Gamification of Personalized Learning through Massive Open Online Courses: Learner-to-AI Enabled Chatbot

Abstract

Massive open online courses (MOOCs) are considered an educational revolution that widens access to quality education and enhances social inclusion. With their fast advancement and potential influence in education, they have been enlisted in the modernization agenda of many universities globally. Nevertheless, the effectiveness of learning through MOOCs is debatable, while student retention rate ranges significantly. The primary reason is considered the lack of interactivity in MOOCs, which urges enhancement of interaction between teachers and students. Another challenge regarding the MOOCs is to find the best resource fitting a learner’s personal profile, interests, background, and learning needs. The first challenge has been addressed from the gamification point of view to measure the impact of gamification on the overall success of MOOCs. However, the second challenge still remains untouched. In online learning setting, course design and interaction with instructors as well as students are the factors that greatly influence students’ perceived learning and satisfaction with the online course. The contribution of chatbot-supported MOOCs recommender system is in two-fold. Hence, the current study aims to extend the gamification concept to chatbot-supported MOOCs recommender system with two types of chatbots, namely text and avatar + text chatbot.

1. Introduction

Artificial Intelligence (AI) improves the quality of decision making and problem solving in various industries by exploiting machine intelligence in an effective way (e.g., neural networks and machine learning) [1]. Commercial applications of AI are mainly healthcare, high tech, financial services, automotive, media, retail, and travel industries [2, 3, 4]. In the context of a private sector is the deployment of applications interacting with users in a conversational as well as human mimic format, which is also referred as “conversational agent,” “bot,” or “chatbot” is the most popular AI trend [5]. Education is another field that through AI it provides tremendous potential for learners. It has been evolved from experimental laboratory scenario to the real-time learning platform with its complexities as well [6]. It is added that firms that offer educational technology, or so-called “EdTech” parallelly enhance Individual Adaptive Learning System (IALS) for the aim of providing personalized learning experience. Not limited to learning, the system also assists in classroom activities, such as management, evaluation, and grading of learners, while also solve language problem. According to Mou [7], from 2008 to 2017, global investments into education powered with AI have reached to over hundreds of billions of US dollars.

Chatbots are tools of AI that are massively applied in banking, healthcare, insurance, government services platforms as well. They are acknowledged as taking actions to increase the probability of success reaching a certain goal in response to perceiving the environment [8]. According to Business Insider [9], there are many popular chatbots that serve as virtual assistants, ranging from Alexa to Siri and Google now. They help people to chat with agencies, make bank transactions, get news, book hotel rooms and so on.

What happens in learning environment if AI powered personalized learning is applied? Personalized learning is an approach of education that customizes learning program based on each learner’s needs and skills, as well as interests [10]. In a classroom environment, the application of AI helps to retrieve history of past learning activities, incorporates with current data and a learner’s profile as well as learning gaps and then suggests what steps need to be taken to improve academic performance.

Trends, challenges, and technology developments in education evolve over time with new perspectives and dimensions every year [11]. For instance, mobile and online learning today are not what they were yesterday. Some research works have investigated and analyzed the use of mobile devices in m-learning environment by university students’ and the effect of personality, which addresses data breach in education and the role of excessive technology use on risky cyber-security behavior of students [12]. Yet another research investigated the effect of educational technologies on the perception of tourism students and their intention to work thereafter. Different M-learning and E-learning platforms were compared to assure the most effective platform for CSR education among the students, which eventually found that e-learning tools are rather more effective [13]. Hence, virtual reality, chatbots, and immersive apps have added more functionality and greater potential for learning. While notable examples of artificial intelligence (AI) are utilized in a classroom setting, administrative tasks also implemented with the use of AI [11]. For instance, some institutions improve teacher evaluations by using AI-enabled chatbots to record, organize, and provide detailed feedback from learners.¹ Georgia State utilized Pounce–a chatbot that helps incoming learners navigate through complicated application process and provides a personalized checklist for completing financial aid and enrolling in courses.²

There are several areas of research interest in HCI: (1) Exploration of interfaces that expand representational possibilities beyond metaphors to the extent that users have already been accustomed, such as desktop icons; (2) technology-aided cooperative work; (3) human perception and cognition coupled with the task analysis, which is a particular focus of the current study due to the fact that little attention has been paid to socio-cognitive nature of human-technology interaction with particular emphasis on chatbots [14].

