Exploring University Students ’ Adoption of ChatGPT Using the Diffusion of Innovation Theory and Sentiment Analysis With Gender Dimension

This study explores the adoption and societal implications of an emerging technology such as Chat Generative Pre-Trained Transformer (ChatGPT) in higher education students. By utilizing a mixed-method framework, this research combines Rogers ’ di ﬀ usion of innovation theory with sentiment analysis, o ﬀ ering an innovative methodological approach for examining technology adoption in higher educational settings. It explores ﬁ ve attributes — relative advantage, compatibility, ease of use, observability, and trialability — shaping students ’ behavioral intentions toward ChatGPT. Sentiment analysis o ﬀ ers qualitative depth, revealing emotional and perceptual aspects, and introduces a gender-based perspective. The results suggest that ﬁ ve innovation attributes signi ﬁ cantly impact the adoption rates and perceptions of ChatGPT, indicating its potential for transformative social change within the educational sector. Gen Zs viewed ChatGPT as innovative, compatible, and user-friendly, enabling the independent pursuit of educational goals. Consequently, the bene ﬁ ts provided by ChatGPT in education motivate students to use the tool. Gender di ﬀ erences were observed in the prioritization of innovation attributes, with male students favoring compatibility, ease of use, and observability, while female students emphasized ease of use, compatibility, relative advantage, and trialability. The ﬁ ndings have implications for understanding how technological innovations such as ChatGPT could be strategically di ﬀ used across di ﬀ erent societal segments, especially in the academic context where ethical considerations such as academic integrity are paramount. This study underscores the need for a demographic-sensitive, user-centric design in generative arti ﬁ cial intelligence (AI) technologies.


Introduction
In November 2022, OpenAI, a leading technology company, unveiled ChatGPT (Chat Generative Pre-Trained Trans-former), one of the most advanced iterations of its artificial intelligence (AI)-driven large language model for text generation [1][2][3].Since its public debut, ChatGPT has captured global attention, with coverage in prominent publications • RQ3: Are there any perceived differences concerning gender regarding ChatGPT enablers?
The importance of ChatGPT in education is paramount because it has wide applicability for enhancing education through content creation, grammar and writing checks, grading, and online syllabus design for teachers.On the other hand, it can significantly help students complete homework and assignments, research projects, and language learning [19].This research explores the determinants guiding university students' proclivity to integrate ChatGPT into higher educational environments.Using an innovative mixed-method design, the investigation combines Rogers' theory of perceived attributes (2003) with sentiment analysis techniques.Rogers' theory provides a structured framework for examining how various attributes, such as relative advantage, compatibility, simplicity, trialability, and observability, might influence the rate of technology adoption.Sentiment analysis will be incorporated to probe qualitative aspects, capturing emotional nuances and user opinions about ChatGPT.Additionally, the study is aimed at exploring gender biases in sentiment classification within the dataset.By leveraging both quantitative and qualitative methodologies, this study seeks to provide a comprehensive understanding of the attributes and perceptions that affect the adoption of ChatGPT in educational contexts.The unique integration of these methods aspires to capture not only measurable variables but also the emotional and demographic intricacies that could play a crucial role in the broader adoption of this technology.Investigating the predictors of ChatGPT adoption among university students through the lens of Rogers' theory of perceived attributes has multifaceted value.Theoretically, this research could validate or extend the applicability of Rogers' attributes to AI-driven educational technology.This becomes pivotal for adjusting existing technology adoption models to the contours of generative AI tools such as ChatGPT.From an application standpoint, identifying these key predictors can directly inform the design elements and deployment strategies, ensuring that ChatGPT aligns with the factors most likely to influence its adoption rate in an educational setting.

Literature Review
2.1.DOI Theory: Relevance for Educational Innovations.In the context of rapid technological advancements, the role of innovation in affecting organizational and societal change has gained significant scholarly attention [21].The transition to digital transformation, marked by developments in the Internet of Things (IOT), big data analytics, and AI, has spurred a variety of strategies for technology adoption and diffusion [22].Among the theoretical frameworks employed to understand this phenomenon, the Rogers' DOI theory stands out for its comprehensive approach to studying how innovations disseminate across different social systems [23].
The DOI posits that five key perceived attributes of innovation influence its adoption rate [20]: • Relative advantage: the degree to which the innovation is perceived as better than the existing solution.
• Compatibility: the degree to which the innovation is perceived as compatible with existing values, practices, and needs.
• Complexity: the degree to which the innovation is perceived as difficult to understand or use.
• Trialability: the degree to which the innovation can be tried on a limited basis before committing to its adoption.
• Observability: the degree to which the results of using the innovation are visible to others.
The adopter categories are central to Rogers' theory, classifying individuals based on their willingness and speed to adopt innovations.These categories include innovators, early adopters, early majority, late majority, and laggards.Innovators are the first to embrace new ideas and are often risk-takers eager to experiment with cutting-edge technologies.Early adopters, the second-fastest group to adopt innovations, are influential opinion leaders who help spread new ideas to a broader audience.The early majority, making up approximately 34% of the population, adopt innovations at a slower pace, considering the experiences of early adopters before committing to the new idea.Approximately 34% of the late majority are generally skeptical of change and only adopt innovations after they have become widely accepted.Finally, laggards are the most resistant to change and adopt innovations only when necessary or when traditional alternatives are no longer available.Understanding these adopter categories helps organizations and innovators strategize the successful diffusion of their products, services, or ideas to different segments of society.Early users of the ChatGPT, who began using the AI model soon after its release and were eager to explore its potential, likely belonged to the innovator or early adopter category.These individuals are often more open to experimenting with new technologies and appreciate the potential value of AI-driven language models.
Numerous factors influence the acceptance, adoption, and intention to use technology [6,8].Innovation is one such factor that drives technology adoption [24,25].Encouraging students to develop self-efficacy and learning autonomy can facilitate e-learning adoption and the use of virtual platforms such as Moodle as a Learning Management System [26].These educational tools can enhance students' self-efficacy and skills, enabling them to manage these technologies better [1,27].As a result, using virtual platforms can improve individuals' perceptions of self-efficacy and abilities.
Technology adoption in education, particularly students' intentions to use e-learning systems [26,28], virtual laboratories [29], and online proctoring systems [30], has been well studied based on Rogers' theory.According to the theory, innovations with favorable features are more likely to be adopted [20].In other words, innovations that offer advantages, perceived compatibility with existing practices and beliefs, low complexity, potential trialability, and observability will have a broader and more rapid diffusion rate [31].Alrahmi et al. [32] conducted a study and found that relative advantage, compatibility, complexity, trialability, and observability influenced students' behavioral intentions to use e-learning systems in universities.Similarly, Pinho, Franco, and Mendes [26] examined Portuguese university students and discovered that Moodle LMS's favorable features positively impacted its usage, as Rogers [20] proposed.In addition to LMSs, blogging has become a popular tool for students [33], with teachers and students increasingly using blogs worldwide [34].Perceived compatibility positively affects students' attitudes toward blog usage.Students who find blogs compatible with their thought processes and learning styles are more likely to use them for educational purposes [35].This finding supports the notion that when students perceive high compatibility with learning technologies, their attitudes toward using them are positive [14,35].
Some examples illustrate the rapid growth and diffusion of educational technology innovations driven by accessibility, ease of use, and the ability to address the evolving needs of learners and educators, for example,  [36].
A critical issue to address is the challenge of accepting and acknowledging ChatGPT's usage.Patel and Lam [37] expressed concerns for medical practitioners, noting that although ChatGPT can quickly generate discharge summaries using AI-enabled automation, patients may feel detached from 3 Human Behavior and Emerging Technologies their doctors' care, potentially leading to resistance to adopting the technology [38].Similarly, concerns about the use of ChatGPT in education have increased; for example, Choi et al. [39] examined the ability of AI to grade exams without human assistance.When these exams were graded blindly, such as regular class assignments, ChatGPT's performance was average and comparable to that of a C+ student.This mediocre performance may result in low intentions to use the technology, and according to Ram and Sheth [40], it can be considered resistance to technology acceptance and adoption.
DOI theory provides a multidimensional lens through which to examine technology adoption, particularly in educational settings.Unlike other models, such as the theory of reasoned action (TRA) [41] or the theory of planned behavior (TPB) [42], DOI focuses on the spread or adoption of innovation rather than individual decision-making.It analyzes how innovations are introduced, disseminated, and adopted across phases to understand the diffusion process.The TRA and TPB focus on individual attitudes, beliefs, and intentions, neglecting societal influences that can affect adoption.Furthermore, DOI theory recognizes that people and groups acquire new ideas and technology at different rates.Innovativeness divides adopters into innovators, early adopters, early majority, late majority, and laggards.This classification recognizes adoption behavior variety and provides personalized efforts to target specific populations.The TRA and TPB presume a homogeneous decisionmaking process and ignore adoption pattern heterogeneity.Similarly, the TAM and the UTAUT were not considered for this study because they primarily focus on individuallevel variables, such as perceived ease of use and perceived usefulness, without adequately accounting for social and systemic influences [43,44].

