CR for rating system community relation and the corresponding user shopping intention is represented by SI(D), SI(L), SI(RS), and SI(GR).

(3) Correlation analysis between independent variable-perceived pleasantness and dependent variable-attitude.

Figure 4.

Results of correlation analysis.

As can be seen from Table 5, there is a significant positive correlation between perceived pleasantness and consumers’ attitudes towards live shopping behavior, with a correlation coefficient of 0.550, which preliminarily confirms Hypothesis 9.

Table 5.

Results of correlation analysis.

	Perceived pleasure	Attitude
Perceptual pleasure	1
Attitude	0.550**	1

Note: (1) ** Significant correlation at the 0.01 level (two-tailed).

Regression analysis

On the basis of the correlation analysis of the influencing factors of consumers shopping intention in the mode of live banding that has been carried out in the previous section, in order to further understand the degree of influence of each factor on consumers’ shopping intention and between the factors, this paper carries out a linear regression analysis, and the specific results are shown in Table 6 and Table 7.

Table 6.

Model Fit Indicators for regression equation.

Model	M1	M2	M3	M4	M5	M6	M7	M8
R²	0.682	0.645	0.601	0.589	0.563	0.548	0.626	0.711
Adjusted R²	0.674	0.637	0.592	0.581	0.654	0.639	0.617	0.603

Table 7.

Regression analysis and hypothesis testing results.

Model		β Value Standard error		Standardized coefficient beta	T Value	Salience	Test hypothesis
Implicit variable	Independent variable	β Value Standard error		Standardized coefficient beta	T Value	Salience	Test hypothesis
Perceived ease of use	Algorithm technology	0.581	0.056	0.539	10.434	0.000	H1 established
Constant		2.198	0.180	—	12.234	0.000	—
Perceptual-behavioral control	Algorithm technology	0.620	0.049	0.609	12.530	0.000	H2 established
Constant		2.032	0.160	—	12.733	0.000	—
Promotional discount	Algorithm technology	0.601	0.052	0.580	11.618	0.000	H3 established
Constant		2.105	0.167	—	12.628	0.000	—
Shopping intentions - datamining	Digitization	0.777	0.044	0.731	17.469	0.000	H4 established
Constant		0.915	0.183	—	4.990	0.000	—
Shopping intentions - labeling	Labeling	0.939	0.015	0.967	61.561	0.000	H5 established
Constant		0.036	0.034	—	1.053	0.293	—
Shopping intentions - scoring system	Marking scheme	0.742	0.046	0.707	16.312	0.000	H6 established
Constant		1.097	0.188	—	5.841	0.000	—
Shopping intentions - community relationships	Community relation	0.325	0.040	0.443	8.061	0.000	H7 established
Constant		3.115	0.112	—	27.860	0.000	—
Attitude	Perceptual pleasure	0.522	0.049	0.550	10.731	0.000	H9 established
Constant		2.414	0.157	—	15.409	0.000	—

In the first three models, the algorithmic technology p-value is 0.000, which is less than 0.05, and the linear regression coefficient is greater than 0. This indicates that the algorithmic technology has a significant positive impact on perceived ease of use, perceived behavioral control, and promotional discounts, and H1, H2, and H3 hold. There is a linear regression relationship between algorithmic technology (X) and perceived ease of use (Y1), perceived behavioral control (Y2), and promotional discounts (Y3), and the regression equation:

Y_{1} = 0.581 X + 2.198

Y_{2} = 0.620 X + 2.032

Y_{3} = 0.601 X + 2.105

In models 4 to 7, the p-values for datafication, labeling, rating system, and group relationship are 0.000, which is less than 0.05, and the linear regression coefficients are greater than 0. This indicates that datafication has a significant positive impact on shopping intention-datafication, labeling on shopping intention-labeling, rating system on shopping intention-rating system, and group relationship on shopping intention-group relationship, and H4, H5, H6, and H7 are established. There is a linear regression relationship between datafication (X4) and shopping intention-datafication (Y4), labeling (X5) and shopping intention-labeling (Y5), scoring system (X6) and shopping intention-scoring system (Y6), and group relationship (X7) and shopping intention-group relationship (Y7) with the regression equation:

