Determinants of student satisfaction with AI-powered chatbots in higher education: a mixed-methods study at Khanh Hoa University

I. INTRODUCTION

The integration of artificial intelligence (AI) into education has reshaped teaching and learning paradigms globally, with AI-powered chatbots emerging as transformative tools for instructional delivery and learner support (Okonkwo & Ade-Ibijola, 2021). In Vietnam, where digital transformation is aligned with national education development strategies (Nguyen et al., 2023), AI-based chatbots offer scalable solutions to personalize learning, enhance engagement, and mitigate disparities in educational access. Despite their potential, user satisfaction with chatbots remains inconsistent due to concerns about accuracy, adaptability, and trustworthiness (Sáiz-Manzanares et al., 2023). This study examines the factors influencing student satisfaction with learning support chatbots at Khanh Hoa University, addressing critical gaps in the localized implementation of AI-enhanced educational technologies.

1.1. Chatbots in education: opportunities and challenges

Chatbots simulate human-like interaction through natural language processing, enabling real-time responses to learner inquiries, guiding problem-solving, and supporting self-directed learning (Pérez et al., 2020). A growing body of international research supports the effectiveness of these approaches in improving student motivation, engagement, and academic outcomes (Roca et al., 2024). For instance, Vanichvasin (2022) reported that chatbot-facilitated expert dialogue models improved conceptual understanding by 22% among STEM learners. In the Vietnamese context, adoption is rising Ngo et al. (2024) found that 68% of students using GPT-based chatbots reported improved knowledge retention.

Learners often question the reliability of chatbot-generated content due to its reliance on unverifiable or non-academic sources (Bui & Nguyen, 2024), and educators face challenges in integrating it pedagogically (Tuan & Phuong, 2018). These limitations highlight the need for context-specific evaluations of user satisfaction and adoption factors, particularly within emerging educational ecosystems.

1.2. Research gap and objectives

Although the educational potential of chatbots has been widely examined in Western contexts (e.g., Cunningham-Nelson et al., 2019), relatively few studies have focused on Southeast Asia, where digital infrastructure, educational policy, and learner behavior differ significantly. At Khanh Hoa University, preliminary institutional data indicate skepticism toward chatbot efficacy, with only 32.6% of students rating chatbot-generated content as “highly accurate” (see Section 4.2). To address this gap, this study investigates the following research questions:

- How do students evaluate the usability and utility of learning support chatbots?

- Which factors (e.g., accuracy, response speed) most significantly influence student satisfaction?

- What is the perceived impact of chatbot usage on learning outcomes?

1.3. Theoretical and practical contributions

This study is grounded in two established frameworks: the Technology Acceptance Model (TAM) (Davis, 1989) and the Information Systems Success Model (ISSM) (DeLone & McLean, 2003). TAM emphasizes the role of perceived usefulness and ease of use in technology adoption. At the same time, ISSM extends this by incorporating system quality measures such as accuracy, response time, and information reliability.

Findings from this research aim to guide Vietnamese universities and regional institutions in:

- Designing AI chatbots with localized content validation mechanisms to improve trust and accuracy;

- Training educators to integrate chatbots effectively into teaching practices.

- Prioritizing user-centered features such as native language support and discipline-specific knowledge bases.

By connecting theoretical insights with real-world educational needs, this study contributes to the broader discourse on AI in education. It proposes scalable, context-sensitive strategies for technological integration in emerging economies.

A journal article introduction should clearly and concisely explain the research topic, its significance, and why the study was undertaken. It should also provide a brief overview of the research methods and findings. A well-written introduction guides the reader, establishing the context for the rest of the paper.

II. THEORETICAL FRAMEWORK

2.1. Chatbots in education

Chatbots, defined as AI-powered virtual assistants (Pérez et al., 2020), have evolved significantly from early rule-based systems such as ELIZA (Weizenbaum, 1966) to sophisticated, context-aware tools that support learning in modern educational environments (Okonkwo & Ade-Ibijola, 2021). Within education, chatbots play a multifaceted role in facilitating self-regulated and autonomous learning. Key educational functions include:

- Providing instant responses to academic queries (Cunningham-Nelson et al., 2019).