2.2.
ChatGPT: Initial Development.In 2022, ChatGPT was launched as a generative AI tool that can provide humanlike responses to a different set of inputs and largely functions as a conversational AI [45].The tool encompasses a suitable grouping of AI, Internet resources, natural language generation, and processing coupled with machine learning to develop answers to queries [46,47].The development of conversational AI platforms differs greatly from that of conventional chatbots.ChatGPT has an architecture based on deep learning that facilitates the generation of contextspecific natural responses.Therefore, ChatGPT has a foundation based on the GPT-3 group of large language models and can develop natural language text and responses similar to those of humans [48].With inbuilt chatbot capabilities led by deep learning and GPT-based language models, ChatGPT can naturally provide answers to user-generated queries [49].With 175 billion parameters, the GPT-3 model enables ChatGPT to focus on several diverse text sources, such as webpages, books, research articles, and social conversations [10].
ChatGPT has gradually become one of the most debated AI tools for different sectors based on its wide usage, including education [6,13], healthcare [16,50], banking, tourism, and marketing [51,52].Concurrently, controversies regard-ing its deployment and adoption intentions have been triggered.While Baidoo-Anu and Owusu Ansah [53] underscored specific capabilities such as personalized tutoring and feedback, along with automated essay grading [54] for students, [55] stressed the unique tailored learning experiences for students that can help to improve reading comprehension and performance.In this regime, Lim et al. [25] further identified ChatGPT's potential to provide a more holistic transformation in the higher education industry.While ChatGPT can be a prominent resource for helping students and professionals in essay writing, business correspondence, research and development, and software testing [52], it is becoming a slow threat to content creators and teachers.With increasing ethical concerns regarding the use of ChatGPT's in academics and research, a significant divide exists among teachers and researchers [1].While the New York City Department of Education has blocked it, fearing that it could disrupt education as we know it [5], Mogavi et al. [49] suggested that ChatGPT can erode students' social and critical thinking while promoting superficial learning; Elbanna and Armstrong [56] promoted a balanced view of using ChatGPT within a restricted framework to realize the benefits for students in automating routine tasks while still offering personalized learning.
Similarly, ChatGPT can act as an online educator for learners, reforming education through smartphones and IOT devices and promoting group work; however, some questions can significantly result in inaccurate data and the hiding of falsified data where originality is critical [1].Rejeb et al. [57] further explored the sentiment of students and educators and reconsidered the focus on the restricted and policy-guided use of ChatGPT, both for educators and students/learners.Furthermore, adverse effects on students, such as an eroding learning culture, degrading social and critical thinking, reduced importance of original content, and prominent dependence on technology, have also been noted [1,57].Furthermore, studies have revealed certain agreed-upon challenges, such as integrity violations, plagiarism, the spread of misinformation, inaccurate data, data privacy, and security issues [49,58].
Researchers have found value in using ChatGPT for academic and scientific research writing, hypothesis formation, and resource searching [59].ChatGPT has sometimes even been credited as a coauthor [60].Furthermore, the ChatGPT can substantially help students and professionals translate reading materials to customized requirements and may satisfy the needs of different sectors [53].Despite these advantages, concerns remain regarding research ethics, plagiarism, AI-authored content [61,62], and the potential for errors, privacy issues, and data biases [53].As a result, there are mixed opinions on the use of ChatGPT in academia and other sectors.Another study utilized ChatGPT to evaluate the quality of predatory and hijacked journals, ultimately finding it unreliable for such assessments [63].
As ChatGPT becomes more widespread in education, its ethical and practical implications remain debated [49,64], and research on its adoption is still emerging.For example, Steiss et al. [48]  of ChatGPT than that of humans in providing students with feedback on essay writing.Students' adoption intention of ChatGPT based on the TAM perspective was found to be strongly influenced by perceived ease of use, perceived usefulness, positive attitudes toward technology, social influence, behavioral/cognitive elements, low anxiety, and minimal perceived risk [13]; these allied negative effects are not ignored [15].Using TAM perspectives, multiple studies have explored the adoption intentions of ChatGPT.While Liu and Ma [6] found that the perceived ease of use of ChatGPT does not shape attitudes directly but rather through usefulness for teachers, Iqbal, Ahmed, and Azhar [7] found a significant negative attitude toward ChatGPT adoption intentions owing to ethical concerns.Yilmaz et al. [8] found prominent gender differences regarding "perceived ease of use" regarding ChatGPT among male and female students, with further significant differences regarding "social influence" based on academic discipline.Similarly, Saif et al. [15] reported that perceived ease of use and usefulness are significantly shaped by stress and anxiety among students.Students often become stressed as part of the exam or assignment deadlines approach, and they adopt tools such as ChatGPT, as it is easy to use and useful as a mechanism to reduce their anxiety [15].To this end, Sallam et al. [16] further added that risk perceptions, usefulness, ease of use, attitudes toward technology, and behavioral factors must be carefully considered to determine usage intentions for ChatGPT as a tool for healthcare education.By exploring the research landscape regarding the use of ChatGPT, Raman et al. [65,66] obtained a prominent score using the Altmetric Attention Score (AAS).Their study further highlighted the United States, Japan, and the United Kingdom as prominent countries focusing on ChatGPT research, including major journals such as Nature and Science, and concluded that the fields of "information and computing sciences" and "biomedical and clinical sciences" were the most engaged [65,66].
It may be noted that the current manuscript builds on the preprint version by the same set of authors [65,66].The remarkable capabilities of ChatGPT, a cutting-edge technological innovation, have led to the rapid use of AIenabled text generation in education.ChatGPT and similar large learning models can perform various tasks, including answering detailed questions and composing essays, stories, plays, and poems on any topic; however, these models display gender biases or neglect negation [64].

