Exploring User Intentions for Virtual Memorialization: An Integration of TAM and Social Identity in Immersive Environments

Abstract

As immersive technologies reshape how people experience identity, emotion, and loss, virtual memorialization is emerging as an important application of virtual reality. This study examines the psychological mechanisms influencing user intentions to engage in virtual memorialization by extending the Technology Acceptance Model (TAM) to incorporate Avatar Attachment and Social Identity theories. A survey of 437 participants with diverse experiences in virtual worlds and memorial practices was analyzed using structural equation modeling. The results show that Avatar Attachment (AA) and Social Identity (SI) significantly predict perceived usefulness (PU), Perceived Role Importance (PRI), and behavioral intention (BI), with PU and PRI mediating these effects. Perceived ease of use (PEOU) directly influences both PU and BI. Furthermore, perceived human-likeness (PHL) moderates the effect of AA on PU, indicating that anthropomorphic avatars enhance the perceived emotional value of memorialization. However, PHL does not moderate the AA–PRI pathway, suggesting that the salience of avatars in mourning contexts relies more on narrative identity than visual realism. This research advances the application of TAM in immersive environments and contributes to digital thanatology by highlighting the interplay between identity, emotion, and technology. The findings provide design implications for creating user-friendly and emotionally meaningful virtual memorial platforms within emerging VR ecosystems.

1. Introduction

In the digital age, the boundary between life and death has become increasingly unclear. The rise in virtual worlds, digital avatars, and posthumous digital presences has led to new socio-technical practices of mourning, remembrance, and identity extension [1,2]. Individuals now invest significant emotional and social capital in virtual characters that function as extensions of their identity and social relationships. Consequently, when these digital entities disappear or die—whether real “death” or symbolic “death”—users may experience a sense of grief and the need for commemoration [3].

Virtual memorialization broadly refers to the creation and use of digital spaces—such as websites, social media pages, VR environments, and avatar-based rituals—where deceased individuals or virtual entities are remembered, mourned, and commemorated [4,5]. These spaces, known as digital cemeteries or VR memorial halls, allow for sustained emotional bonds, collective memory, and ritualized expressions of grief [6,7]. In addition, virtual memorial practices have emerged as significant tools for navigating loss, grief, and remembrance. Functionally, virtual memorialization supports emotional expression, therapeutic reflection, identity reinforcement, and communal support—offering continuity in the form of a symbolic presence beyond physical loss [8,9]. From website and social media pages to gaming and immersive environments, online mourning has evolved from static content to dynamic, interactive environments [10,11].

Recent innovations that have expanded such practices from static online tributes to immersive and interactive formats include a range of features. Personalization (e.g., adjusting an avatar’s tone, speech, or memory triggers) has been shown to increase user identification and immersion [12]. VR and AR technologies allow users to reconstruct the deceased’s living environment or interact with AI-generated avatars [13]. VR Memorial, such as Gather. Town memorials [14], or the “Living Memories” system developed by the MIT Media Lab allow users to “enter” digitally reconstructed environments and interact with avatars modeled after the deceased [15]. These systems support emotional regulation, foster symbolic presence, and address the human need for continuity when facing loss. In addition, stereoscopic presentation can also alter basic motor performance, suggesting that immersive manipulations can directly shape user behavior and fundamentally transform user experience [16]. Therefore, creating high-quality VR environments and optimizing the stereoscopic effects of objects within them can enhance human motor control performance and perceptual abilities, thereby improving perceived ease of use (PEOU) and user satisfaction. In addition, blockchain-based memorialization systems enable decentralized control over digital legacies, while gamified rituals (e.g., virtual offerings and memorial leaderboards) offer users accessible and culturally resonant commemorative experiences [17]. The use of advanced technology enhances the way users’ data is stored in the cloud, and immersive experiences, along with personalized virtual memorial spaces, offer users a richer experience, allowing them to interact with their deceased loved ones over a longer period across space and time.

Despite growing interest, the literature remains fragmented, as research into virtual memorialization for digital death represents an academic gap. Existing research tends to emphasize narrative practices or community dynamics, but rarely models the psychological mechanisms behind users’ intention to engage in virtual memorialization—particularly in relation to attachment, identity, and system perceptions [18,19]. Moreover, few studies have integrated psychological constructs, such as Avatar Attachment, perceived usefulness, or Social Identity into predictive models of memorial behavior. Additionally, most existing research has overlooked avatar memorialization, especially when avatars act as surrogates for real or imagined identities [20,21]. Virtual avatars represent identity projection and emotional investment [20]. The “death” of an avatar can provoke complex emotional responses, similar to grief in real life, particularly when users see the avatar as an extension of themselves [12]. Regarding the relationship between virtual avatars and human behavioral intentions, existing research mainly examines how virtual avatars directly influence behavioral intentions within the metaverse. Examples include the impact of virtual avatars on Behavior Intentions such as consumption under the Extended Technology Acceptance Model (ETAM), as well as users’ acceptance of the metaverse. The TAM has not been fully applied to the field of digital mourning, and the psychological mechanisms behind human behavior remain unclear. Therefore, to fill this gap, this study combines the TAM with AA and SI to conduct scientific research on avatars in virtual memorial scenarios, exploring the underlying psychological mechanisms behind them.

In summary, given the multiple pathways and diverse factors through which virtual avatars in virtual memorialization may influence people’s behavioral intentions, we employed surveys. The collected data were analyzed using AMOS 24.0 to verify their validity. By integrating TAM and SI, we developed a novel conceptual framework to investigate the psychological mechanisms driving user participation in virtual memorialization. Although the original Technology Acceptance Model (TAM) primarily focuses on cognitive evaluations of usefulness and ease of use, it fails to capture the emotional and social dimensions that are critical in memorial practices. Therefore, this study extends TAM into the domain of virtual mourning by incorporating both Avatar Attachment (AA) and Social Identity (SI) to overcome the limited explanatory power of traditional TAM variables. Such integration provides an innovative contribution to both TAM research and the broader literature on digital death.

The structure of the paper is as follows: First, the Section 2 reviews the development and key findings of theories, such as TAM, SIT, and AA, and proposes 12 hypothetical research pathways. Second, the Section 3 describes the variables, outlines the methods used to collect valid samples, and explains how data were measured and analyzed using a five-point Likert scale. Third, the Section 4 describes the validity analysis of the model, which was conducted to demonstrate its high goodness of fit. Finally, the Section 5 and Section 6 summarize the research findings, explore the psychological mechanisms underlying user participation in virtual memorial activities, and highlight the study’s limitations and directions for future research.