- Offering personalized guidance tailored to learner profiles (Roca et al., 2024).

- Stimulating critical thinking through Socratic-style questioning (Vanichvasin, 2022).

The use of chatbots in education aligns with Vietnam’s national strategy for digital transformation in higher education (Nguyen et al., 2023). However, due to cultural and technological differences, evaluating their effectiveness in the local context is essential.

2.2. User satisfaction and integrated theoretical framework

User satisfaction, commonly defined as “the fulfillment of needs or desires,” is a key construct in understanding technology adoption in education (Oliver, 1980). In this study, satisfaction is evaluated through an integrated lens combining two prominent models: the Technology Acceptance Model (TAM) and the Information Systems Success Model (ISSM).

The TAM framework, developed by Davis (1989), identifies “perceived usefulness (PU)” and “perceived ease of use (PEOU)” as the two primary predictors of user satisfaction. PU reflects the extent to which users believe a system enhances their task performance, while PEOU measures the effort required to use the system effectively. While TAM has been widely adopted in educational research, it primarily emphasizes user perceptions. It does not explicitly address system-level factors such as technical reliability and information quality, which are also important in learning environments.

To address these limitations, the ISSM framework (DeLone & McLean, 2003) provides a more system-oriented perspective. It includes dimensions such as “system quality” (e.g., response speed, interface usability), “information quality” (e.g., accuracy, relevance), and “service quality” (e.g., supportiveness, responsiveness). While ISSM excels in measuring objective performance, it tends to underrepresent user cognition and affect, which are central to understanding student attitudes in learning environments.

By integrating TAM and ISSM, this study proposes a hybrid framework that combines user-centered and system-centered perspectives. Specifically, system quality (from ISSM) is linked to PEOU (from TAM), as chatbot responsiveness and interface design directly affect the ease of interaction. Information quality, particularly the accuracy of chatbot-generated content, is emphasized as a standalone construct. Service quality is also mapped to PU, as the pedagogical relevance of chatbot responses contributes to users’ perceived value of the tool.

Figure 1. Integrated TAM-ISSM Framework

Figure 1 illustrates the synthesized model. The blue dimension represents system quality (linking TAM’s PEOU with ISSM’s technical factors), the green dimension captures information quality (core to ISSM), and the red dimension reflects service quality (closely aligned with TAM’s PU). This hybrid framework is particularly well-suited for evaluating chatbot tools in Vietnamese higher education, where both technical performance and perceived educational value are essential. It also accounts for user perspectives shaped by collectivist learning cultures (Bui & Nguyen, 2024).

2.3. Operationalization of constructs

Table 1. Operationalization of TAM-ISSM Constructs

To operationalize the model, survey instruments were designed to measure each theoretical construct using a 5-point Likert scale, ranging from 1 (Strongly Disagree) to 5 (Strongly Agree). These items assess students’ perceptions of chatbot accuracy, response speed, interface usability, and educational support, ensuring alignment with both TAM and ISSM dimensions.

III. RESEARCH METHODOLOGY

3.1. Research design

This study employed a convergent parallel mixed-methods design (Creswell & Plano Clark, 2018), integrating quantitative survey data with qualitative literature analysis to triangulate findings on determinants of chatbot satisfaction. The quantitative phase operationalized constructs from the Technology Acceptance Model (TAM) (Davis, 1989) and the Information Systems Success Model (ISSM) (DeLone & McLean, 2003). The qualitative phase contextualized these results through thematic synthesis of relevant prior studies.

3.2. Participants and sampling

- Population. Undergraduate students at Khanh Hoa University who had used learning support chatbots (e.g., Zalo AI, Messenger bots) for at least three months.

- Sampling. Stratified random sampling was employed by academic year (freshman to senior) and discipline (humanities and social sciences) to ensure representativeness.

- Sample Size. A total of 132 valid responses were collected (87.4% response rate; 151 invited). This sample exceeds the minimum requirement of 98 for detecting a medium effect size (f² = 0.15, α = 0.05, power = 0.95), based on G*Power 3.1 calculations.