Research Model-ChatGPT as an Educational Technology Innovation
The DOI and the TAM are widely recognized adoption theories.Many studies, such as those on mobile learning [27] and mobile payment [67], have integrated DOI and TAM constructs to examine behavioral intentions to use technology [68].However, few studies have explored the antecedents of ChatGPT acceptance and adoption.Thus, it is crucial to investigate DOI attributes, which include relative advantage, compatibility, complexity/ease of use, trialability, and observability, as antecedents of the intention to use the ChatGPT (Figure 1).
3.1.Relative Advantage.Rogers [20] defines relative advantage as the extent to which an innovation is perceived as superior to the idea it replaces.An innovation may be advantageous when it provides economic benefits, social status, and gratification [69].For instance, Ali et al. [70] studied how ChatGPT could create easily understandable clinical letters using generative AI, enhancing efficiency, consistency, accuracy, patient satisfaction, and cost-effectiveness in healthcare systems.If students perceive ChatGPT as a cost-free resource that can economically benefit them, enhance their academic writing skills for daily assignments, and help them improve their grades and reputation among peers, they may be more likely to adopt it [71].A high perception of relative advantage can make ChatGPT a preferred option compared to alternatives, and students are more likely to have a stronger intention to use it.Therefore, we hypothesize the following: H1: The relative advantage of ChatGPT positively affects students' intention to adopt it.[20] defines compatibility as the extent to which an innovation aligns with potential adopters' existing values, past experiences, and needs.Consequently, compatibility reflects the degree of congruence between innovation and potential adopters' values, needs, and lifestyles [72].This study suggested that students may be inclined to adopt ChatGPT, as it could enhance their virtual learning experiences.The COVID-19 pandemic has significantly increased the adoption of virtual and e-learning processes, making them an essential aspect of students' lives.The study posits that students' proficiency in an e-learning environment is a critical factor influencing their intention to adopt ChatGPT.Therefore, we hypothesize the following: H2: The compatibility of ChatGPT positively affects students' intention to adopt it.

3.3.
Complexity/Ease of Use.Innovations are often quickly associated with being either easy or difficult to use [20].Davis [43] defines "ease of use" as the degree to which an individual perceives a system as effortless to use.Consequently, innovations demanding new skills and knowledge from adopters are typically adopted more slowly than simpler innovations [65,66].This study proposes that university students will likely perceive ChatGPT as a user-friendly and convenient tool for their learning needs.Therefore, ease of use influences individuals' intention to adopt ChatGPT.We hypothesize the following: H3: The complexity of ChatGPT negatively affects students' intention to adopt it.

Trialability.
Trialability refers to the degree to which an innovation can be experimented with on a limited basis [20].According to Rogers's theory, trialability pertains to how easily an innovation can be tested before adoption.For individuals considering adopting innovation, trialability is essential because it allows them to experience it before fully committing [69].To promote the adoption of ChatGPT among university students, enhancing the technology's trialability by providing training sessions, demonstrations, or presentations could help students improve their academic writing abilities, increase their knowledge, and foster personal growth.Consequently, trialability will influence their intention to adopt the ChatGPT.We hypothesize the following: H4: The trialability of ChatGPT positively affects students' intention to adopt it.

Observability.
Observability is another aspect of innovation, referring to the extent to which the innovation's results are visible to others [20].The visibility of an innovation's outcomes can influence potential adopters' willingness to adopt it.Agag et al. [73] argue that visible innovations can encourage individuals to discuss them with friends and neighbors, leading to positive attitudes toward their use.In this study, we suggest that when students observe their peers expressing interest in and using ChatGPT, they are more likely to form an intention to adopt the technology.Therefore, we hypothesize the following: H5: ChatGPT's observability positively affects students' intention to adopt it.

Research Methods
This study investigated the potential enablers of students' "intention to use ChatGPT."Through an extensive literature review, this study identified five potential antecedents: relative advantage, compatibility, ease of use, trialability, and observability.Several studies have been conducted to ensure the validity of the findings.First, we develop the measures through expert review and pretesting, followed by hypothesis testing using covariance-based structural equation modeling and partial least squares.The identified factors were operationalized as first-order factors, with their measures collected from established scales and validated for measurement through expert review and pretesting with 172 perceptual responses that satisfied the filtering criteria.With satisfactory measures, the study then collected another round of perceptual responses (N = 288) to validate the measurement and structural model.The samples in both the pretest and the main test were deemed suitable because they agreed on two key filter questions: "familiarity with ChatGPT" and "effectiveness of ChatGPT as a learning tool."Purposive sampling was adopted to collect perceptual responses (pretest and main test) from university students from India studying in three different states in the cities of Kochi, Amritapuri, Coimbatore, and Visakhapatnam.Subsequently, we conducted sentiment analysis to complement the findings from the quantitative analysis.
As a mixed-method approach, the study incorporates both quantitative and qualitative data.As such, this study provides deeper insight into the factors that affect the intention of students to adopt ChatGPT.On the one hand, quantitative analyses served as a foundation for most of the  research.For instance, first-stage analysis, that is, confirmatory factor analysis (CFA), was utilized to determine the reliability and validity of survey instruments.The second stage analyzed the possible causal influence of the five posited enablers based on the DOI theory.In this way, it was possible to examine the premise of the study and evaluate whether the relationships between presumed constructs such as ease of use, compatibility, and students' intention to use ChatGPT supported the hypotheses.
A sentiment analysis complements the findings from the quantitative analysis.For the sentiment analysis, we utilized the "bert-base-nli-mean-tokens" sentence transformer model [74] to convert responses from open-ended text responses into 768-dimensional sentence embeddings that represent their semantic content.After preprocessing, the data were clustered using the K-means algorithm [75,76] into three sentiment groups: positive, negative, and neutral.Cosine similarity was used to evaluate the closeness of sentiments within clusters.In addition, the key themes for each gender and sentiment were identified through frequency analysis of tokenized words.The findings were used for triangulation to inform theory and implications.The results of the two types of analyses were combined to provide a comprehensive view of the factors that drive adoption while enabling a deep look at the sentiments students have toward technology.
We acknowledge our study participants for being willing to participate in the study.Informed consent was obtained from all individual participants involved in the study.All the constructs, namely, five independent variables and one dependent variable (the intention to use ChatGPT), were operationalized as first-order factors.The measures for the first-order constructs were developed from the extant literature and were subjected to expert review and feedback.A panel of ten experts was appointed.The experts were given a working definition of each construct and were asked to indicate the suitability of the items for measurement as 1 (not suitable), 2 (somewhat suitable), and 3 (highly suitable).Only those items with a highly suitable rating from the expert panel were retained for further pretesting.Using purposive sampling and certain filtering criteria, 198 undergraduate and postgraduate students at a university in Tamil Nadu, a state located in south India, from February to April 2023 were contacted to pretest and validate the constructs' measurement items, using an online questionnaire prepared using Google Forms.After several follow-ups, 172 responses were received, satisfying the filter questions ("familiarity with ChatGPT" and "effectiveness as a learning tool") and subjected to principal component analysis in SPSS 28 to check whether the items measured the intended constructs.Bartlett's test of sphericity and the KMO test were significant, suggesting correlations among the items.Table 1 shows the rotated factor loadings.Furthermore, the rotated factor loadings clearly showed that one item (of relative advantage) was problematic and was excluded from further consideration.The rotated factor and loadings are shown in Table 1.The items for relative advantage were labeled ra1-ra5, those for compatibility were labeled comp1-5, those for ease of use were labeled eou1-4, those for trialability were labeled trial1-4, those for observability were labeled obs1-3, and those for intention to use were labeled int1-3.One problematic item (ra6-I am usually the first to try out innovations such as ChatGPT) did not load appropriately and was removed to develop a clear pattern.