According to Class Central³, approximately 23 million new learners started using massive open online courses (MOOC) in 2017, taking the total number of learners reaching to 81 million. The top MOOC providers are Coursera, edX, XuetangX, Udacity, and FutureLearn. Among them, edX is particularly aimed for professional education. Currently, MOOCs cover almost all study disciplines and educational levels. The major motivation for the creation of MOOCs is pedagogical and educational innovation [15, 16]. MOOC refers to online course that involves learners in learning process, offers a way for learners to connect and collaborate, and provides a platform where course materials are shared among learners.

The effectiveness of learning through MOOC is debatable, while learner retention rate ranges between 10 and 50% [17]. The primary reason is considered the lack of interactivity in MOOC, which urges enhancement of interaction between teachers and learners [18]. Another challenge regarding the MOOC is to find the best resource fitting learner’s personal profile, interests, background, and learning needs [19]. The study proposed “MOOCBuddy” chatbot that acts as MOOC recommender system based on learner’s interests as well as social media profile and it is based on Facebook messenger platform. It is shown that the one way that chatbots transform education is providing teaching assistance.⁴ For instance, Georgia Institute of Technology introduced Jill Watson—a new teaching assistant that responds to learner inquires in a fast and accurate way. It is powered by IBM’s AI system collecting more than 40,000 forum posts.

1.1 Research aim

Although chatbots are gaining massive interest in education, much systematic and empirical work is still needed to approach human-chatbot interaction from psychological aspect and develop theoretical as well as practical guidelines for the design and development of this technology, especially with consideration of communication style alignment between humans and chatbots. Drawing from the previous studies, this research addresses:

• What are the salient aspects of learners’ perceptions toward the chatbots that assist them in finding relevant course-driven MOOCs as part of their studies.

• How does learner-chatbot communication style similarity affect the above aspects?

• How does learners’ intention to use chatbots affect their academic performance?

This research particularly focuses on text-stream chatbots. Because, Ciechanowski et al. [14] found that users experience less negative feeling in engagement with text chatbot in comparison with animated avatar chatbot. Finally, academic performance is tested as an outcome of learner-chatbot interaction. Fryer et al. [20] found a significant drop in learners’ task interest with chatbot compare to human in terms of language learning. It may be related to the area of chatbot application. As such, previous sections highlighted the successful application of chatbots in different fields, while chatbot was found to be less effective in language learning. Therefore, it is worthy to investigate whether learners will have positive or negative perceptions of chatbots in finding course-driven MOOCs based on their specific needs.

1.2 The learner-Chatbot interaction process

If the effectiveness is the focus point, the chatbot should be designed in a way that it can convince a learner to be an appropriate tool to perform learning task and to be beneficial in delivering the learning outcomes. According to Ajzen [21], there is a consistency between a user’s belief and intention to perform a task and reach a target.

Figure 1 illustrates a learner’s interaction process with chatbot. According to Al-Natour and Benbasat [22], the factors of outcome, occurrences during learner’s interaction with chatbot, and results of the interaction are identified in the whole process. Moreover, object-based and behavioral beliefs combined with the task (online learning) represent the cognitive processing of learner’s behavior. The proposed model has a guiding role in design of system process that chatbot may have a broader capacity that will not lead learners to perceive it less useful in different tasks as well as different phases of interaction from initiation to completion, and to stop using it.

Figure 1.

The learner-chatbot interaction flowchart.

1.3 Similarity-attraction theory and learners’ interaction with Chatbot

Recently, information system (IS) research has mainly focused on the effect of cognitive and emotional aspects on different behaviors [23]. Both negative (e.g., anxiety and helplessness) and positive (e.g., enjoyment) emotions facilitate user behavior and openness to information reception [24]. In a functional context, when users trust the information they receive and feel informed well, it encourages their tendency for seeking information [25]. Moreover, AI chatbots with higher utilitarian and hedonic features and similarity of chatbot-learner communication style can be expected to enhance learner’s adoption of this new communication channel. On the contrary to utilitarian features, hedonic features heavily focus on fun-aspect of IS usage, and encourage prolonged usage rather than productive usage [26].