2. Literature and Hypothesis

2.1. Theoretical Foundation

2.1.1. Technology Acceptance Model (TAM)

The Technology Acceptance Model (TAM), proposed by Davis in 1989, remains a foundational framework in human–computer interactions [22]. It posits that PU and PEOU are key predictors of BI when using technology. When users perceive a system as both useful and easy to use, they are more likely to adopt and engage with it.

In virtual environments, PU refers to the extent to which a system supports users’ goals—emotional expression, identity reinforcement, or social connection—while PEOU pertains to a system’s usability, intuitiveness, and operational clarity [23]. TAM has also been applied in contexts involving avatars. Research demonstrated that stronger avatar–user bonds, characterized by immersive interaction and expressive self-representation, elevated both PU and usage intention by increasing cognitive fluency [24]. These effects are particularly prevalent in emotionally charged settings such as virtual memorialization, where users may prioritize affective resonance over technical efficiency.

2.1.2. Social Identity Theory (SIT)

Social Identity Theory (SIT) posits that individuals derive a sense of self from their membership in social groups and aspire to maintain positive group distinctiveness [25]. In virtual spaces, users construct and negotiate identity through digital embodiment, particularly via avatars [26,27]. These avatars often act as visual and behavioral representations of group belonging, such as guild membership in games or participating in shared grief roles in virtual memorialization.

Intersectionality theory has been applied in research to demonstrate how avatar traits (e.g., gender and race) influence identity alignment or tension in online settings [28]. In commemorative spaces, users may adopt avatars designed to reflect the deceased or community norms—such as wearing black virtual garments or carrying digital offerings—thus reinforcing in-group cohesion and symbolic solidarity [29].

Participation in virtual mourning communities, such as joining memorial pages or VR gatherings, allows users to express their identity as grievers or supporters of a shared memory. This identity is often visually reinforced through avatar esthetics, interaction patterns, or platform rituals [30].

This study adopts SIT as its theoretical foundation to explain the role of group identity in the context of virtual memorialization. In the proposed model, the construction of SI is based on this theory by measuring users’ perceived sense of belonging, group affiliation, and identification within commemorative virtual communities.

2.1.3. Avatar Attachment (AA)

Avatar Attachment (AA) refers to users’ emotional, cognitive, and behavioral dependence on their virtual characters. It reflects the degree to which users view avatars as extensions of themselves or as emotionally significant entities [31,32]. Measurement tools such as the Avatar Attachment Scale [32] assess this construct through emotional (e.g., loss when the avatar disappears), cognitive (e.g., self-concept extension), and behavioral (e.g., time investment) dimensions.

Several factors that contribute to AA are included: (1) Customization: The more personalized an avatar’s appearance or behavior, the stronger the user’s identification and attachment [33]. (2) Interaction and Co-Presence: Social engagement—such as participating in shared rituals or receiving community feedback—can reinforce avatar-based bonding [34]. (3) Immersion and Control: Technologies such as VR or voice interaction can enhance users’ sense of embodiment, increasing psychological ownership and emotional resonance [35].

In virtual memorial contexts, AA may be formed not only with one’s own avatar but also with avatars representing a deceased individual. Users may actively design, maintain, or interact with such avatars as a form of continued bonding, making the avatar a proxy for memory and emotional expression [36]. Sah et al. (2021) further emphasize that avatars serve dual roles—as expressive tools for the self and as symbolic vessels for others [24].

2.1.4. Integrating TAM, SIT, and AA

This study proposes an integrated framework in which the TAM’s core predictors (PU and PEOU) are influenced by avatar-related factors (AA) and SI. For instance, users who strongly identify with a mourning group and feel attached to their avatars may perceive the memorial platform as more emotionally useful, even if it lacks optimal usability [37,38].

PU as mediation: AA and SI may influence BI indirectly by enhanced PU, where users feel that the system enables them to maintain identity, expression, and connection.

PRI as mediation: SI not only strengthens emotional bonds but also increases the PRI of one’s avatar in a memorial context—signaling commitment and behavioral engagement [39]. Prior research suggests that congruence between avatar-based self-expansion and identity influences users’ commitment and sustained participation in virtual environments [40,41]. In SIT, the avatar’s role becomes central when embedded in a memorial group; this perceived importance may further drive intention.

PHL as a moderator: As avatars become more anthropomorphic, their emotional and cognitive resonance with users deepens. PHL has been found to moderate the relationship between customization and emotional engagement [18,29], suggesting its potential as a boundary condition in TAM-based models of affective technology use. Perceived human-likeness may intensify the effect of AA on PU and PRI by increasing affective realism and symbolic credibility [42].

Building on the integrated framework of the Technology Acceptance Model (TAM), Social Identity Theory (SIT), and Avatar Attachment (AA), this study explores how users’ psychological and perceptual experiences affect their willingness to engage in virtual memorialization. As such, this study aims to answer the following three research questions:

RQ1: How do AA and SI influence users’ behavioral intentions to engage in virtual memorialization?

RQ2: Do PU, ease of use, and role importance mediate these relationships?

RQ3: Does perceived human-likeness moderate the effects of AA?

2.2. Research Hypothesis Development

Note on Hypothesis Numbering: Unlike the traditional sequential numbering of hypotheses (e.g., H1, H2, and H3), this study organizes its hypotheses based on clearly articulated theoretical pathways rather than numerical order. This approach is intended to enhance readability and promote deeper theoretical understanding by directly reflecting the relationships illustrated in the conceptual research model (Figure 1). By grouping hypotheses according to theoretically coherent pathways, readers can more intuitively grasp the logical foundations of each hypothesis, understand the interconnections between constructs, and follow the underlying rationale of the proposed relationships with greater ease.

2.2.1. PU, PEOU, and PRI Positively Influence BI

Within TAM, PU and PEOU are consistently found to be direct precursors of BI [22]. In emotionally charged contexts such as virtual memorialization, users evaluate the significance of their virtual identity based on the commemorative process, which we conceptualize as PRI. In emotionally rich systems like virtual memorialization, the PRI of an individual’s avatar may serve a similar motivational function, particularly in community-based rituals or when avatars embody identity-laden roles [39]. When virtual identities embody both personal identity and memory within rituals, their significance often stimulates users’ motivation to participate. Therefore, in memorial contexts, beyond traditional TAM factors, as an additional driver of BI.

H2: PU positively influences users’ BIs to engage in virtual memorialization.

H10: PEOU positively influences BI.

H4: PRI positively influences BI [27,39].

2.2.2. AA Positively Influences PU and PRI

AA refers to users’ emotional bond with their digital avatars [20]. When individuals perceive avatars as extensions of themselves, commemorating these virtual identities ceases to be a neutral act and instead becomes a ritual imbued with personal meaning. Such attachment can enhance PU, as memorialization serves as a means of maintaining identity continuity and enabling emotional expression. Similarly, Avatar Attachment may strengthen PRI, since avatars are regarded as central to one’s digital life and social interactions.