Study 2:
Main Data Collection.With these satisfactory measures, the study next prepared a questionnaire in Google Forms for further data collection.The form included the purpose of the study, the definition and explanation of necessary items, and the assurance of anonymity.The authors' team then circulated the developed questionnaire to different students who were primarily studying at universities located in Coimbatore, Kochi, Vishakhapatnam, and Bhubaneshwar from June to August 2023.Two filter questions were included in determining whether the respondents were "very familiar/somewhat familiar/not familiar at all" with ChatGPT and whether they felt that they could use ChatGPT effectively for learning.After several follow-ups and reminders, 336 complete responses were received.However, since the study primarily aimed to understand the attitudes of university students toward ChatGPT for education and learning, the study excluded 20 school students who responded to the survey.Furthermore, 32 respondents indicated that they were unfamiliar with ChatGPT, with 15 indicating that they are not confident that ChatGPT can help them as a learning tool.Therefore, the study excluded these respondents from further consideration and had 288 complete responses for the final analysis.Next, the study used CFA in AMOS 28 to analyze the reliability and validity of the questionnaire.Table 2 shows the final sample profile.The final sample profile has demographic variability.There were 140 female and 148 male responses.Furthermore, 171 respondents indicated that they were very familiar with the ChatGPT, and 117 respondents indicated that they were somewhat familiar with the ChatGPT.5.1.2.1.Measurement Model Evaluation.CFA was conducted using AMOS 28 to evaluate the reliability and validity of the measures corresponding to the latent variables (Table 3).Reliability was assessed through Cronbach's alpha and composite reliability, while validity was assessed by examining standardized loadings, average variance extracted (AVE), and discriminant validity [77].All items had loadings greater than 0.70 and AVE > 0 50, indicating adequate convergent validity [78].With the measures of the construct deemed reliable and valid, the study next evaluated the validity of the proposed structural paths.Furthermore, the square root of the AVE of the constructs was greater than that of the intercorrelations, suggesting satisfactory discriminant validity [78] (Table 4).

Structural Model Evaluation.
With adequate reliability and validity of the measurement items for each construct, the study next evaluated the structural model in AMOS 28.As shown in the structural model (Figure 2), all the posited enablers of students' intention to use ChatGPT were supported, as validated by our responses.The path values and their corresponding significances are summarized in Figure 2. Furthermore, our study compared the proposed model after evaluating the measurement and structural models using partial least squares in ADANCO 2.3.2 [79].The measurement items have also shown adequate reliability and validity in partial least squares.Furthermore, the structural model in partial least squares also supports the earlier main model findings, suggesting increased stability and robustness of our results.