Based on the similarity-attraction theory the more similar a learner’s features and beliefs are to those of other parties, the more likely that learner will be attracted to and build better perceptions of that parties [27]. In the context of virtual intelligent agents, humans regularly perceive them to have human-like behaviors and personalities [22]. Similarity reduces uncertainty, while increases validation and enjoyment of interaction [28]. In their study, Al-Natour and Benbasat [22] conceptualized user-IT artifact relationship and proposed that object-based beliefs, such as individualistic (e.g., information quality and interactivity) and dyadic (e.g., personality similarity and complementarity) factors create behavioral and relationship beliefs. In another study, Al-Natour et al. [29] found that personality and behavioral similarities between users and online shopping assistants positively lead to user evaluation of the technological artifact. Al-Natour et al. [30] distinguished between personality and decision process similarity and found that personality similarity affects trusts, while decision process similarity influences enjoyment and ease of use of online shopping assistants.

To sum up, it is assumed that the similarity-attraction paradigm can be applied in the adoption of AI chatbots by the learners in learning delivery in the context of MOOCs. Hence, this research categorizes task-related and channel-related factors [31], in order to guide prospective system designer’s MOOC system by incorporating above aspects in chatbot-supported communication channel.

1.3.1 Direct relationships

Task value (TV) has been extensively used in educational context for evaluating learners’ achievement and academic performance. It refers to evaluating how important, interesting, and beneficial the task is [32]. TV is related to attachment to perform well on a given task, pleasure gained from it, and its contribution to long-term goals, as well as cost and energy invested in performing the task [33].

Informativeness (INFO) is explained with the user perception of the system to effectively provide appropriate information [34, 35]. It expresses the feeling of being informed about a specific feature of products or services, such as technical capabilities, which leads users to be positively associated with using that products or services [36]. In the context of virtual health advisory system, Li and Mao [23] proposed that communication similarity between user and virtual health advisory system could reduce uncertainty and make users feel their need for information is easily fulfilled through the system usage.

Similarity might affect building pleasurable interactions [23]. Decision process and personality similarity significantly impacts perceived enjoyment (ENJ) [29]. In addition, Li and Mao [23] found that perceived ENJ positively and significantly impacts both reuse intention of virtual health advisory system and social presence.

Transparency (TRS) is significantly and positively influenced by communication similarity between user and virtual health advisory system, while it also influences reuse intention [23]. In their study, Baxter and West [37] highlighted that similarity creates predictability between partners and enables them to interact with greater confidence.

Similarity in communication style enhances feeling of credibility (CRE) [38]. In consumer research, it has been found that internal similarity increases buyer’s disposition to trust salespeople and follow their supervision [30]. However, it was found that credibility does not influence reuse intention of virtual health advisory system and social presence [23].

Trust on internet (TI) is regarded as e-government users’ belief on the reliability of Internet for information accuracy and transaction security [39]. Learners must trust Internet as e-enabler to keep their information secure and private to accept and adopt e-government services [40]. In the context of this study, it is proposed that chatbots can be considered as e-enablers, while learners’ trust on them can be crucial in their motivation e-government service adoption.

Irreplaceability (IRR) refers to the extent that a certain product has a symbolic meaning to a person that is not apparent in other products, even if they are physically identical [41]. When a product is irreplaceable, consumers are less likely to replace it, whereas more likely to retain it for a long term [41]. To our best knowledge, there are not many studies utilized irreplaceability in the MOOC learning context. It is proposed that IRR of chatbots in e-learning service delivery could have significant influence on the adoption of this technology by learners.

Drawing from the discussion above, the current research proposes the following hypotheses:

H1. Communication style similarity will significantly impact task-related factors in learner-chatbot interaction.

H2. Communication style similarity will significantly impact channel-related factors in learner-chatbot interaction.

H3. Task-related factors will significantly impact learner’s intention to use chatbot-supported e-learning service in MOOC.

H4. Channel-related factors will significantly impact learner’s intention to use chatbot-supported e-learning service in MOOC.

1.3.2 Moderating effect

Ciechanowski et al. [14] studied human-chatbot interaction with different types of interfaces. The two types of chatbots, namely simple text and avatar chatbots were used in the experiment. The major emphasis was on feeling of eeriness and discomfort about the medium in human-machine communication, and results showed that users experienced less negative feeling in association with simple text chatbot compare to more complex (animated avatar) chatbot. It can be proposed that before introducing the chatbot, it is essential to understand user psychology and design better chatbots to contribute to human-machine interaction. Therefore, in the current study, text-based chatbot is tested as a main moderator (see Figure 2). Drawing from the findings, the following hypotheses are proposed:

Figure 2.