Y_{4} = 0.777 X_{4} + 0.915

Y_{5} = 0.939 X_{5} + 0.036

Y_{6} = 0.742 X_{6} + 1.097

Y_{7} = 0.325 X_{7} + 3.115

In the 8th model, the p-value of perceived pleasantness is 0.000, which is less than 0.05, and the linear regression coefficient is greater than 0, which indicates that perceived pleasantness has a significant positive correlation effect with attitude, and H9 is established. There is a linear regression relationship between perceived pleasantness (X8) and attitude (Y8), and the regression equation.

Y_{8} = 0.522 X_{8} + 2.414

Mediated effects test

In order to test hypotheses 8 and 10, this study used Bootstrap method to conduct 5000 self-sampling tests for mediation effects using Model 4 (Model 4 is a simple mediation model) in PROCESSv4.1 version in SPSS. The results are shown in Table 8, Model 1 refers to the effect of the independent variable-algorithmic technology on the dependent variable anchor effect (

β = 1.201, Ρ = 0.000

). Therefore, algorithmic techniques can significantly and positively influence the anchor effect during consumers’ live shopping, H8 holds. Algorithmic techniques can significantly and positively influence attitudes (

β = 1.201, Τ = 9.934, Ρ < 0.01

). From Model 3, when both algorithmic techniques and attitudes enter the regression equation, algorithmic techniques (

β = 1.252, Τ = 28.641, Ρ < 0.01

) significantly positively influences the anchor effect, attitude (

β = - 0.093, Τ = - 2.290, Ρ < 0.05

) has a negative impact.

Table 8.

Regression analysis of the relationship between variables in the intermediation model (N = 268).

Variable	Model 1		Model 2		Model 3
Variable	β	T	β	T	β	T
Algorithm technology	1.201	31.887	0.559	9.934	1.252	28.641
Attitude	—	—	—	—	−0.093	−2.290
R²	0.793		0.271		0.797
F	1016.790		98.676		519.132

As can be seen from Table 9, the total effect value of the independent variable-algorithmic technology and the dependent variable anchor effect is 1.201, and its 95% confidence interval is [1.127 and 1.275], indicating that the total effect is significant; the direct effect is 1.253, and its 95% confidence interval is [1.167 and 1.339], and the direct effect is significant; and the indirect total effect is −0.052, and the 95% confidence interval is [−0.121 and 0.015], containing 0, indicating that the mediating effect is not significant, the hypothesis H10 does not hold, that is, the algorithmic technology is not able to positively affect the anchor effect in the process of users live shopping through the positive effect on the attitude, and the attitude does not play a mediating role in the algorithmic technology and the anchor effect in the process of users’ live shopping. In summary, H8 is valid and H10 is not valid. Although Hypothesis 10 assumed a positive effect of attitude on the anchor effect, the regression result showed a small negative coefficient (β = −0.093). This unexpected direction may stem from suppression effects or conceptual overlap between algorithmic intervention and users’ attitudinal responses.

Table 9.

Analysis of intermediation effects.

	Effect value	BootSE	BootLLCL	BootULCI
Total effect	1201	0.038	1.127	1.275
Direct effect	1.253	0.044	1.167	1.339
Total indirect effect	−0.052	0.035	−0.121	0.015

Conclude

Main conclusions

(1) Algorithmic technology significantly enhances consumers’ perceptions of ease of use (β = 0.581, r = 0.539, p < 0.001) by streamlining access to live-stream content, thereby improving overall user experience and attitudes toward live shopping. It also positively influences perceived behavioral control (β = 0.620, r = 0.609), indicating that users feel more capable of engaging in live-stream purchases when aided by algorithmic recommendations. These findings validate the applicability of the Technology Acceptance Model (TAM) and the Theory of Planned Behavior (TPB) in algorithm-driven environments.