Stage 2:
Sentiment Analysis.This study incorporates sentiment analysis to examine students' perceptions of ChatGPT to complement the quantitative findings by elucidating the qualitative nuances behind the statistical measures.For instance, while quantitative data may indicate a high level of adoption based on usage metrics, sentiment analysis can reveal dissatisfaction due to complexity, directly informing the "simplicity" attribute as per Rogers' DOI theory.Additionally, sentiment analysis can reveal latent variables-such as usability concerns or feature preferences-that are not captured by traditional quantitative metrics but are integral for a comprehensive understanding of ChatGPT adoption.
Sentiment analysis, a natural language processing (NLP) technique, identifies the emotional tone of a text as positive, negative, or neutral.It employs machine learning algorithms to classify sentiment based on language use, contextual clues, and other features.The applications of sentiment analysis span various text types, such as product reviews, social media posts, customer feedback, and news articles.Businesses, governments, and researchers use sentiment analysis to understand customer opinions, analyze public opinion, and study trends in language use and attitudes, respectively [80].Several approaches to sentiment analysis exist, including rulebased systems, machine learning models, and deep learning models.Rule-based systems rely on predefined rules for sentiment classification, while machine learning models learn from labeled data.Deep learning models, such as neural networks, can learn complex text data patterns and achieve state-of-the-art performance on sentiment analysis tasks [80,81].Deep learning architectures such as word2vec; recurrent neural networks (RNNs); and transformer networks such  as BERT, RoBERTa, and GPT-3 generate high-quality text representations [81,82].Sentiment analysis data were derived from 255 of 288 responses to an open-text question ("any other feedback") within the structured online survey.Students' feedback was anonymously collected to maintain confidentiality.To address the reliability of the sentiment analysis results, the collected text data were subjected to a series of preprocessing steps, such as text cleaning, tokenization, and the elimination of redundant terms, in preparation for sentiment analysis.A transfer learning approach was employed to identify latent sentiments in the unlabeled textual data.
To perform any NLP tasks on raw text data, the data must be converted to a machine-readable format, typically real-valued vectors that mimic the meaning and context of a sentence.Using sentence transformers, this work converts the raw text data into real-valued vectors (sentence embeddings).A transfer learning approach is considered here, as the size of the dataset is small, and a pretrained sentence transformer ("bert-base-nli-mean-tokens") is used to generate vectors (768 dimensions) for the given raw text.The generated sentence vectors (the unlabeled data points) are clustered into three distinct clusters-positive, negative, and neutral-using the K-means clustering algorithm.Kmeans clustering is used to group sentence embeddings (vectors) based on their semantic similarity, effectively clustering sentences with similar sentiments together.The premise is that the sentence embeddings (vectors) with similar sentiments will be clustered in the semantic space.K-means is an unsupervised learning algorithm used for clustering data points into groups or clusters based on similarity.The goal is to partition the data into K clusters where each data point belongs to the cluster with the nearest mean, serving as the cluster's centroid.The following are the steps involved in computing clusters: • Step 1: K-means starts by randomly initializing K centroids, which are the centers of the clusters.
• Step 2: Each data point is assigned to the nearest centroid, forming K clusters.
• Step 3: The centroids are recalculated as the mean of all the data points assigned to each cluster.
• Step 4: The data points are reassigned to the nearest centroid based on the updated centroids.
• Step 5: Steps 3 and 4 are repeated until convergence, where the centroids no longer change significantly or a maximum number of iterations is reached.
In the above equation, J is the total within-cluster variance, K is the number of clusters, C i represents the i th , and μ i represents the centroid of the i th cluster.
The sentiment analysis framework is shown in Figure 5.
The cosine similarity score measures the distance between two vectors in the semantic space regardless of their magnitude.As the raw text (sentences) is represented as vectors in semantic space, the cosine similarity will present with the most similar sentences for the given sentence.The cosine similarity metric ranges between −1 and 1, where score "1" indicates the vectors that are very close by (similar sentences), "0" indicates the orthogonal vectors, and "−1" indicates the vectors that are opposite (most dissimilar sentences) in the semantic space.The cosine similarity scores presented in Tables 5, 6, and 7 represent the ten most similar sentences for the given sentence in each generated cluster.Similar sentences are induced based on the cosine similarity metric.
Table 5 (a) presents cosine similarity scores for freetext comments obtained from a survey on the intention to use ChatGPT.They show the three focal comments: "nil," "risky app," and "great invention."The focal comment is "nil," and the scores indicate how semantically similar other comments are to this term.The high similarity scores for comments such as "nope," "nopes," and "no" (ranging from 0.8520 to 0.8750) indicate that a significant subset of respondents expressed a negative or null intention toward using ChatGPT, reinforcing the neutrality or nonengagement suggested by "nil."Specifically, they could indicate a lack of perceived relative advantage or compatibility of ChatGPT within the educational setting.Phrases such as "did not try" and "cannot believe it," which scored approximately 0.78, may reflect skepticism and could be associated with perceived complexity or lack of trialability, implying that students might not find it easy to experiment with the ChatGPT before deciding to adopt it fully.The less similar phrases, such as "I do not know about it" and "I used it for assignments," suggest ambiguity or      "risky application" and "dangerous tools," with high cosine similarity scores close to 1 and 0.85, may reflect a perceived lack of relative advantage and compatibility.They suggest that respondents may find ChatGPT to be too risky to offer any meaningful benefit, particularly in an academic context.Comments such as "scared what is next with ai" and "there are high chances of getting inaccurate answers," scoring approximately 0.77, hint at concerns that could be mapped to the attribute of observability.These findings indicate a general concern about the unobservable or uncertain outcomes related to ChatGPT and AI.The sentences "scary but useful tool is ChatGPT" and "ChatGPT can be scary and terrific simultaneously," with scores above 0.75, represent a dichotomy.While they acknowledge the potential utility (relative advantage), they also express apprehension (lack of compatibility) that may hinder the trialability of the system.Finally, comments such as "Yes will replace humans.be ready" and "it becomes a troublemaker sometimes, by showing irrelevant information," scoring approximately 0.71, indicate more nuanced concerns that could relate to both relative advantage and compatibility.These comments could imply that while ChatGPT may be useful, its capabilities are viewed as potentially disruptive or problematic, affecting its overall perceived value and adoption.Table 5 (c) illustrates sentences with high semantic similarity to the phrase "great invention," which all fall under a cluster characterized by positive sentiment.The phrases "excellent invention" and "it is a great innovation" display high cosine similarity scores (above 0.96), indicating strong agreement with the comment "great invention."This finding suggests that users perceive a clear relative advantage in using ChatGPT, possibly considering it superior to existing solutions within educational contexts.Sentences such as "very useful," "very much useful," and "very helpful," scoring approximately 0.89-0.90,can be mapped to the attributes of relative advantage and compatibility.These terms suggest that users find the tool advantageous and compatible with their existing workflows or educational processes.Pharma such as "very good" and "very good platform," scor-ing above 0.88, also indicates relative advantage and simplicity.These comments indicate that the respondents found the ChatGPT to be effective and easy to use, contributing to its potential for quick adoption.Finally, the comments "amazing  software" and "very creative tool," with scores of approximately 0.88, perhaps speak to relative advantage and observability.These terms suggest that the benefits of ChatGPT are recognizable and highly valued, making the tool's advantages easily observable to potential adopters.In summary, the cosine similarity scores in this cluster reinforce the positive perceived attributes of ChatGPT, including its relative advantage, compatibility, and simplicity.These insights provide a robust understanding of the factors that may promote the adoption of ChatGPT among students in higher education.
For further analysis, the mean and standard deviation are calculated to evaluate the quality of the generated clusters.The mean, also known as the centroid, refers to the center of a cluster and is calculated by averaging all the data points in the cluster.The centroid is the point that minimizes the sum of the squared distances between all points in the cluster and itself [83].The standard deviation, a measure of the data spread within a cluster, is calculated as the square root of the variance.The variance is the average of the squared distances of each point from the mean.In Kmeans clustering, the objective is to minimize the sum of the squared distances between each data point and its assigned cluster's centroid [83].The mean and standard deviation are used to assess how well the data points are clustered around their respective centroids.The clusters are considered well separated and well defined if the means are close to the centroids and the standard deviations are minor.
Table 6 provides the mean and standard deviation for three sentiment clusters-neutral, negative, and positivegenerated using K-means clustering.
• Cluster 1 (neutral) had a high mean of 0.90 with a relatively low standard deviation of 0.08.This suggests a strong, concentrated sentiment around neutrality, possibly mapping Rogers' attribute of trialability.Here, the users might be evaluating ChatGPT but have not yet fully recognized its benefits or drawbacks, leading to a densely packed neutral sentiment.
• Cluster 2 (negative) has a lower mean of 0.70 and a standard deviation of 0.09, indicating a less favorable view.This may be related to perceived attributes such as complexity or incompatibility, according to Rogers' theory.A lower mean and greater spread could imply that negative sentiment is not universally shared but could be significant enough to warrant attention for potential system design or functionality improvement.
• Cluster 3 (positive) exhibited a mean of 0.80 and a standard deviation of 0.09.Compared to Cluster 2, this higher mean aligns with attributes such as relative advantage and compatibility.The broader spread, as indicated by the standard deviation, may suggest differing levels of enthusiasm, possibly due to varying experiences with observability or simplicity as per Rogers' framework.
Table 7 indicates that the sentiment distribution based on gender found through the frequency analysis of tokenised words revealed distinct sentiment distributions across genders.The results highlight gender differences in perceptions and discussions about the ChatGPT.Males focused more on the utility of the ChatGPT, while females expressed concerns regarding its reliability.
For positive sentiments, males frequently use words such as "good," "useful," and "chatgpt," highlighting the utility and general appreciation for the technology.In contrast, females in the positive category emphasize terms such as "easy," "using," and "helpful," focusing more on usability and the practical assistance the technology offers.This suggests that while males are concerned with functionality, females value ease of use and direct benefits.
For negative sentiments, males commonly use words such as "irrelevant," "questions," and "information," which could indicate dissatisfaction with the relevance or utility of the information provided by the technology.Conversely, females use more critical terms, such as "dangerous," "errors," and "wrong," indicating concerns about the accuracy and safety of the technology and highlighting more critical engagement with potential risks.
For neutral sentiments, males list words such as "chatgpt," "using," and "lot," suggesting a balanced or varied interaction with the technology that does not sway strongly toward positive or negative.Females, however, use words such as "none," "nil," and "nan," indicating a lack of strong feelings or a significant impression left by the technology, possibly reflecting indifference.