MOOC learner-chatbot interaction model.

H5. Text chatbot type will have moderating role between task-related factors and learner’s intention to use chatbot-supported e-learning service in MOOC.

H6. Text chatbot type will have moderating role between channel-related factors and learner’s intention to use chatbot-supported e-learning service in MOOC.

2. Methods

Primarily, the current study aimed to investigate the perceptions of learners toward the AI supported chatbot. To attain this aim, in total of 32 participants selected randomly based on the population were given open-ended survey questionnaires, through which the researcher could identify the most attractive channel for learners to use in e-learning process. For instance, it could be through online courses, YouTube channels, various websites with learning resources, and so on. Henceforth, six criteria (C) for evaluation were identified in regards to three learning channels (Ch), namely, (1) Web search, (2) YouTube learning channels, and (3) online courses official websites. Those criteria are (1) Information retrieval time; (2) Request on personalized learning; (3) Sense of human touch; (4) Comfortability; (5) Credibility; (6) Ease of use of channel.

Results based on a mixed methodology approach revealed significant difference among the six criteria across three e-learning channels (F (3, 28) = 0.21, p > 0.05). In terms of C1, web search is more effective (M = 4.08, SD = 0.87) than YouTube (M = 3.72, SD = 0.85), and online courses (M = 3.19, SD = 0.91). The learners also think that online learning is more personalized (M = 4.12, SD = 0.85) in YouTube channels, giving sense of human touch, (M = 4.09, SD = 0.92) through online courses, and easy to use (M = 3.98, SD = 0.82) in web search. The details are demonstrated in Figure 3.

Figure 3.

E-learning by interaction characteristics of learning channels.

What is learnt from the above finding is that with the full potential of AI integration and learning capability of chatbots, MOOCs may benefit from this technology to motivate learners to switch from traditional to modern learning channel.

2.1 Quantitative study

A quantitative method was used for the data collection procedure due to the extent that online survey fosters the geographical distribution of the questionnaire with relatively less cost and higher time efficiency [42].

This study used AMOS v.24 software for the statistical analysis of the proposed hypotheses and moderation effect. Initially, measurement model was assessed, followed by structural model, as suggested by Hair et al. [43]. A 5-point Likert scale ranging from 1 being “strongly disagree” to 5 being “strongly agree” is used.

2.2 Sampling process and collection of data

This study was conducted in the Middle East during the period 2021–2022. The data was collected from the learners who mostly prefer using digital channels to access to learning materials and conduct learning activities. The subjects of the study were distributed the online questionnaire with the use of emails and official pages of MOOCs on social media (e.g., Facebook, Twitter, and others).

Prior to the final survey, a pre-test was carried out with 14 participants, who had experience with chatbot in other platforms. It helped us to refine the questionnaire in terms of clarity and language. Because, questions were firstly translated into Arabic and then back to English language to ensure that the meanings were not lost. The items representing CE, TI, and ENJ were re-worded in order for increasing the generalizability of the findings. Cronbach’s alpha (α) was used to test reliability of the pre-test results, and it was found that α values of the study variables were higher than 0.7, indicating the adequate reliability [44]. The final questionnaire (Appendix) comprised of 42 questions, 6 of them being related to demographics.

Data was analyzed using AMOS v.24 software with structural equation modeling (SEM) technique. Firstly, measurement model was assessed, which was followed by structural model of testing the proposed hypotheses [43]. The overall analysis included construct validity including convergent and discriminant validity, and model testing including model fit and hypotheses.

3. Data analysis

3.1 Demographics of respondents

The demographic analysis showed that majority of respondents are male (58.4%), while they belong to 26–30 years old age group (35.9%) and with undergraduate degree (49.6%). In the context of Internet usage, they reported that more than half of them use Internet for above 21 hours in a weekly basis, with mobile devices (52.3%). However, nearly 60% of them are unfamiliar with chatbot use in the context of online activities, not limited to learning process (see Table 1).