The methods section of a research paper provides the information by which a study’s validity is judged. The method section answers two main questions: 1) How was the data collected or generated? 2) How was it analyzed? The writing should be direct and precise and written in the past tense.

3.3. Instruments and measures

A structured questionnaire was developed and pretested with 20 students. All scales demonstrated strong internal consistency (Cronbach’s α > 0.8). The final version was administered via Google Forms from March 11 to March 18, 2025. The questionnaire consisted of five sections:

- Section 1: Demographics. Gender, academic year, and major (categorical variables).

- Section 2: Chatbot Usage Patterns. Frequency (daily/weekly/monthly) and purposes of use (e.g., assignment help, concept clarification).

- Section 3: Satisfaction Determinants. Five-point Likert scale items measuring constructs from the TAM-ISSM framework (see Table 1).

- Section 4: Learning Impact. One open-ended question (e.g., “Describe how chatbots improved your grades”).

- Section 5: Open-Ended Suggestions. Participants were invited to provide additional feedback for improving chatbot services.

3.4. Data analysis

Quantitative data were analyzed using SPSS 26.0. The following procedures were conducted:

- Descriptive Statistics. Means (M), standard deviations (SD), and frequency distributions were computed.

- Pearson Correlation. Bivariate correlations were tested among key variables: PU, RESP, ACCUR, EASE, and SAT.

- Multiple Linear Regression. A model was constructed to assess predictors of satisfaction:

SAT = β₀ + β₁(PU) + β₂(RESP) + β₃(ACCUR) + β₄(EASE) + ε.

- Binary Logistic Regression. Odds ratios (OR) were calculated to assess whether chatbot usage predicted perceived learning improvement (binary outcome: yes/no).

IV. FINDINGS AND DISCUSSION

4.1. Demographic characteristics

This study analyzed responses from 132 undergraduate students at Khanh Hoa University using a mixed-methods approach. Key demographic characteristics are summarized in Table 2.

- Gender. The sample consisted of 84.1% females (n = 111), 15.2% males (n = 20), and 0.8% other gender (n = 1).

- Academic Year. The majority of respondents were first-year students (75.0%, n = 99), followed by third-year (17.4%, n = 23), second-year (5.3%, n = 7), and fourth-year students (0.8%, n = 1).

- Discipline. Students represented the humanities and social sciences (45.5%, n = 60), foreign languages (28.8%, n = 38), pedagogy (19.7%, n = 26), tourism (2.3%, n = 3), and natural sciences (3.0%, n = 4).

- Chatbot Usage Frequency. A total of 37.1% (n = 49) reported daily chatbot use, 33.3% (n = 44) used them weekly, 28.0% (n = 37) used them rarely, and 1.5% (n = 2) used them monthly.

Table 2: Participant Demographics (N = 132)

Key observations

The gender imbalance (84.1% female) reflects enrollment trends at Khanh Hoa University, particularly in humanities-related fields. Future research should consider stratifying by gender to assess potential biases better.

The predominance of first-year students (75.0%) may limit the generalizability of the findings to more advanced cohorts, underscoring the need for longitudinal follow-up studies to capture evolving user experiences over time.

A substantial portion of participants (37.1%) reported daily interaction with chatbots, suggesting a high level of adoption for learning support purposes. It is noteworthy that students from the natural sciences and tourism accounted for less than 5% of the sample combined and were thus excluded from subsequent subgroup analyses.

Overall, the sample was skewed toward first-year female students in the humanities. These demographic characteristics should be taken into account when interpreting the satisfaction results, especially within the context of Vietnam’s gender-disaggregated enrollment patterns.

4.2. Descriptive statistics

A survey of 132 undergraduate students at Khanh Hoa University evaluated five key dimensions of chatbot experiences: Perceived Usefulness (PU), Response Speed (RESP), Accuracy (ACCUR), Ease of Use (EASE), and Overall Satisfaction (SAT). All items were measured using a 5-point Likert scale (1 = Strongly Disagree, 5 = Strongly Agree). Descriptive statistics are presented in Table 3.