Discussion
We discuss the findings of our study more closely in sync with our proposed research questions.
RQ1: What are the key enablers of ChatGPT adoption for students from an innovation theory perspective?
Our study explored the usage intentions of ChatGPT based on the diffusion of the innovation model.The findings from the quantitative analysis are discussed in the following subsections.

Relative Advantages of
ChatGPT.ChatGPT has several potential advantages over other educational technologies, including the following.
• Rapid information accessibility: ChatGPT grants students swift and convenient access to extensive topic- related data regardless of geographical location [84].This particularly benefits students in isolated areas lacking easy access to conventional educational resources [80].
• Customized learning experiences: Through interactions with ChatGPT, students can achieve a tailored learning journey [56].Such interactions include posing queries, receiving work feedback, and delving into varied learning trajectories, allowing students to study at a pace and method tailored to them.
• Time-saving flexibility: ChatGPT can help students conserve time and facilitate concentration on other academic facets by managing routine tasks such as searching and editing [85].
The quantitative sentiment analysis further supported these advantages, particularly given the gender differences in positive sentiment, where males reported more positive sentiments than females did.This indicates a stronger perceived relative advantage among male students.In Halaweh's [86] study, students identified the ChatGPT as an innovative tool for independent educational pursuits.Noted advantages such as instantaneous information access, customized learning, and time-saving flexibility drive students toward ChatGPT, indicating that ChatGPT is superior to other educational technologies.Consequently, the adoption of ChatGPT is forecasted to surge in future years.

Compatibility of ChatGPT.
Early adopters, who are typically conversant with tech and amenable to novel technologies, find that ChatGPT aligns well with their educational technology needs.The system's capabilities-such as instant information retrieval, personalized learning, and time savings-are not lost.ChatGPT can seamlessly complement existing digital learning tools [48].It can be utilized for myriad purposes, including generating practice questions, essay feedback, and exam preparation to enhance learning efficacy and efficiency [15,54].Nonetheless, reservations have been aired concerning the potential misuse of ChatGPT.Studies by Cotton, D. R., Cotton, P. A., and Shipway et al. [64] and Steiss et al. [48] highlighted possible pitfalls such as content plagiarism or completing tasks without genuine comprehension.Addressing these concerns mandates mechanisms to forestall the potential misapplication of ChatGPT.Potential solutions include educating students on responsible AI usage, crafting ChatGPT-specific plagiarism detectors, and endorsing student support [1,19].Nevertheless, early educational tech adopters find ChatGPT to be a fitting innovation, as evidenced by the positive yet modest coefficient of the effect of compatibility on student intentions.Despite the potential for misuse, ChatGPT is still a highly compatible innovation for early adopters of educational technology [56].This is reflected in the positive, significant, yet low path coefficient of the influence of compatibility on students' intentions.