Demographics (N = 373)		Frequency	Percentage (%)
Gender	Male	218	58.4
	Female	155	41.6
Age	<25	116	31.1
	26-30	134	35.9
	31 - 35	67	18.0
	36 - 40	39	10.5
	> 41	17	4.6
Education	Undergraduate	185	49.6
	Master	109	29.2
	PhD	79	21.2
Frequency of Internet use/weekly	> 5 hours	11	2.9
	5 - 10 hours	17	4.6
	11 - 15 hours	39	10.5
	16 - 20 hours	117	31.4
	> 21 hours	189	50.7
Device type for Internet use	Desktop PC	38	10.2
	Laptop	56	15.0
	Tablet	84	22.5
	Mobile phone	195	52.3
Familiarity with chatbot use	Fully familiar	43	11.5
	Somehow familiar	107	28.7
	Not familiar at all	223	59.8

Table 1.

Demographic profile of study respondents.

3.2 Measurement model

The constructs and their underlying items were assessed for reliability and validity. Reliability is to measuring the consistency between multiple measurements of a construct [45], which must be carried out before validity test. Reliability was assessed with the use of Cronbach’s α, which is a commonly applied measure [44]. There are four cut-off points that define the reliability level of the constructs, recommended by Hinton et al. [46]:

• Excellent reliability >0.90.

• High reliability 0.70 ∼ 0.90.

• Moderate reliability 0.50 ∼ 0.70.

• Low reliability <0.50.

In this study, four constructs, namely UV (α - 0.79), INFO (α – 0.76), TI (α – 0.71), and CS (α – 0.73) show high reliability, while other constructs showed moderate reliability.

In the next stage, convergent validity is assessed. There are three measures that utilize the convergent validity test: (1) Confirmatory factor analysis (CFA) for measuring scale validity with indicator factor loadings. The indicator loadings must exceed threshold of 0.5 [47]; (2) Composite reliability (CR) that must be over 0.6 acceptance level [48]; and (3) Average variance extracted (AVE), which must be higher than 0.5 [49]. Table 2 shows that all criteria regarding the convergent validity test are met. CR values ranged between 0.71 and 0.86, while AVE values ranged between 0.54 and 0.69. In CFA analysis, IRR3, ENJ2, and TRS1 items did not load on their underlying constructs and therefore excluded.

Constructs & items	Mean	SD	Standardized factor loadings	Cr α	CR	AVE
Task value
Intrinsic or interest value (IV)				0.65	0.82	0.62
IV1	3.07	0.81	0.72	3.07	0.81	0.72
IV2	3.12	0.82	0.75	3.12	0.82	0.75
IV3	2.94	1.13	0.79	2.94	1.13	0.79
IV3	3.14	1.18	0.77	3.14	1.18	0.77
Attainment value (AV)				0.63	0.84	0.69
AV1	3.26	0.86	0.83	3.26	0.86	0.83
AV2	2.87	0.79	0.82	2.87	0.79	0.82
AV3	3.11	1.04	0.91	3.11	1.04	0.91
Utility value (UV)				0.79	0.71	0.55
UV1	3.87	1.07	0.79	3.87	1.07	0.79
UV2	3.74	1.24	0.71	3.74	1.24	0.71
UV3	3.42	0.76	0.70	3.42	0.76	0.70
Informativeness (INFO)				0.76	0.75	0.56
INFO1	3.14	1.28	0.71	3.14	1.28	0.71
INFO2	3.08	1.13	0.74	3.08	1.13	0.74
INFO3	2.95	0.79	0.78	2.95	0.79	0.78
Enjoyment (ENJ)				0.71	0.72	0.64
ENJ1	3.56	0.84	0.82	3.56	0.84	0.82
ENJ2	3.27	0.88	0.79	3.27	0.88	0.79
Transparency (TRS)				0.69	0.73	0.57
TRS1	3.11	0.83	0.75	3.11	0.83	0.75
TRS2	3.06	1.05	0.73	3.06	1.05	0.73
Credibility (CRE)				0.76	0.82	0.56
CRE1	2.85	1.13	0.77	2.85	1.13	0.77
CRE2	2.91	1.27	0.71	2.91	1.27	0.71
CRE3	3.04	1.08	0.82	3.04	1.08	0.82
Trust of Internet (TI)				0.71	0.86	0.68
TI1	3.17	0.74	0.80	3.17	0.74	0.80
TI2	3.26	0.81	0.83	3.26	0.81	0.83
TI3	3.10	0.89	0.91	3.10	0.89	0.91
Irreplaceability (IRR)				0.68	0.82	0.69
IRR1	3.28	0.83	0.86	3.28	0.83	0.86
IRR2	3.16	0.77	0.79	3.16	0.77	0.79
Personalization (PER)				0.66	0.81	0.66
PER1	3.47	0.92	0.83	3.47	0.92	0.83
PER2	3.21	0.96	0.85	3.21	0.96	0.85
PER3	3.18	0.84	0.87	3.18	0.84	0.87
Communication style alignment (CS)				0.73	0.81	0.65
CS1	2.78	0.85	0.82	2.78	0.85	0.82
CS2	2.69	1.06	0.79	2.69	1.06	0.79
CS3	2.81	1.17	0.78
Intention to use e-government service (INT)				0.64	0.75	0.57
INT1	3.29	3.29	3.29
INT2	3.17	3.17	3.17
INT3	3.06	1.03	0.77
Chatbot type (CT)				0.67	0.71	0.54
CT1	3.17	0.85	0.72
CT2	3.39	0.83	0.74