Table 3: Descriptive Statistics of Chatbot Experience Dimensions (N = 132)

Key findings and discussion

Perceived usefulness emerged as the most positively rated dimension, with the highest mean score (M = 3.91, SD = 0.97). A total of 72% of participants either agreed or strongly agreed that chatbots were useful in supporting their learning. This finding aligns with Hypothesis 1 and reinforces the idea that students recognize the practical utility of chatbots, even if other aspects of the experience remain suboptimal.

In contrast, accuracy was the dimension with the lowest proportion of “strongly agree” responses (9.8%) and the highest proportion of neutral responses (53.8%). This pattern suggests a degree of uncertainty or hesitation among users regarding the reliability of chatbot-generated content. Qualitative responses echoed this concern, with several students noting that they “often need to verify answers,” indicating a lack of complete trust in the system’s informational accuracy.

Ease of use also received a relatively neutral reception, with a mean score of 3.31 and more than 60% of participants selecting “neutral.” Only 35.6% of respondents rated the usability positively, suggesting that the interface and interaction design of current chatbots may not be fully optimized for student needs. These findings highlight opportunities for enhancing user experience (UX) design, particularly in mobile access and discipline-specific customization.

Taken together, the results reveal a “utility-trust paradox.” While students perceive chatbots as useful tools for learning (high PU scores), they simultaneously harbor concerns about the accuracy of information and ease of use. This paradox is not unique to this context and echoes global challenges associated with integrating AI into educational settings (Roca et al., 2024). The findings also resonate with the cultural tendency in Vietnam to prioritize functional outcomes over design perfection (Bui & Nguyen, 2024), reinforcing a form of cultural pragmatism in educational technology adoption.

The relatively neutral evaluations of ease of use suggest the need for more intuitive, context-aware interfaces. Educational institutions and developers should consider mobile-first UX redesigns and incorporate customizable features tailored to different disciplines, such as simplified interaction formats for the humanities and advanced input-output structures for STEM fields.

4.3. Correlation analysis

To examine the relationships among chatbot experience dimensions and overall student satisfaction, a Pearson correlation analysis was conducted (N = 132, α = 0.05). The results are summarized in Table 4 and discussed below.

Table 4. Pearson correlation matrix of chatbot satisfaction determinants

Notes.

*** < 0.001 (two-tailed). PU = Perceived Usefulness, RESP = Response Speed, ACCUR = Accuracy, EASE = Ease of Use, SAT = Satisfaction.

- Coefficients ≥ 0.70 (bolded) indicate strong correlations (Cohen, 1988).

Key findings and interpretation

Perceived Usefulness (PU) as the strongest predictor of satisfaction: The strongest correlation observed was between Perceived Usefulness and Satisfaction (r = .927, p < .001). This finding supports Hypothesis 1 and aligns with the Technology Acceptance Model (TAM; Davis, 1989), suggesting that students place a high value on the chatbot’s ability to deliver functional benefits. Qualitative feedback reinforced this result, with many students emphasizing the chatbot’s utility in reducing their workload and improving access to learning materials.

Response speed and accuracy: a trade-off in expectations: Response Speed also showed a strong positive correlation with Satisfaction (r = .894), indicating that students appreciated quick replies. However, its slightly weaker correlation with Accuracy (r = .753) suggests that students may tolerate minor errors if the chatbot responds promptly. This is consistent with Nguyen et al. (2024), who found that in Vietnamese higher education, students tend to prioritize efficiency in service tools over perfect accuracy. Some students noted that, although the answers were not always completely correct, the quick feedback still supported their learning process.

Ease of use and interface design limitations: Ease of Use (EASE) had the weakest correlation with Satisfaction (r = .778) among the four predictors. Although still statistically strong, this finding aligns with descriptive results showing that a majority of students rated the interface neutrally. Several students commented that the layout was “cluttered but manageable,” pointing to room for improvement in the chatbot’s user experience (UX). This highlights a potential design gap: while the system functions adequately, the visual and interactive design elements may hinder deeper engagement.