H1: AA positively influences PU [18,24,33].

H3: AA positively influences PRI [31,32,36].

2.2.3. SI Positively Influences PU and PRI

Social Identity Theory posits that individuals derive meaning and motivation from their sense of belonging in a group [25]. Within virtual memorial communities, avatars function as symbols of belonging and shared empathy in mourning. When users develop a stronger SI with their community, their motivation to participate increases; as such, participation fulfills the need for social identification and continuity. At the same time, SI reinforces PRI; since avatars embody group roles (e.g., mourner or supporter), they render their presence in rituals more meaningful (e.g., by wearing mourning attire and participating in group processions).

H5: SI positively influences PU [25,30,43].

H6: SI positively influences PRI [29,39].

2.2.4. PU, PRI, and PEOU Serve as Mediators

Although AA and SI may directly influence BIs, their effects are often realized through users’ evaluations of system quality and the significance of their online roles. PU serves a cognitive evaluative function—when users perceive a memorial system as beneficial for identity affirmation and emotional expression, they are more likely to participate in it. PRI reflects the symbolic significance of avatars within rituals, transforming emotional attachment and social identification into behavioral participation. PEOU may enhance PU, forming an indirect chain to BI.

H7: PU mediates the relationship between AA/SI and BI.

H8: PRI mediates the relationship between AA/SI and BI.

H9: PEOU positively influences PU.

2.2.5. PHL Moderates Relationships Between AA and PU/PRI

As avatars become more human-like, PHL enhances the psychological realism of avatars. When avatars more closely resemble humans in appearance and behavior, emotional attachment is more readily translated into PU, as users perceive avatars as socially and emotionally credible entities [18,42]. However, hyper-realistic avatars may also trigger discomfort if they approach the “uncanny valley” [44]. Therefore, theoretically, PHL serves as a boundary condition that can strengthen (or in some cases weaken) the effect of AA.

H11: PHL moderates the effect of AA on PU; the relationship is stronger when PHL is high [18,45].

H12: PHL moderates the effect of AA on PRI; the relationship is stronger when PHL is high [29,46].

2.2.6. Proposed Research Model

Based on the theoretical framework and prior literature, we formulated the following hypotheses (see Figure 1 and

Table 1. Summary of Hypotheses.

Hypothesis	Path	Description
H1	AA → PU	Avatar attachment positively influences perceived usefulness.
H2	PU → BI	Perceived usefulness positively influences behavioral intention.
H3	AA → PRI	Avatar attachment positively influences perceived role importance.
H4	PRI → BI	Perceived role importance positively influences behavioral intention.
H5	SI → PU	Social identity positively influences perceived usefulness.
H6	SI → PRI	Social identity positively influences perceived role importance.
H7	AA → PU → BI	Perceived usefulness mediates the relationship between avatar attachment and behavioral intention.
H8	SI → PRI → BI	Perceived role importance mediates the relationship between social identity and behavioral intention.
H9	PEOU → PU	Perceived ease of use positively influences perceived usefulness.
H10	PEOU → BI	Perceived ease of use positively influences behavioral intention.
H11	AA × PHL → PU	Perceived human-likeness positively moderates the effect of avatar attachment on perceived usefulness.
H12	AA × PHL → PRI	Perceived human-likeness positively moderates the effect of avatar attachment on perceived role importance.

for the proposed model).

By combining cognitive, emotional, and social constructs, this research provides a novel explanatory model to explore virtual mourning intentions, expanding both the TAM literature and digital death scholarship. Findings are expected to inform designers, technologists, and scholars interested in affective computing, human–avatar interactions, and the socio-psychological architecture of virtual worlds.

2.3. Technology Framework

As shown in Figure 2, this study applies TAM, AA, and SIT to virtual memorialization, proposing 12 hypothetical pathways and developing a scientific model through empirical research to explore the psychological mechanisms underlying user participation in virtual memorial activities. Research data were collected from online communities, yielding 437 valid responses, which were measured on a five-point Likert scale. The data analysis was conducted using SPSS 26.0 and AMOS 24.0, followed by mediation tests and moderation analyses. The results indicated that the structural model exhibits strong convergent and discriminant validity.

3. Methodology

3.1. Participants and Procedure

This study did not involve the use of any biological materials, reagents, or laboratory instruments. The online survey was created by a Chinese online data collection platform Wenjuanxing (Changsha Ranxing Information Technology Co., Ltd., Changsha, China). Statistical analyses were conducted using IBM SPSS Statistics 26.0 and AMOS 24.0 (IBM Corp., Armonk, NY, USA). Data visualization and figure preparation were performed using Python 3.10 with the Matplotlib 3.7.1 library (Python Software Foundation, Wilmington, DE, USA), and Adobe Illustrator 2024 (Adobe Inc., San Jose, CA, USA).

A total of 437 valid responses were collected through an online survey distributed via social media, online forums, and virtual community groups. Participants were informed of the study’s academic purpose and voluntarily consented to participate. The sample was relatively well-balanced by gender (51.49% male, 48.51% female) and diverse age groups, with the majority of participants aged between 26 and 40 years (36.38%), followed by those aged between forty-one and sixty years (24.94%) (

Table 2. Frequency analysis results.

Variable	Selection	Count	%
Gender	Male	225	51.49
	Female	212	48.51
Age	<18	38	8.70
	18–25	77	17.62
	26–40	159	36.38
	41–60	109	24.94
	>60	54	12.36
Education Level	High school/under	157	35.93
	Undergraduate	188	43.02
	Master/above	92	21.05
Experience with virtual worlds and virtual characters	Occasionally used	246	56.29
	Frequently used	117	26.77
	Long-term use	74	16.93
Experience with virtual memorial activities or services	Never participated	1	0.23
	Participated 1–2 times	331	75.74
	Participated 3 or more times	105	24.03
Total		437	100.0

). Educational backgrounds ranged from high school to postgraduate levels, with 43.02% holding a bachelor’s degree and 21.05% holding a master’s degree or higher. Although the sample was diverse in terms of age, gender, and experience, participants were primarily drawn from a Chinese sociocultural context. Consequently, cultural norms surrounding mourning, spiritual beliefs, and technology adoption may have influenced their responses.

Participants were asked about their experience with virtual worlds (e.g., games and metaverse platforms) and virtual memorial practices. Notably, 75.74% had engaged in virtual memorial services at least once, and 24.03% had engaged more than three times, suggesting broad exposure to digital mourning contexts.

3.2. Measures

All items were adapted from validated scales and measured on a five-point Likert scale (1 = strongly disagree; 5 = strongly agree) (

Table 3. Variable description.