Use of ChatGPT.
Although recently developed, the ChatGPT has been intentionally constructed for simplicity.It facilitates interaction through straightforward language and commands and furnishes feedback to refine user efficacy.Amjad et al. [27] elucidated that students can easily engage with ChatGPT for query resolution.Students could rapidly and effortlessly adapt to ChatGPT's mechanisms regardless of prior exposure to AI technology.Moreover, Wu et al. [3] underscored its proficiency as a tool for spawning diverse creative text formats, from poetry to scripts.The predominant ease associated with the deployment of the ChatGPT positions it as a significant prospect in higher education milieus [8,27].
6.4.Trialability and Observability.ChatGPTs present a myriad of exploration opportunities for discerning their pedagogical advantages.It enables students to produce text, transition between languages, craft varied content, and derive insightful answers without preliminary engagements.Academic studies emphasize the prominence of trialability and observability in educational tech adoption.According to Thorp [62], trialability notably influences student intent toward ChatGPT utilization.Concurrently, findings from Rudolph et al. [61] and Mogavi et al. [49] point to the observable merits of ChatGPT-ranging from enriched learning trajectories to heightened student participation-as catalysts for its integration.The platform's autonomy, which allows students to independently generate exercises or procure feedback, accentuates its trialability [69].Differences in sentiment perceptions, especially greater neutral sentiments among females, suggest the need for more observable and trialable features that directly demonstrate the benefits of ChatGPT, potentially increasing its attractiveness and usability among female students.Furthermore, the demonstrable outcomes of ChatGPT, which are evident when tutors craft bespoke learning resources or initiate interactive modules, augment its observability.
RQ2: What are the sentiments of students toward ChatGPT enablers for education?
The sentiment analysis suggested three divergent segments or clusters among students who view ChatGPT as "not useful," a "risky app," and a "great invention" [87].With the first set of respondents expressing negative feedback regarding the use of ChatGPT, it may be inferred that a population in educational settings has yet to realize relative advantage and/or compatibility in this educational context [1].The negative sentiment comments about ChatGPT in education emphasize the potential reduction in critical thinking ("ChatGPT is like spoon feeding us with answers"), ethical worries over academic dishonesty ("A ban should be imposed at least in the college WiFis"), frustrations about reliability ("Sometimes it gives wrong answers"), and a preference for traditional methods ("I still like searching for codes on Google than ChatGPT").Furthermore, a subsection also represented sufficient skepticism, which also indicates that many students do not have enough scope [57] to explore ChatGPT before deciding to use it.Some segments also suggest that there is a lack of observability of the technology's benefits in the educational context, as some ambiguity is present.
The second cluster of respondents indicated that some respondents feel that it does not offer substantial benefits  [88] mandating adoption, thereby again representing a lack of perceived relative advantage and/or compatibility.Furthermore, this cluster also showed that there is a general stress prevailing among the students in education settings regarding consequences related to ChatGPT and AI.While they acknowledge the potential utility (relative advantage), they also express apprehension (lack of compatibility) that may hinder the trialability of the system.Neutral comments on ChatGPT express a mix of caution and moderate approval.Users acknowledge its helpfulness but caution against overreliance ("It is great and easy to use, but I feel like it must be limited").Concerns are raised about its potential to diminish student diligence ("using chatgpt may decrease a student's hard work").Still, some see its utility for specific tasks like improving presentations or overcoming learning hurdles, though with more comprehensive capabilities ("ChatGPT is a good platform but needs to be updated").ChatGPT may be perceived as useful because its capabilities viewed as potentially disruptive or problematic, affecting its overall perceived value and adoption [9,64].
The third cluster of respondents exhibited more positive findings, suggesting that ChatGPT is a superior tool to other conventional tools within educational settings.In the third cluster, many users find the tool advantageous and compatible with their existing workflows [15,80] or educational processes.
The positive comments highlight ChatGPT's effectiveness as an educational tool, enhancing learning and simplifying tasks like coding and writing ("ChatGPT will make writing/coding assignments a better experience").Users appreciate its quick, accurate responses and its ability to correct mistakes, which enhances learning efficiency ("It's really a great experience with ChatGPT").Many see ChatGPT as a valuable partner in education, aiding in personalized learning and facilitating a deeper understanding of subjects ("ChatGPT helps a lot in learning about coding or other subjects").Additionally, within this segment, many respondents found ChatGPT to be effective and easy to use, contributing to its potential for quick adoption.
RQ3: Are there any perceived differences concerning gender regarding ChatGPT enablers?6.5.Gender-Based Differences in Attribute Preferences.Although male students' alignment toward observability might be rooted in a quest for societal validation as pioneering users, females accentuate the palpable superiorities of new tools (relative advantage) and their proclivity for preliminary assessments (trialability).Such divergences possibly mirror males' historical tech affiliations and broader cultural or societal imprints on females [89].The findings from the sentiment analysis, particularly the neutral and positive sentiments, show that while both genders recognize the potential of ChatGPT, their enthusiasm for male students and cautiousness toward female students significantly increase.These findings suggest that males and females have different expectations and experiences with technology.Males appear to focus on functional aspects, while females are more attuned to operational and ethical implications.These gender-specific preferences could influence how educational technologies such as ChatGPT are adopted and utilized in different learning environments.In summary, both genders underscore the imperative of cohesive tech integration, albeit driven by different motivations: men by societal affirmation and women by utilitarian and precautionary considerations [90].This finding resonates with prior scholarly investigations into the nuances of gendered tech adoption, epitomized by Venkatesh et al. [44].
Several scopes exist for further analysis regarding the deployment of ChatGPT for students in education.First, future studies can capture the potential differences in how different academic disciplines and allied students perceive ChatGPT usage.For example, engineering and management students will likely perceive ChatGPT differently based on their selective and specific usage [12].Across different domains, studies must focus on having a comparable sample to capture allied gender-driven differences in usage intentions.Second, a comprehensive comparison of the enablers from the DOI theory perspective gender wise across developed countries [1], namely, the US and the UK, which have emerging economies such as India, would be interested in capturing meaningful insights.Third, future studies can examine how certain other psychological factors, such as stress and anxiety, further shape the enablers from the DOI theory [20] and affect behavioral intentions for ChatGPTs.This study also has to capture how gender differences affect the development of relative advantage, compatibility, observability, and others.Fourth, studies also have to capture the potential differences between Gen Z learners and Gen Y learners and how the same varies for male and female learners among the respective generations.Finally, our study has a sufficient sample size for the empirical validation of the proposed model, which relies on perceptual responses collected from university students in India.This limits the generalizability of the findings to only related emerging economies [1].Therefore, future studies should aim to collect data from students studying in four different locations based on the prevailing prominent diversity present among East, West, North, and South India.Further research should focus on a longitudinal cross-comparison of the ChatGPT [91] with other emerging tools, such as Gemini and Microsoft copilots.A longitudinal study among emerging versus developed countries regarding the usage behavior of ChatGPT with Gemini and Microsoft Copilot would further reveal interesting insights, helping to understand the changing Gen Z landscape and their orientation toward AI tools for education.6.6.Implications.Using Rogers' [20] DOI theory as a primary lens, our study revealed that relative advantage, compatibility, trialability, observability, and ease of use are prominent enablers of students' adoption intention of ChatGPT.Our study also explored gender differences regarding the role of enablers in shaping adoption intentions.While male students find ChatGPT easy to observe and comprehend for their daily use and, therefore, easier to explore (trial), female students find ChatGPT to have a greater advantage than other tools for day-to-day usage and easy to understand, explore, and use.However, similar to males, females do not feel that ChatGPT, as a tool, is easy to learn simply through observation.This suggests that, in general, both genders prefer easy-to-use and easy-tounderstand ChatGPT and that the ability to explore ChatGPT is a key attribute influencing their usage/adoption intentions.Although extant studies have shown that perceived ease of use does not shape one's attitude toward ChatGPT directly but rather acts through its impact on perceived usefulness for teachers [6], our study showed that for students in education, ease of use directly shapes usage intentions within the DOI theory framework.While Yilmaz et al. [8], using TAM, found significant gender differences regarding "perceived ease of use" and perceived differences regarding social influence driven by academic disciplines, our study further complemented these findings by empirically suggesting that male and female students perceive social influence on them differently.
Apart from the empirical model, our study further used sentiment analysis, which furnished qualitative insights into the attitudes and emotional perspectives of the students toward this AI-driven educational tool.The sentiment analysis of student opinions on ChatGPT reveals a complex of attitudes significantly influenced by gender biases [7,8].The highly male-centric nature of the responses [92], particularly in the neutral and positive clusters, raises questions about the generalizability of the study's findings.According to Rogers' theory of perceived attributes, these skewed data may primarily capture male perspectives on relative advantage, compatibility, and other attributes, thereby missing the nuances that a more balanced gender distribution might offer.This study underscores the need to use diverse and representative datasets for more reliable and equitable technological adoption assessments, as emphasized by Jim et al. [80].
Gender biases in sentiment analysis skew the representation of innovation attributes and hinder the development of inclusive technological solutions in higher education settings.Therefore, future research must focus on eliminating these biases [89,92] to better understand how different demographic groups perceive and adopt innovations such as the ChatGPT.The confluence of these quantitative and qualitative methodologies strengthens the robustness of the findings and offers a nuanced understanding that is beneficial for both researchers and practitioners in educational technology.Integrating the DOI theory [20] with sentiment analysis in this research contributes to the existing body of literature in several ways.First, it offers a multidimensional approach to understanding technology adoption, blending behavioral and emotional aspects.This amalgamation enriches traditional TAMs [6][7][8]15] that predominantly rely on quantitative factors, providing a more holistic understanding.Second, by demonstrating the relevance and applicability of the five innovation attributes of DOI theory in the context of ChatGPT adoption among university students, this study broadens the scope of the theory beyond its traditional applications.
From a practical standpoint, the findings provide actionable insights for educational institutions, policy-makers, and developers of AI-driven educational tools such as ChatGPT [15,52].Understanding that all five innovation attributes significantly influence students' intention to use ChatGPT can guide the design and implementation strategies of such technologies [19].For instance, ease of use can be prioritized in interface design [6], while trialability can be facilitated through pilot programs or free trial periods [13].Sentiment analysis outcomes can address specific concerns or preferences among user groups [80], thus enabling a more tailored approach to technology integration in educational settings.Moreover, the identified gender bias in sentiment analysis underscores the importance of considering demographic factors [13,89,92] when designing and evaluating AIbased tools, thus informing more equitable technology implementation strategies.
ChatGPT, an advanced AI-driven language model, has garnered significant attention and rapid adoption across various sectors, including education.The potential benefits and applications of ChatGPT in education, such as personalized tutoring, automated essay grading, and adaptive learning experiences [1,58], have been highlighted in multiple studies.However, concerns regarding ethical issues, academic integrity, privacy, and dependency [56] on technology persist, leading to a divide in opinions on adopting the ChatGPT in academic settings [58].As ChatGPT gains more widespread use and recognition, users who adopt the technology after seeing its success and hearing positive experiences from early adopters could be classified as part of the early majority.These individuals may be more cautious about new technologies but are open to adopting them once they see the benefits [24,58].As AI language models such as ChatGPT become increasingly mainstream and integrated into various applications [48], users who adopt the technology because it is widely accepted and utilized likely fall into the late majority category.The rapid use of ChatGPT necessitates a thorough understanding of the factors contributing to its acceptance and use among students [1,7].By identifying these factors, stakeholders in education can make informed decisions on the implementation and integration of ChatGPT [15,27], striking a balance between harnessing its potential benefits and mitigating associated risks and challenges [16].
We can draw parallels between ChatGPT's diffusion and the diffusion of other technological innovations, such as social media platforms, instant messaging apps, and smartphone adoption [24,25].Social media platforms such as Facebook, Twitter, and Instagram have experienced rapid diffusion over the past two decades.Like ChatGPT, social media platforms have changed how people communicate, interact, and share information [27,56].These platforms have gained widespread popularity due to their ease of use, accessibility, and ability to connect users with their friends, family, and broader communities.Both innovations appeal to early adopters who recognize the potential benefits and are eager to try new communication and informationsharing methods [25,48].As social media platforms became more widely adopted, they transitioned from early adopters to the early majority and beyond.By analyzing the factors that contributed to the diffusion of social media platforms, we can gain insights into the potential trajectory of ChatGPT adoption and growth [46,52].
The study's findings illustrate the importance of customizing AI tools like ChatGPT for diverse educational settings and student populations.Considering the regional specificity of the sample, practitioners must study and adapt AI integration to local educational norms.The gender differences observed highlight the need for these tools to accommodate diverse learning styles and preferences.Finally, AI technologies should enhance accessibility for students with disabilities through inclusive design and adaptable interfaces.
When integrated into various disciplines within academia, such as engineering, medicine, legal, art, sciences, and business, among others, ChatGPT would pose a unique set of challenges and barriers for learners.For instance, students may lack the necessary finesse to solve complex problems accurately.AI tools may not handle specialized design, or students may not be able to trust them with reliability in critical projects, in the case of engineering students.Medical students may have challenges with the sensitivity of data and the accuracy of medical advice due to the high stakes of clinical information accuracy.Likewise, legal students will be torn between upholding ethical considerations through plagiarism and prohibiting an overreliance on technology, which may hinder the development of critical thinking and independent research.Arts students may become silenced to their nuance or complexity in creative thought culture and history, relying on design as a scalar.Likewise, a science student may have an insufficient tool to conduct proper experiments, and its utility may not be very deep in doing practical research.Business students may have a hard time how interpreting complex data and analysis as they may also be overreliance.Common challenges in all the fields cut across the variable level of digital literacy, skepticism on AI's accuracy, and rate of pace of how they can keep up with the technological changes.The integration between traditional learning and ChatGPT should strike a proper balance.