(2) Beyond cognitive factors, algorithmic technology also evokes emotional responses by shaping perceived pleasantness through scene construction and anchor effects (β = 0.522). This emotional engagement strengthens users’ attitudes toward live shopping. In parallel, algorithmic support for promotional discounts demonstrates a strong effect on purchase motivation (β = 0.601 and r = 0.580), reinforcing their behavioral relevance under the TAM framework.

(3) Four specific algorithmic mechanisms—datafication, labeling, scoring systems, and group relations—exert substantial influence on consumers’ purchase intentions (β = 0.777, 0.939, 0.742, and 0.325, respectively). These mechanisms personalize the user experience, facilitate decision-making, and foster social identity within platform communities. However, the mediating role of attitude between algorithmic technology and the anchor effect was not statistically supported, suggesting that emotional influence may be more directly driven by algorithmic design rather than through attitudinal shifts alone. This interaction warrants further investigation.

(4) This study provides practical insights for platforms such as TikTok and Kuaishou. Personalized recommendations that reduce access effort may enhance perceived ease of use and user engagement. Likewise, optimizing anchor performance and scene design can improve perceived pleasantness and positively influence shopping attitudes. Moreover, datafication, labeling, and scoring mechanisms should be applied with caution to avoid bias and reinforce meaningful social connections within user communities.

Management recommendation

Based on the findings of this study, the following recommendations are proposed from the perspectives of anchors, consumers, and e-commerce enterprises.

(1) Leverage the anchor effect and enhance scene design. Anchors serve as crucial intermediaries in live-stream commerce. Their professional competence and credibility can shape consumer attitudes and drive purchasing behavior. Anchors should strengthen product knowledge and interactive skills, while platforms should ensure that scene design aligns with brand positioning and consumer expectations, incorporating interactive tools (e.g., pop-ups and likes) to boost engagement.

(2) Promote consumer literacy and rational decision-making. Public campaigns and educational programs should raise awareness about the risks and rights in live-stream shopping. Consumers are encouraged to critically assess anchor credibility, product reviews, and community feedback. Strengthening peer group communication can also foster informed, data-driven consumption choices.

(3) Empower traditional enterprises to adopt live-stream marketing. In the post-pandemic context, “e-commerce + livestreaming” has become a viable strategy for offline enterprises. By integrating live product demonstrations with online promotion, businesses can enhance brand visibility, attract new consumers, and expand digital retail channels.

Limitations and future research

This study has several limitations. First, the sample is limited to a single data collection period and platform environment, which may affect generalizability. Second, although TAM and TPB provide a strong theoretical foundation, alternative models such as UTAUT could be explored. Future research could include longitudinal studies, cross-platform comparisons, or hybrid models to further investigate algorithmic effects in live-streaming commerce.

Footnotes

ORCID iD

Yuyue Qiao

Funding

The author received no financial support for the research, authorship, and/or publication of this article.

Declaration of conflicting interests

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

References

Ying

. Feasibility study on livestreams selling of agricultural products on tik tok platform — a case study of Yingkou city in Liaoning Province. Malays E Commer J 2022; 6(1): 20–23. DOI: 10.26480/mecj.01.2022.20.23.

Tian

Chen

Pan

. Research on the influence of product information on consumers’ willingness to shop online. Adv Ind Eng Manag 2023; 12(1): 14–17.

Sarah Jannat

Abbas Al Sammarraie

Mubuke

. Exploring the mediating effect of consumers’ behavioral intention to shop online in the relationship between facilitating conditions and online shopping in developing Nations empirical evidence from Uganda’s E-commerce sector. Malays E Commer J 2023; 7(2): 83–89. DOI: 10.26480/mecj.02.2023.83.89.

Pingsheng

Yongdong

. Influence mechanism of live bandwagon marketing model on consumer purchase decision. China Circulation Economy 2020; 34(10): 38–47.

Gefen

Heart

. On the need to include national culture as a central issue in e-commerce trust beliefs. J. Glob. Inf Manag 2006; 14: 1–29.