Variable	Item	Statement
Avatar Attachment (AA) [20]	AA1	6.1: I believe that the experiences of my virtual character are like my own.
	AA2	6.2: I believe my virtual character reflects my real or ideal self.
	AA3	6.3: My virtual character is an extension of myself in the virtual world.
	AA4	6.4: My virtual character holds special emotional meaning to me.
	AA5	6.5: I feel a strong sense of loss when my virtual character disappears.
Social Identity (SI) [25,43]	SI1	7.1: Using virtual characters (avatars) helps me integrate into specific communities.
	SI2	7.2: I identify with my role in virtual communities/groups.
	SI3	7.3: I actively participate in the activities of virtual communities where my avatar belongs, and group recognition is important to me.
	SI4	7.4: I believe the virtual community activities that participated by my virtual character (avatar) could influence my real life.
	SI5	7.5: I feel a sense of belonging in my virtual communities.
Perceived Usefulness (PU) [22,47]	PU1	8.1: I believe virtual memorial activities help me express emotions for my virtual character.
	PU2	8.2: I feel that using virtual memorialization can ease the sense of loss caused by a character’s disappearance.
	PU3	8.3: I think virtual memorialization enhances my sense of belonging to a virtual community.
	PU4	8.4: I believe virtual memorialization helps me better understand my identity in the virtual world.
Perceived Ease of Use (PEOU) [22]	PEOU1	9.1: I find the virtual memorial activities I’ve experienced easy to use.
	PEOU2	9.2: I can use virtual memorial functions fluently without much effort.
	PEOU3	9.3: I believe anyone can easily learn to use virtual memorial features.
	PEOU4	9.4: I can quickly find the features I need when using or participating in virtual memorial services/activities.
Perceived Role Importance (PRI) [20]	PRI1	10.1: I believe virtual characters play an important role in my digital life.
	PRI2	10.2: My virtual character significantly affects my digital social relationships.
	PRI3	10.3: I believe my virtual character represents certain aspects of my identity or values.
	PRI4	10.4: The disappearance or death of a virtual character has a major impact on my virtual life.
Behavioral Intention (BI) [48]	BI1	11.1: I am willing to use virtual memorial platforms to commemorate important virtual characters.
	BI2	11.2: I am willing to pay for virtual character memorial features.
	BI3	11.3: I am willing to participate long-term in memorial activities for virtual characters.
	BI4	11.4: If my friends participate in virtual memorialization, I am more likely to join as well.
Perceived Human-Likeness (PHL) [18,49]	PHL1	12.1: I feel that virtual characters resemble real human beings.
	PHL2	12.2: When interacting with virtual characters, I experience a sense of real interpersonal interaction.
	PHL3	12.3: I am more likely to form emotional bonds with virtual characters that look and behave like humans.

).

3.3. Data Analysis

All statistical analyses were conducted using SPSS 26.0 and AMOS 24.0. Descriptive statistics were first calculated to examine the distributional properties of the data, including means, standard deviations, skewness, and kurtosis. The results indicated acceptable levels of normality for all constructs (skewness < |3|, kurtosis < |10|), confirming their suitability for SEM analysis. Reliability was assessed using Cronbach’s alpha, with all constructs exceeding the recommended threshold of 0.70, indicating high internal consistency.

To address the proposed hypotheses and research questions, we employed structural equation modeling (SEM) as it enables the simultaneous estimation of multiple direct, mediating, and moderating pathways within a single theoretical model, which was essential for testing the extended TAM framework in this study. The analysis proceeded through four sequential steps aligned with the research questions:

Measurement validation: Confirmatory factor analysis (CFA) was first conducted to examine the reliability and validity of all latent constructs (AA, SI, PU, PEOU, PRI, BI, and PHL). Composite reliability (CR) and average variance extracted (AVE) were also computed to confirm convergent validity. This step ensured that the scales adequately captured the intended theoretical dimensions.

Structural model testing (RQ1): The direct pathways hypothesized between AA, SI, PU, PEOU, PRI, and BI were tested using SEM. This step evaluated the core relationships in the extended TAM.

Mediation analysis (RQ2): Bootstrapping procedures (5000 resamples and 95% confidence intervals) were employed to examine the mediating roles of PU, PEOU, and PRI in the connection between both AA and SI and BI. This allowed us to assess whether cognitive and role-related appraisals transmit the effects of emotional and social antecedents to behavioral outcomes.

Moderation analysis (RQ3): Hierarchical regression was conducted to test the moderating effects of PHL on the AA → PU and AA → PRI pathways. This step evaluated whether anthropomorphic features strengthen or weaken the translation of Avatar Attachment into perceived system value or role significance. To test for potential multicollinearity, variance inflation factors (VIFs) and tolerance values were calculated for both construct-level and item-level predictors. All construct-level VIFs ranged from 1.177 to 1.478 (Tolerance = 0.677–0.850), and item-level VIFs ranged from 1.757 to 2.414, well below the conventional threshold. These results suggest that multicollinearity is unlikely to bias the moderation analyses (see

Table A1. Collinearity Test.

Level	Variable	VIF	Tolerance
Construct	AA	1.319	0.758
Construct	SI	1.461	0.685
Construct	PU	1.310	0.764
Construct	PEOU	1.337	0.748
Construct	PRI	1.370	0.730
Construct	BI	1.478	0.677
Construct	PHL	1.177	0.850
Item	AA1	2.181	0.459
Item	AA2	2.414	0.414
Item	AA3	2.052	0.487
Item	AA4	2.278	0.439
Item	AA5	2.044	0.489
Item	SI1	2.142	0.467
Item	SI2	2.006	0.498
Item	SI3	2.274	0.440
Item	SI4	1.959	0.511
Item	SI5	1.988	0.503
Item	PU1	2.011	0.497
Item	PU2	1.757	0.569
Item	PU3	1.878	0.533
Item	PU4	1.799	0.556
Item	PEOU1	1.861	0.537
Item	PEOU2	1.944	0.514
Item	PEOU3	1.991	0.502
Item	PEOU4	1.903	0.525
Item	PRI1	1.791	0.558
Item	PRI2	1.996	0.501
Item	PRI3	1.954	0.512
Item	PRI4	2.021	0.495
Item	BI1	2.119	0.472
Item	BI2	2.385	0.419
Item	BI3	2.229	0.449
Item	BI4	2.296	0.436
Item	PHL1	1.775	0.563
Item	PHL2	1.816	0.551
Item	PHL3	1.811	0.552

for details).

This design ensured a systematic correspondence between the research questions, the hypotheses, and the analytic procedures, moving between construct validation to direct effects, mediation, and moderation.