Conclusions
Using a mixed-method approach through the lens of DOI theory complemented by sentiment analysis, our study contributes to a comprehensive understanding of adoption intentions for ChatGPT for education students.Based on the perceptual responses of 288 university students, our study revealed that relative advantage, compatibility, ease of use, observability, and trialability are significant enablers of student behavioral intentions for ChatGPT.Compared to existing tools, students have found ChatGPT to be quite easy to learn due to its innovative features, and it can have wider applicability.Furthermore, students have a significant ability to explore ChatGPT's capabilities through different trials; therefore, learning and adopting the same approach becomes easier.As the visibility of the benefits of using ChatGPT in education increases, so does students' intention to use it.Furthermore, the sentiment analysis suggested a complex behavioral landscape across genders and education sectors.While male students find ChatGPT easy to observe and comprehend for their daily use and, therefore, easier to explore (trial), female students find ChatGPT to have a greater advantage than other tools for day-to-day usage and easy to understand, explore, and use.However, similar to males, females do not feel that ChatGPT, as a tool, is easy to learn simply through observation.Theoretically, our study contributes significantly by offering a multidimensional lens combining behavioral and emotional aspects through sentiment analysis with the DOI theory.
This study acknowledges limitations that may potentially limit its generalizability.The participants were predominantly from specific regions within India, limiting the applicability of the findings to different geographic or cultural contexts.The sample was restricted to university students, which may not reflect the perceptions and adoption rates in other educational settings.

Figure 1 :
Figure 1: Research model for student intentions to use ChatGPT.

Figure 2 :
Figure 2: ChatGPT structural model ((a) CB-SEM model; (b) PLS model).# p = one-sided p value, as we know the direction of the hypothesis.

Table 3 :
Measurement items: reliability and validity.
hbet, 2024, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2024/3085910byUmea University, Wiley Online Library on [18/07/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)onWiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License conditional usage, possibly indicating a lack of clear observability of the technology's benefits in the educational context.Table 5 (b) shows the top ten sentences semantically similar to the comment "risky app," belonging to a cluster characterized by negative sentiment.Sentences such as 11 Human Behavior and Emerging Technologies hbet, 2024, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2024/3085910by Umea University, Wiley Online Library on [18/07/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Downloaded from https://onlinelibrary.wiley.com/doi/10.1155/2024/3085910by Umea University, Wiley Online Library on [18/07/2024].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License