Discussion and implications

These correlation patterns validate the integrated TAM-ISSM model presented earlier (Figure 1). System quality (represented by Response Speed and Ease of Use) and information quality (Accuracy) jointly influence Satisfaction. Preliminary regression analysis indicates that these factors explain approximately 85% of the variance in Satisfaction (R² = .85), which exceeds the explanatory power typically observed in TAM-alone studies (60-70%; Venkatesh et al., 2003).

Culturally, the results reflect Vietnam’s pragmatic approach to educational technology. The strong PU-SAT relationship underscores a utilitarian mindset in which students prioritize tools that deliver tangible academic outcomes (Bui & Nguyen, 2024). Despite aesthetic or interface concerns, functionality remains paramount in the adoption and continued use of learning support technologies.

4.4. Regression analysis

To assess the relative influence of chatbot experience dimensions on overall student satisfaction, we conducted a multiple linear regression analysis. The model incorporated five independent variables derived from the Technology Acceptance Model (TAM) and the Information Systems Success Model (ISSM): perceived usefulness (PU), response speed (RESP), accuracy (ACCUR), ease of use (EASE), and usage frequency.

Model specification:

Satisfaction = β₀ + β₁(PU) + β₂(RESP) + β₃(ACCUR) + β₄(EASE) + β₅(Frequency) + ε

- Dependent Variable: Satisfaction (Q12)

- Independent Variables:

- Perceived Usefulness (PU, Q8)

- Response Speed (RESP, Q9)

- Accuracy (ACCUR, Q10)

- Ease of Use (EASE, Q11)

- Usage Frequency (Q7)

Regression Results

Table 5. Standardized Regression Coefficients Predicting Chatbot Satisfaction (N = 132)

Model fit:

- Model Fit: R² = 0.68; Adjusted R² = 0.66

- F(5,126) = 32.7, p < .001

- Durbin-Watson = 1.92, indicating no significant autocorrelation.

The results validate our integrated TAM-ISSM framework (Figure 2):

Figure 2: Standardized regression coefficient plot and statistical significance

Key findings

Perceived usefulness emerged as the strongest predictor of satisfaction (β = 0.42, p < .001), lending strong support to the foundational TAM framework (Davis, 1989). This suggests that students are primarily motivated by the chatbot’s functional value-namely, whether it helps them achieve their learning goals efficiently.

Accuracy (β = 0.35, p < .001) was the second most influential factor, reflecting the importance of reliable information in building student trust. This aligns with the ISSM dimension of information quality (DeLone & McLean, 2003) and echoes themes identified in student comments such as “accuracy matters more than speed”.

While response speed also had a significant positive effect (β = 0.28, p = .003), its relatively smaller impact compared to accuracy suggests that students may tolerate moderate delays if the chatbot delivers correct and relevant information. This trade-off between immediacy and reliability is consistent with findings in Vietnamese educational contexts (Nguyen et al., 2024).

Ease of use was marginally significant (β = 0.19, p = .023), indicating that while usability concerns exist, they are secondary to cognitive outcomes like perceived utility. Open-text responses noted interface limitations (e.g., “cluttered layout”) but also expressed adaptability to such constraints.

Notably, usage frequency did not significantly predict satisfaction (β = 0.11, p = .152), suggesting that the quality of chatbot interactions outweighs their quantity.

Theoretical and practical implications

These findings offer empirical validation for our integrated TAM-ISSM framework (see Figure 1). The model explained 68% of the variance in student satisfaction, outperforming traditional TAM-based models, which typically account for 60-70% (Venkatesh et al., 2003). The dominance of perceived usefulness and accuracy supports the cognitive evaluation perspective (Oliver, 1980). It highlights the pragmatic orientation of Vietnamese learners, who emphasize tangible academic outcomes over interface aesthetics (Bui & Nguyen, 2024).

From a practical standpoint:

- For Developers: Focus should be placed on enhancing accuracy through fact-verification tools (e.g., integration with Wolfram Alpha APIs) and ensuring functional clarity rather than elaborate UI design.

- For Educators: There is a need to foster students’ digital literacy and critical evaluation skills, helping them discern trustworthy chatbot responses (Nguyen et al., 2024).