4. Results

4.1. Descriptive Statistics and Reliability

This study employed the Maximum Likelihood Estimation (MLE) method within the framework of structural equation modeling (SEM) to evaluate the relationships between the measured constructs. MLE assumes that the observed variables follow a multivariate normal distribution. This assumption is typically assessed by examining the skewness and kurtosis of each variable, with attention particularly given to the extent these values deviate from zero. In practical terms, a dataset is generally considered to conform to an ideal normal distribution when both skewness and kurtosis are less than 10. However, due to various forms of measurement errors and flaws in real-world data, it is often difficult to achieve perfect normality. As a widely accepted standard in empirical analysis, when both skewness and kurtosis values fall within an absolute value of three, the data can be considered approximately normally distributed.

As shown in

Table 4. Descriptive statistical analysis results.

Variable	Cronbach’s α	Min	Max	Mean	SD	Kurtosis	Skewness	AVE	CR
AA	0.879	1.2	5	3.709	0.921	−0.581	−0.940	0.593	0.879
SI	0.867	1.2	5	3.769	0.992	−0.178	−1.125	0.567	0.867
PU	0.824	1	5	3.805	0.963	0.118	−1.105	0.540	0.824
PEOU	0.834	1	5	3.797	0.978	0.073	−1.127	0.558	0.835
PRI	0.833	1	5	3.810	0.948	0.237	−1.184	0.556	0.834
BI	0.864	1	5	3.590	1.093	−0.871	−0.744	0.614	0.864
PHL	0.797	1	5	3.849	0.967	0.356	−1.151	0.567	0.797

, all constructs demonstrated acceptable skewness (less than |3|) and kurtosis (less than |10|), indicating that the data met the assumptions of approximate normality. The means ranged from 3.590 for BI to 3.849 for PHL, confirming that each item fulfilled the statistical requirements for further analysis. Thus, the survey data were deemed suitable for subsequent statistical procedures.

To assess the reliability of the questionnaire, Cronbach’s alpha was adopted as the primary indicator. This coefficient is widely used in empirical research to measure the internal consistency of multi-item scales. A Cronbach’s alpha value below 0.7 typically indicates poor internal consistency, suggesting that the items may not be measuring the same underlying construct and that the scale may require revision. Values above 0.7 indicate acceptable internal consistency, while values exceeding 0.9 are considered excellent. In addition to Cronbach’s alpha, this study also employed the Corrected Item–Total Correlation (CITC) method to assess the reliability of individual items. In this context, an item is considered for deletion if two conditions are met simultaneously: the CITC value for that item is less than 0.4, and removing the item would result in an increased Cronbach’s alpha value for the corresponding scale.

The results showed a Cronbach’s alpha coefficient of 0.879 for AA, 0.867 for SI, 0.824 for PU, 0.834 for PEOU, 0.833 for PRI, 0.864 for BI, and 0.797 for PHL. All values exceeded the 0.7 threshold, and no items met the deletion criteria based on CITC or an increase in alpha. These findings confirm that the constructs measured in the questionnaire possess high internal stability and that the reliability of the instrument met the requirements for empirical validation.

Notably, BI had the lowest mean value (M = 3.590), suggesting that users were relatively cautious in their willingness to engage in virtual memorialization. In contrast, PHL had the highest mean value (M = 3.849), implying that participants were generally receptive to avatars with anthropomorphic qualities. These findings suggest a balance between users’ emotional openness to human-like avatars and their more tentative behavioral commitment.

4.2. Validity Analysis

This study conducted a confirmatory factor analysis (CFA) using AMOS 24.0 to assess the measurement models of the following constructs: AA, SI, PU, PEU, PRI, BI, and PHL (Figure 3). The analysis evaluated the structures of these multi-item scales and assessed the convergent validity of each construct through composite reliability (CR) and average variance extracted (AVE).

Model fit was first assessed for CFA using the Maximum Likelihood Estimation method. The survey data were imported into AMOS 24.0, and the resulting model fit indices are presented in

Table 5. Model fit of CFA.

Model Fit	CMIN	DF	CMIN/DF	NFI	RFI	IFI	TLI	CFI	GFI	RMSEA
Fit Results	367.246	356	1.032	0.939	0.930	0.998	0.998	0.998	0.947	0.009
Suggestion			<3	>0.9	>0.9	>0.9	>0.9	>0.9	>0.9	<0.08

. CFA indicated an excellent model fit, with CMIN/DF = 1.032, Comparative Fit Index (CFI) = 0.998, and Root Mean Square Error of Approximation (RMSEA) = 0.009. To assess potential common-method bias, we conducted a single-factor CFA by loading all items onto a common factor (see the Figure A1). The single-factor model (

Table A2. Single-factor model fit.

Model Fit	CMIN	DF	CMIN/DF	NFI	RFI	IFI	TLI	CFI	GFI	RMSEA
Fit Results	3157.195	377	8.375	0.474	0.434	0.506	0.465	0.503	0.587	0.130
Suggestion			<3	>0.9	>0.9	>0.9	>0.9	>0.9	>0.9	<0.08

) fit the data very poorly (e.g., CMIN/DF ≈ 8.38, CFI ≈ 0.50, RMSEA ≈ 0.13) and was substantially worse than the seven-factor model, indicating that a single underlying factor does not account for the item covariances. Therefore, common-method variance is unlikely to threaten the validity of our findings. In addition, discriminant validity was verified using the Fornell–Larcker criterion: the square roots of AVE (bold diagonal) exceeded the corresponding inter-construct correlations for all constructs (

Table 6. Discriminant Validity: Fornell–Larcker Matrix.

Variable	AA	SI	PU	PEOU	PRI	BI	PHL
AA	0.770
SI	0.379	0.753
PU	0.329	0.295	0.735
PEOU	0.279	0.421	0.274	0.747
PRI	0.273	0.360	0.365	0.292	0.746
BI	0.386	0.409	0.361	0.361	0.392	0.784
PHL	0.208	0.176	0.252	0.259	0.299	0.272	0.753

Note: The bold numbers on the diagonal represent the square root of the AVE values.

). Taken together, these results suggest that measurement quality is adequate and that CMB is unlikely to bias the structural estimates.

All constructs demonstrated acceptable convergent validity. The AVE values ranged from 0.540 to 0.614, exceeding the recommended threshold of 0.50, while the CR values ranged from 0.797 to 0.879, all exceeding the recommended minimum of 0.70. These results indicate that the measurement model exhibited both strong internal consistency and convergent validity, confirming the reliability and validity of the constructs assessed in this study.

As such, the constructs used in this study reliably capture underlying theoretical concepts. In particular, the high CR values for AA and SI reflect the robustness of these measures in capturing the emotional and social dimensions of avatar-related experiences.

4.3. Structural Model

Based on the results of the measurement model, the structural equation model consists of seven latent variables and a total of 36 observed indicators (Figure 4). The questionnaire data were imported into AMOS 24.0, and the model was estimated using the Maximum Likelihood Estimation (MLE) method. The model fit indices obtained are presented in

Table 7. Model fit of SEM analysis.