Eurologo . Common legal issues faced by E-commerce businesses, 2016. https://www.eurologo.org/common-legal-issues-faced-by-e-commerce-businesses/ Accessed 2021-08-02.

Rantavuo

. Designing for intelligence: user-centred design in the age of algorithms. In: Proceedings of the 5th international ACM in-cooperation HCl and UX conference, Jakarta Surabaya. Bali, Indonesia: ACM, 2019.

Draude

Klumbyte

Lücking

, et al. Situated algorithms: a sociotechnical systemic approach to bias. Online Inf Rev 2019; 44(2): 325–342.

Ajzen

. From intentions to actions: a theory of planned behavior. In: Kuhl

Beckmann

(eds). Action Control: From Cognition to Behavior. New York: Springer, 1985.

10.

Ajzen

. The theory of planned behavior. Organ Behav Hum Decis Process 1991; 50(2): 179–211.

11.

Ajzen

. Consumer attitudes and behavior: the theory of planned behavior applied to food consumption decisions. Rivista di Economia Agraria/Italian Rev Agric Econ 2015; 70(2): 121–138.

12.

Lee

Bonn

Cho

. Consumer motives for purchasing organic coffee. Int J Contemp Hospit Manag 2015; 27(6): 1157–1180.

13.

Cheon

Lee

Crooks

, et al. An investigation of mobile learning readiness in higher education based on the theory of planned behavior. Comput Educ 2012; 59(3): 1054–1064.

14.

Ajzen

Fishbein

. The influence of attitudes on behavior. In: Albarracin

Johnson

Zanna

(eds). The handbook of attitudes. Mahwah, NJ: Eribaum, 2005, pp. 173–221.

15.

Davis

. A technology acceptance model for empirically testing new end-user information systems: theory and results. Cambridge, MA: Massachusetts Institute of Technology Sloan School of Management, 1986. Ph. D. Dissertation.

16.

Rogers

. Logic models in sandra mathison. In: Encyclopedia of evaluation. Beverly Hills, CA: Sage Publications, 2005, p. 232.

17.

Sharp

. Development, extension, and application: a review of the technology acceptance model. Inf. Syst. Educ. J 2007; 5(9): 1–11.

18.

Rajchakit

Sriraman

Boonsatit

, et al. Exponential stability in the Lagrange sense for Clifford-valued recurrent neural networks with time delays. Adv Differ Equ 2021; 2021: 256. DOI: 10.1186/s13662-021-03415-8.

19.

Boonsatit

Rajchakit

Sriraman

, et al. Finite-/fixed-time synchronization of delayed Clifford-valued recurrent neural networks. Adv Differ Equ 2021; 2021: 276. DOI: 10.1186/s13662-021-03438-1.

20.

Wang

Liu

. Fixed-time stability of neural networks via event-triggered control. Adv Differ Equ 2021; 2021: 250. DOI: 10.1186/s13662-021-03367-z.

21.

Zhang

Wang

, et al. Clifford-valued neural networks and synchronization under switching. Adv Differ Equ 2021; 2021: 253. DOI: 10.1186/s13662-021-03415-8.

22.

. Deep learning-based personalized recommendation system for social media platforms. Electronics 2021; 10(12): 1501. DOI: 10.3390/electronics10121501.

23.

Lan

. Survival, cognition, relationships: how algorithms will change us. Journalism 2021; 03: 45–53.

24.

Lee

Olafsson

. Two-way cooperative prediction for collaborative filtering recommendations. Expert Syst Appl 2009; 36(3): 5353–5361.

25.

Dai

Tseng

. A two-stage clustering method to analyze customer characteristics to build discriminative customer management: a case of textile manufacturing business. Expert Syst Appl 2011; 38(1): 7186–7191.

26.

Moon

Kim

. Extending the TAM for a world - wide - web context. Inf Manag 2001; 38(4): 217–230.

27.

Ahuja

Khazanchi

. Creation of a conceptual model for adoption of mobile apps for shopping from E - commerce sites – an Indian context. Procedia Comput Sci 2016; 91: 609–616.