Limitations and future directions

The cross-sectional nature of this study precludes causal inference. Longitudinal or experimental designs would be beneficial for examining how user satisfaction evolves with continued chatbot exposure. Additionally, qualitative follow-ups could explore how cultural and contextual factors mediate perceptions of usefulness and trust.

4.5. Learning impact analysis

To evaluate whether students’ chatbot experiences significantly predict perceived improvement in learning efficiency, we performed a binary logistic regression analysis. The dependent variable was a dichotomous outcome based on student responses to Q14: “Did the chatbot improve your learning efficiency?”

Model specification

The logistic regression model included six independent variables drawn from the TAM and ISSM frameworks:

- Perceived Usefulness (PU)

- Response Speed (RESP)

- Accuracy (ACCUR)

- Ease of Use (EASE)

- Satisfaction (SAT)

- Usage Frequency

Model:

Logit (Improvement) = β₀ + β₁(PU) + β₂(RESP) + β₃(ACCUR) + β₄(EASE) + β₅(SAT) + β₆(Frequency)

Regression results

Table 6. Odds Ratios for Predicting Learning Improvement (N = 132)

The chart below illustrates the odds ratios (OR) of factors influencing learning improvement when using chatbots

Figure 3: Forest plot of “Odds Ratios (95% CI)”

Note:

- Bold columns (OR > 1 and marked with ***, **, *) indicate statistically significant effects.

- The vertical line at OR = 1 is the reference threshold: if the column passes this level and is significant, the factor has a significant positive effect.

- Factors like “Accuracy” and “Usefulness” have the highest OR and a substantial impact.

- “Interface” and “Frequency of use” did not exceed the significance threshold (noted as ns — not significant).

Key findings

Primary predictors of learning improvement

- Accuracy (OR = 3.42, p < .001): Students who perceived the chatbot as accurate were over three times more likely to report learning improvement.

- Perceived Usefulness (OR = 2.85, p < .001): Reinforces the centrality of functional value in learning outcomes, as predicted by TAM (Davis, 1989).

Secondary but significant factors

- Satisfaction (OR = 2.16, p = .004): Supports the expectation-confirmation model (Oliver, 1980), where positive experience drives perceived improvement.

- Response Speed (OR = 1.72, p = .018): Indicates that timeliness contributes to learning outcomes, albeit less strongly than accuracy.

Non-significant predictors

- Ease of use and usage frequency (both OR = 1.24, p > .10): Neither interface simplicity nor usage intensity significantly influenced perceived learning improvement.

Discussion

The results affirm that content quality is the primary determinant of perceived learning gains. The strong effects of accuracy and usefulness align with prior findings in AI-supported education (Roca et al., 2024), emphasizing that students prioritize informational trustworthiness and relevance over visual or interactional features.

Interestingly, while satisfaction positively influenced learning outcomes, its role was secondary—possibly reflecting the pragmatic orientation of Vietnamese learners (Bui & Nguyen, 2024), who value tangible educational benefits over hedonic experience. This may also explain why ease of use and interface design were not significant, despite their importance in Western contexts.

Implications for practice

For Developers: Enhance algorithmic reliability and factual correctness over interface aesthetics. Consider integrating domain-specific knowledge bases to improve accuracy.

For Educators: Promote AI literacy programs to help students critically evaluate chatbot outputs and understand their limitations.

Limitations and future research

This study relies on self-reported perceptions of learning gains, which may be subject to recall or confirmation bias. Future research should integrate objective performance indicators (e.g., exam scores, assignment quality) and explore interaction effects between satisfaction and other predictors. Additionally, cross-cultural comparisons could shed light on whether these findings generalize beyond the Vietnamese context.

4.6. Key findings and implications

#### 4.6.1. Determinants of satisfaction and learning outcomes

The integrated analysis of the Technology Acceptance Model (TAM) and the Information Systems Success Model (ISSM) identified four key predictors of user satisfaction with chatbot-based learning support. The strongest determinant was perceived usefulness (β = 0.42; OR = 2.85), reaffirming its central role in TAM and the adoption of educational AI technologies (Davis, 1989). Accuracy (β = 0.35; OR = 3.42) also emerged as a significant predictor, highlighting its importance in building trust and aligning with the ISSM’s emphasis on information quality (DeLone & McLean, 2003).