Model Fit	CMIN	DF	CMIN/DF	NFI	RFI	IFI	TLI	CFI	GFI	RMSEA
Fit Results	336.619	286	1.177	0.939	0.930	0.990	0.989	0.990	0.946	0.020
Suggestion			<3	>0.9	>0.9	>0.9	>0.9	>0.9	>0.9	<0.08

.

Before conducting pathway analysis, we first analyzed the model’s fit. As shown in Table 5, the CMIN/DF value is 1.177, which is less than 3; the RMSEA value is 0.020, which is less than 0.05; and the values of NFI, RFI, IFI, TLI, CFI, and GFI are all greater than 0.9, indicating that the model has a good level of fit.

As shown in

Table 8. Path coefficient test of the SEM.

Path			Regression Weight	S.E.	C.R.	p	β	Conclusion
PU	←	AA	0.247	0.057	4.349	***	0.269	Supported
PU	←	SI	0.155	0.063	2.458	0.014	0.167	Supported
PRI	←	AA	0.166	0.059	2.812	0.005	0.167	Supported
PRI	←	SI	0.367	0.063	5.848	***	0.367	Supported
PU	←	PEOU	0.152	0.064	2.379	0.017	0.155	Supported
BI	←	PEOU	0.188	0.068	2.739	0.006	0.161	Supported
BI	←	PU	0.189	0.067	2.838	0.005	0.159	Supported
BI	←	PRI	0.235	0.062	3.797	***	0.213	Supported
BI	←	AA	0.206	0.062	3.307	***	0.189	Supported
BI	←	SI	0.186	0.072	2.576	0.010	0.169	Supported

Note: “←” indicates the direction of the structural path from the independent variable to the dependent variable; *** refers to p < 0.001; β = standardized regression weights.

, all path coefficients yielded critical ratios (CRs) with corresponding p-values below 0.05, indicating that all structural paths are statistically significant at the 95% confidence level. These results suggest a high degree of structural fit within the model.

Following the confirmation of an acceptable model fit, hypothesis testing was conducted on the proposed structural relationships using AMOS 24.0. The results of the pathway analysis are presented in Table 6. Several direct effects were found to be statistically significant.

Specifically, AA had a significant positive effect on PU, with a standardized path coefficient of 0.269 (p < 0.001). SI also exerted a significant positive influence on PU, with a standardized coefficient of 0.167 (p = 0.014). Furthermore, AA significantly influenced PRI, with a standardized path coefficient of 0.167 (p = 0.005), while SI also had a strong positive effect on PRI, with a coefficient of 0.367 (p < 0.001).

The influence of PEOU had a significant impact on PU (β = 0.155, p = 0.017) and BI (β = 0.186, p = 0.002). PU also significantly influenced BI, with a path coefficient of 0.161 (p = 0.006). In addition, PRI showed a significant positive effect on BI (β = 0.213, p < 0.001).

AA also directly affected BI, with a standardized coefficient of 0.169 (p = 0.010), while SI showed a significant positive relationship with BI as well (β = 0.196, p = 0.001).

To facilitate interpretation, the figure below (Figure 5) presents a forest plot of standardized path coefficients with 95% bias-corrected bootstrap confidence intervals extracted from AMOS. The total effects chart (Figure 6) displays the combined direct and indirect effects of predictors on BI, with error bars representing 95% confidence intervals.

These results support the hypothesized relationships among the latent variables and confirm the robustness of the theoretical model in explaining users’ behavioral intentions toward virtual memorialization.

4.4. Mediation Tests

This study employed the Bootstrap method to test the mediating effects of Organizational Commitment and Job Satisfaction. The sample size was set to 5000, with a 95% confidence level, using the biased correction confidence interval as the standard, and the upper and lower bounds were observed. When the biased corrected confidence interval of the indirect effect did not include 0 and p < 0.05, it indicated the presence of a mediating effect. The results are provided in the table below.

As shown in

Table 9. Mediation test results.

Mediation Path	Effect Type	Effect Value	Lower	Upper	p
AA => PU => BI	Indirect Effect 1	0.043	0.010	0.095	0.008
	Direct Effect 1	0.189	0.049	0.324	0.008
	Total Effect 1	0.231	0.101	0.356	0.001
PEOU => PU => BI	Indirect Effect 2	0.025	0.002	0.074	0.017
	Direct Effect 2	0.161	0.036	0.303	0.014
	Total Effect 2	0.186	0.057	0.330	0.007
SI => PU => BI	Indirect Effect 3	0.027	0.005	0.071	0.015
	Direct Effect 3	0.169	0.011	0.321	0.040
	Total Effect 3	0.196	0.037	0.348	0.015
AA => PRI => BI	Indirect Effect 4	0.036	0.010	0.086	0.004
	Direct Effect 4	0.189	0.049	0.324	0.008
	Total Effect 4	0.220	0.060	0.387	0.007
SI => PRI => BI	Indirect Effect 5	0.078	0.027	0.159	0.002
	Direct Effect 5	0.169	0.011	0.321	0.040
	Total Effect 5	0.247	0.095	0.393	0.002

Note: “=>” indicates the direction of mediation paths from the independent variable to the dependent variable.

, several significant mediation effects were identified in the model. First, in the path from AA to BI through PU, the direct effect was 0.189 (p < 0.05), and the indirect effect was 0.043 (p < 0.05), yielding a total effect of 0.231 (p < 0.05). The 95% confidence intervals for all effects did not include zero, indicating that PU significantly mediated the relationship between AA and BI. Therefore, the mediation hypothesis is supported.

In the path from PEOU to BI via PU, the direct effect was 0.161 (p < 0.05), and the indirect effect was 0.025 (p < 0.05), with a total effect of 0.186 (p < 0.05). Since all confidence intervals excluded zero, PU was shown to play a partial mediating role in the relationship between PEOU and BI, confirming the hypothesis.

For the mediation pathway from SI to BI through PU, the direct effect was 0.169 (p < 0.05), and the indirect effect was 0.027 (p < 0.05), resulting in a total effect of 0.196 (p < 0.05). Again, all confidence intervals excluded zero, indicating that PU served as a significant mediator between SI and BI.

Regarding the mediating role of PRI, the path from AA to BI through PRI showed a direct effect of 0.189 (p < 0.05), an indirect effect of 0.036 (p < 0.05), and a total effect of 0.220 (p < 0.05). The confidence intervals for all effects did not include zero, supporting the hypothesis that PRI mediates the relationship between AA and BI.

Similarly, for the pathway from SI to BI via PRI, the direct effect was 0.169 (p < 0.05), and the indirect effect was 0.078 (p < 0.05), resulting in a total effect of 0.247 (p < 0.05), with all confidence intervals excluding zero. This provides strong evidence that PRI mediates the effect of SI on BI.

In summary, the results highlight that PRI is a central mechanism through which AA and SI influence BI. PU partially mediates the effect of AA on BI, suggesting that attachment enhances perceptions of usefulness, which, in turn, increases user intentions. PU partially mediated the effect of PEOU on BI, indicating that ease of use enhances usefulness, which subsequently shapes intention. PU also mediated the relationship between SI and BI, confirming that social belonging is translated into the adoption of online practices via cognitive appraisals of usefulness. These results demonstrate that PU and PRI are critical bridges between affective/social antecedents and behavioral outcomes in digital mourning contexts.

4.5. Moderation Analysis

4.5.1. Moderations on PU

To examine the moderating role of PHL, hierarchical regression analysis was conducted using three models (

Table 10. (AA × PHL → PU) Moderation analysis result.

Variable	Model 1	Model 2	Model 3
Constant	3.805 ** (87.388)	3.805 ** (89.072)	3.785 ** (86.743)
AA	0.310 ** (7.271)	0.273 ** (6.370)	0.268 ** (6.262)
PHL		0.191 * (4.234)	0.186 ** (4.119)
AA × PHL			0.097 * (2.070)
Sample Size	437	437	437
R2	0.108	0.144	0.152
Adjust R2	0.106	0.140	0.146
F	F (1435) = 52.875, p = 0.000	F (2434) = 36.431, p = 0.000	F (3433) = 25.900, p = 0.000
ΔR2	0.108	0.035	0.008
ΔF	F (1435) = 52.875, p = 0.000	F (1434) = 17.930, p = 0.000	F (1433) = 4.286, p = 0.039

Note: Dependent variable = PU. * p < 0.05 ** p < 0.01 (t).

):

Model 1 included the independent variable, AA, to examine its direct effect on the dependent variable, PU, without considering any moderating influences. The primary objective of this model was to assess the impact of AA on PU in the absence of the moderator, PHL, and establish a baseline relationship.

Model 2 then introduced PHL as a potential moderator, allowing its direct effect on PU to be evaluated alongside AA. Finally, Model 3 incorporated the interaction term (AA × PHL) to test whether PHL significantly moderates the relationship between AA and PU. By comparing the changes in explained variance (R²) and examining the significance of the interaction term, the moderation effect of PHL was verified.

The interaction between AA and PHL significantly predicted PU (β = 0.097, t = 2.070, p < 0.05), indicating a high-level moderation effect. The change in R² from Model 2 to Model 3 was also statistically significant (ΔR² = 0.008, ΔF = 4.286, p = 0.039). This suggests that when users perceive higher levels of human-likeness in virtual characters, the positive influence of AA on PU becomes stronger.

4.5.2. Moderations on PRI

As shown in the table below (

Table 11. (AA × PHL → PRI) Moderation analysis results.

Variable	Model 1	Model 2	Model 3
Constant	3.810 ** (87.219)	3.810 ** (90.167)	3.810 ** (87.833)
Avatar Attachment	0.253 ** (5.908)	0.204 ** (4.819)	0.204 ** (4.804)
PHL		0.249 ** (5.559)	0.249 ** (5.541)
AA × PHL			0.001 (0.023)
Sample Size	437	437	437
R2	0.074	0.136	0.136
Adjust R2	0.072	0.132	0.130
F	F (1435) = 34.899, p = 0.000	F (2434) = 34.100, p = 0.000	F (3433) = 22.681, p = 0.000
ΔR2	0.074	0.062	0.000
ΔF	F (1435) = 34.899, p = 0.000	F (1434) = 30.902, p = 0.000	F (1433) = 0.001, p = 0.982

Note: Dependent variable = PRI. ** p < 0.01 (t).

), moderation analysis was conducted using a three-step hierarchical regression approach. Model 1 included only the independent variable, AA, to examine its direct effect on the dependent variable, PRI, without accounting for the influence of the moderator. Model 2 added the moderator, PHL, to assess its direct effect on PRI alongside AA. Finally, Model 3 incorporated the interaction term (AA × PHL) to test whether PHL significantly moderates the relationship between AA and PRI.

The results indicate that the interaction effect between AA and PHL on PRI was not statistically significant (β = 0.001, t = 0.023, p = 0.982). Furthermore, there was no notable change in the explained variance between Model 2 and Model 3 (ΔR² = 0.000, ΔF = 0.001, p = 0.982), suggesting that the inclusion of the interaction term did not improve the model’s explanatory power.

This suggests that while anthropomorphic design features enhance perceived usefulness, they do not necessarily alter perceptions of the importance of online roles, which are shaped more by symbolic and communal meanings than by visual realism.

The figure shown below (Figure 7) presents the simple slopes plots for the moderation analyses. As shown in Plot 1 (moderation on PU), the slope of AA predicting PU increased from low (−1 SD) to high (+1 SD) levels of PHL, indicating that avatar attachment translated more strongly into perceived usefulness when avatars were perceived as more humanlike. In contrast, Plot 2 (moderation on PRI) shows that the slopes of AA predicting PRI were nearly identical across PHL levels, suggesting no moderation effect.

These results confirm that PHL is a selective moderator, influencing cognitive but not symbolic evaluations.

5. Discussion

5.1. Key Findings and Interpretations

This study reveals three core findings. First, both AA and SI exert significant influence on PU and PRI. Second, PU and PEOU remain the key determinants of BI, underscoring the robustness of the traditional TAM framework even in emotionally driven contexts. Third, PHL moderates only the AA → PU pathway, indicating that while anthropomorphic realism can enhance perceived usefulness, it does not affect the user’s perception of role importance.

First, the significant influence of AA suggests that virtual memorial platforms are perceived more as emotional extensions than as functional tools. When users experience a deep integration between their avatars and personal identity, commemorative acts gain value by sustaining continuity between an individual’s real and digital selves. This study also found that SI exerts the strongest influence on PRI, highlighting the collective dimension of mourning. Group belonging within virtual communities endows avatars with ritualistic meaning—through wearing mourning attire or participating in collective remembrance—thereby reinforcing their symbolic function in mourning practices. These results underscore the crucial role of emotional and social bonds between virtual personas in shaping users’ perceptions of value and their commitment to virtual memorialization, echoing findings in the digital identity literature [20,48].

Second, PU and PEOU act as significant mediators in multiple pathways, and PRI mediates the relationships between both AA and SI and BI. These findings highlight the influence of cognitive appraisals—particularly usefulness and the importance of digital roles—which act as psychological bridges linking affective and social antecedents to behavioral outcomes in digital mourning contexts. Importantly, PEOU demonstrated direct effects on both PU and BI, emphasizing the pivotal role of system usability and operational simplicity in facilitating engagement with virtual memorial platforms. This expands and reinforces the applicability of the TAM framework within the relatively underexplored context of digital death and commemoration.