While response speed (β = 0.28) contributed moderately to satisfaction, it was secondary to accuracy, indicating that learners valued correctness over immediacy. Interface quality (β = 0.19) was marginally significant, suggesting that design aesthetics or usability had a lesser influence on user satisfaction in this context.

From these findings, two sets of practical recommendations are proposed:

- For higher education institutions: Institutions should implement verification protocols for chatbot outputs, such as integrating institutional databases or curated reference sources. Additionally, training modules should be developed to help students critically assess chatbot-generated information.

- For developers of educational chatbots, future iterations should incorporate accuracy-enhancing mechanisms, such as fact-checking APIs (e.g., Wolfram Alpha), and offer adaptive learning pathways that reflect learners’ individual competencies and progress.

#### 4.6.2. Non-significant factors

Neither interface design (OR = 1.24, p > .05) nor usage frequency (OR = 1.24, p > .05) significantly predicted learning improvement. This finding reinforces the conclusion that service quality, particularly the accuracy and relevance of content, outweighs the frequency of use or interface aesthetics in determining the perceived educational value of chatbot systems.

#### 4.6.3. Limitations and directions for future research

The study’s findings must be interpreted within the context of several limitations. First, the sample was drawn from a single Vietnamese university (N = 132), with limited representation from STEM disciplines (12.1%). Second, all measures were based on self-reported data, raising potential concerns about social desirability bias and subjective recall.

Moreover, the current analysis did not explore affective dimensions such as users’ emotional responses (e.g., anxiety or confidence) when interacting with chatbots. It also did not investigate discipline-specific variations in user needs or expectations.

Future research should expand the scope through:

- Large-scale, multi-institutional studies across diverse academic fields;

- Mixed-method approaches incorporating behavioral log data;

- Affective computing techniques to capture emotional engagement in real time.

#### 4.6.4. Conclusion

This study provides empirical evidence for the role of chatbots in enhancing learning experiences at Khanh Hoa University. The effectiveness of such tools, however, depends primarily on their accuracy and usefulness, rather than interface design or frequency of use. These findings provide a practical framework for enhancing AI-driven educational technologies, particularly in resource-constrained settings. Broader validation remains essential to confirm the generalizability of these insights.

V. CONCLUSION AND RECOMMENDATIONS

In the era of digital education, AI-powered learning support tools, such as chatbots, have become integral to the learning processes of university students. Early adoption of these tools, when combined with critical thinking skills in modern educational environments, enhances students’ adaptability, digital literacy, and ability to personalize their learning journeys — ultimately fostering greater engagement and creativity. This study examined student satisfaction with AI-powered learning chatbots at Khanh Hoa University through an integrated TAM-ISSM framework. The results demonstrate that perceived usefulness (β = 0.42, p < 0.001) and accuracy (OR = 3.42, p < 0.001) are the strongest determinants of both satisfaction and learning improvement, confirming the centrality of functional utility and information quality in the adoption of educational technology. While response speed (β = 0.28) and interface design (β = 0.19) had lesser impacts, the nonsignificance of usage frequency (OR = 1.24, p = 0.152) suggests that students prioritize the quality of interaction over the quantity of use. These findings carry important implications: for developers, they underscore the need to integrate verified knowledge sources and personalize learning features, rather than focusing on aesthetic enhancements; for educators, they highlight the necessity of training students in critical AI literacy skills to effectively evaluate chatbot outputs. Although limited by its single-institution sample and self-reported data, this study provides a foundational model for optimizing chatbot-assisted learning in Vietnamese higher education. Future research should expand to multidisciplinary contexts and incorporate emotional engagement metrics to capture the user experience fully. Ultimately, these insights contribute to Vietnam’s digital transformation goals by bridging technological innovation with pedagogical best practices for self-directed lifelong learning.