Third, PHL significantly moderates the relationship between AA and PU: the more human-like a virtual character is perceived to be, the stronger the influence of emotional attachment is on PU. This finding aligns with the media equation theory and prior work in human–computer interactions [18], suggesting that anthropomorphic design features can enhance both emotional resonance and perceived functionality in digital mourning systems. However, PHL did not moderate the link between AA and PRI, indicating that while anthropomorphic avatars heighten emotional resonance and usefulness, the symbolic weight of avatars in memorialization is not dependent on visual realism. Instead, the influence of a user’s online role draws on narrative identity, communal participation, and ritual performance—factors that transcend surface-level human-likeness.

5.2. Theorical Contributions

The results indicate that incorporating AA and SI into the TAM provides greater explanatory power compared to the original structure. By integrating these variables, this study moves beyond the TAM’s purely cognitive perspective and instead covers the emotional and social mechanisms underlying digital mourning behaviors. Although AA and SI represent distinct dimensions, they jointly shape behavioral intentions in complementary ways; on the other hand, PHL functions only as a boundary condition and not as a universal enhancer. The integration of SI and PRI further reveals how group belonging and the importance of online roles both mediate the adoption of technology in identity-driven contexts. While PU and PEOU remain strong predictors of BI, both AA and SI account for additional variance by explaining users’ emotional bonds with avatars and their collective identification in mourning rituals. This suggests that when the TAM is applied to emotionally charged and identity-salient contexts such as virtual memorialization, the incorporation of psychosocial constructs can substantially strengthen its explanatory capacity.

5.3. Practical Implications

These findings have several practical implications for designers and platform developers in the virtual world and metaverse industries.

Foster emotional engagement through avatar-linked memory construction: Virtual memorial systems should prioritize strengthening AA by offering features such as shared memory logs, co-created narratives, and interactive remembrance journals. Facilitating long-term emotional bonds between users and their virtual characters can enhance the perceived value and authenticity of the memorial experience.

Accessibility and inclusivity: Accessibility and cost remain critical barriers to the widespread adoption of VR memorialization. Immersive head-mounted displays (e.g., Oculus Quest and HTC Vive) require substantial financial investment. Although companies such as PICO in China are making efforts to lower economic and usability thresholds, high-performance hardware is still unevenly distributed across socioeconomic groups. This disparity particularly restricts opportunities for older populations and those in developing regions to experience fully immersive rituals. To address these challenges, platform designers should adopt inclusive strategies, such as enabling multi-device compatibility (VR headsets, PCs, and mobile devices) to reduce entry barriers. Case studies, such as Gather. Town memorial spaces and the MIT Media Lab’s Living Memories project, demonstrate how platforms can balance usability and emotional resonance. Enhancing usability by optimizing interfaces, simplifying navigation, and streamlining interaction processes can significantly improve perceived ease of use (PEOU), which, in turn, fosters greater behavioral intention (BI). In emotionally sensitive contexts such as mourning, where users may lack the patience or capacity to overcome technical friction, ensuring low-threshold participation through intuitive design and cross-platform accessibility is particularly crucial.

Balance anthropomorphism with meaning-making: While human-likeness features (e.g., expressive facial animations, voice synthesis, and memory persistence) can reinforce emotional attachment and enhance PU, designers must move beyond mere visual realism. Instead, focus should be placed on constructing meaning—facilitating emotional expression, identity projection, and ritualized interactions. The effectiveness of a memorial avatar lies more in its symbolic resonance than in its physical resemblance to human beings. For instance, VRChat memorial ceremonies often rely on symbolic gestures, shared spaces, and co-created rituals rather than hyper-realistic avatar design. Participants use stylized or even fantastical avatars to perform mourning practices, suggesting that the emotional depth of these experiences derives from collective meaning-making and narrative resonance rather than surface-level anthropomorphism [50].

Embed SI cues and community-based features: Platforms should strengthen group-based memorial functions, such as collective ceremonies, shared mourning spaces, and community remembrance walls. These features support SI formation and collective grieving, validating the user’s role in a wider digital community and enhancing the perceived importance of virtual memorialization.

By integrating these design principles, VR memorial platforms can significantly enhance inclusivity, emotional engagement, and cultural adaptability.

5.4. Limitations and Future Research

Several limitations should be acknowledged. First, the cross-sectional nature of this study prevents causal inferences. Future research could employ longitudinal or experimental designs to trace the development of attachment and memorial behavior over time. Second, this study was conducted within a Chinese sociocultural context, where traditional values—such as collectivism and Confucian, Buddhist, and Taoist beliefs—may shape attitudes toward memorial practices and death. Therefore, caution should be exercised when generalizing these findings to other cultural contexts. Comparative studies across cultures (for example, contrasts between East–West and secular and religious contexts) could enrich our understanding of different global digital grief practices. Third, our model did not assess emotional states such as the intensity of grief, nostalgia, or guilt, which are known to influence mourning behavior. Incorporating affective predictors may offer a more comprehensive view of user intention. Finally, future studies could explore algorithmic mechanisms, including how AI-driven avatars (e.g., those mimicking deceased individuals) shape users’ perceptions of usefulness and ethical boundaries.

6. Conclusions

This study provides empirical support for an expanded TAM framework that incorporates the emotional and social dimensions of avatar use in virtual memorial contexts. Our model demonstrates how AA, SI, and PRI shape users’ willingness to engage in commemorative practices for digital entities.

Our findings highlight AA as a critical antecedent, directly and indirectly shaping users’ PU, PRI, and BI. SI also emerged as a powerful predictor, particularly in shaping users’ visibility, reaffirming the importance of collective belonging in virtual commemoration. The cognitive appraisals of PU and PEOU were confirmed as mediating variables, while PHL played a moderating role in enhancing the effect of AA on PU, not on PRI. These nuanced results suggest that, while emotional realism may enhance perceived functionality, deeper perceptions of users’ roles and their meaning in virtual contexts rely more on social continuity than visual anthropomorphism.

By extending the TAM framework to the emotionally complex context of digital death, this research contributes theoretically to studies on human–computer interactions, digital grief, and avatar psychology. Practically, it offers guidance for platform developers and designers aiming to build emotionally resonant, user-friendly, and socially meaningful virtual memorial experiences. Future research should examine the longitudinal development of digital mourning, cross-cultural variations in virtual grief rituals, and the ethical implications of algorithmically generated posthumous avatars.

In a world where identity, memory, and loss increasingly unfold in digital spaces, understanding how and why we mourn through avatars not only deepens our insight into human–technology relations, but can also help shape more empathetic and humane virtual futures.