REFERENCES

[1] Bui, T. T., & Nguyen, M. T. (2024). Nghiên cứu ảnh hưởng của trí tuệ nhân tạo trong giáo dục tới hoạt động học tập của sinh viên. Tạp chí Giáo dục, 6-11. Retrieved from https://tcgd.tapchigiaoduc.edu.vn/index.php/tapchi/article/view/1863

[2] Cohen, J. (1988). Set correlation and contingency tables. Applied Psychological Measurement, 12(4), 425-434. https://doi.org/10.1177/014662168801200410

[3] Creswell, J. W., & Plano Clark, V. L. (2018). Designing and conducting mixed methods research (3rd ed.). SAGE Publications.

[4] Cunningham-Nelson, S., Boles, W., Trouton, L., & Margerison, E. (2019). A review of chatbots in education: Practical steps forward. In 30th Annual Conference for the Australasian Association for Engineering Education (AAEE 2019): Educators becoming agents of change: Innovate, integrate, motivate (pp. 299-306). Engineers Australia.

[5] Davis, F. D. (1989). Perceived usefulness, perceived ease of use, and user acceptance of information technology. MIS Quarterly, 13(3), 319-340. https://doi.org/10.2307/249008

[6] DeLone, W. H., & McLean, E. R. (2003). The DeLone and McLean model of information systems success: A ten-year update. Journal of Management Information Systems, 19(4), 9-30. https://doi.org/10.1080/07421222.2003.11045748

[7] Ngo, T. T. A., Tran, T. T., An, G. K., & Nguyen, P. T. (2024). ChatGPT for educational purposes: Investigating the impact of knowledge management factors on student satisfaction and continuous usage. IEEE Transactions on Learning Technologies, 17, 1341-1352. https://doi.org/10.1109/TLT.2024.3383773

[8] Nguyễn, N.D.T, Nguyễn, V.L.M., Tam, L. C., Huy, T. Q., Nha, T.T., Hiển, L. T. H., ... & Hung, N. V. (2023). ALGOBOT — Một hệ thống chatbot hỗ trợ giải đáp các kiến thức cơ bản về cấu trúc dữ liệu và giải thuật. Tạp chí Khoa học Trường Đại học Sư phạm TP Hồ Chí Minh, 20(2), 205-205. https://doi.org/10.54607/hcmue.js.20.2.3613(2023)

[9] Okonkwo, C. W., & Ade-Ibijola, A. (2021). Chatbots applications in education: A systematic review. Computers and Education: Artificial Intelligence, 2, 100033. https://doi.org/10.1016/j.caeai.2021.100033

[10] Oliver, R. L. (1980). A cognitive model of the antecedents and consequences of satisfaction decisions. Journal of Marketing Research, 17(4), 460-469. https://doi.org/10.1177/002224378001700405

[11] Pérez, J. Q., Daradoumis, T., & Puig, J. M. M. (2020). Rediscovering the use of chatbots in education: A systematic literature review. Computer Applications in Engineering Education, 28(6), 1549-1565. https://doi.org/10.1002/cae.22326

[12] Roca, M. D. L., Chan, M. M., Garcia-Cabot, A., Garcia-Lopez, E., & Amado-Salvatierra, H. (2024). The impact of a chatbot working as an assistant in a course for supporting student learning and engagement. Computer Applications in Engineering Education, 32(5), e22750. https://doi.org/10.1002/cae.22750

[13] Saiz-Manzanares, M. C., Marticorena-Sanchez, R., Martin-Anton, L. J., Diez, I. G., & Almeida, L. (2023). Perceived satisfaction of university students with the use of chatbots as a tool for self-regulated learning. Heliyon, 9(1).

[14] Tuan, N. M., & Phuong, T. V. (2018). Chatbot trong giáo dục. Lời Nói Đầu, 79.

[15] Vanichvasin, P. (2022). Impact of chatbots on student learning and satisfaction in the entrepreneurship education programme in higher education context. International Education Studies, 15(6), 15-26.

[16] Venkatesh, V., Morris, M. G., Davis, G. B., & Davis, F. D. (2003). User acceptance of information technology: Toward a unified view. MIS Quarterly, 27(3), 425-478.

[17] Weizenbaum, J. (1966). ELIZA—A computer program for the study of natural language communication between man and machine. Communications of the ACM, 9(1), 36-45.