Table 2.

Measurement model results.

Finally, discriminant validity test is used. According to Fornell and Larcker [49], square root of AVEs for each construct must be greater than correlation coefficients between the constructs. Table 3 indicates that this criterion is also met.

	1	2	3	4	5	6	7	8	9	10	11	12	13
1	0.77
2	0.32	0.85
3	0.28	0.53	0.73
4	0.17	0.31	0.41	0.83
5	0.23	0.46	0.58	0.56	0.74
6	0.21	–0.04	0.06	0.23	0.05	0.81
7	0.39	0.11	0.02	0.08	0.15	0.43	0.74
8	0.25	0.31	0.12	0.17	0.68	0.09	0.56	0.77
9	0.01	0.48	0.03	–0.23	0.47	0.16	0.34	0.29	0.85
10	0.42	0.62	0.08	0.56	0.31	0.09	0.38	0.17	0.27	0.82
11	0.29	0.49	0.65	–0.07	0.36	0.62	0.26	0.26	0.47	0.17	0.85
12	0.31	0.57	0.49	0.19	0.26	0.43	0.22	0.09	0.71	0.22	0.76	0.73
13	0.49	0.04	0.25	0.09	0.04	0.06	–0.08	0.11	0.51	0.35	0.55	0.39	0.14

Table 3.

Discriminant validity analysis.

3.3 Structural model

Primarily, the structural model between CS and task-related and channel-related factors was tested. CS was found to be positively and significantly related to TV (β = 0.314***, p < 0.001), INFO (β = 0.302***, p < 0.001), ENJ (β = 0.217**, p < 0.01), TI (β = 0.179**, p < 0.01), IRR (β = 0.114*, p < 0.05), and PER (β = 0.356***, p < 0.001), whereas it does not influence perception of transparency (TRS) (β = 0.041, p > 0.05). It may be explained to the extent that transparency is rooted in learners’ prior experience with e-learning service providers. Even though they are introduced with new communication channel, they may still have concerns on trust and transparency, while they find chatbot communication informative, enjoyable, and irreplaceable, and so on. Hence, H1a, H1b, H2a, H2c, H2d, and H2f are confirmed, whereas H2b is rejected.

Next, task-related and channel-related factors predicted INT. It was found that both task-related factors, namely, TV (β = 0.289***, p < 0.001) and INFO (β = 0.252**, p < 0.01) positively and significantly predict learners’ intention (INT) to use chatbot-supported MOOCs. In the context of channel-related factors, ENJ (β = 0.245***, p < 0.001), IRR (β = 0.126*, p < 0.05), and PER (β = 0.276***, p < 0.001) positively and significantly influence INT, whereas TRS (β = −0.057, p > 0.05) and TI (β = 0.021, p > 0.05) are not related to INT. Moreover, H3a, H3b, H4a, H4e, and H4f are supported, while H4b, H4c, and H4d are not supported.

3.4 Moderating impacts

As moderators, CT was included to test whether it strengthens or weakens the main effects. CT was added in line with the assumption that text-based chatbot type may significantly enhance learners’ motivation to use chatbot-supported communication channel, which may also increase learners’ perceptions of e-learning service features that could ultimately lead to switching from traditional to digital communication channel. In the context of task-related factors, it was found that CT plays a strong moderating role between TV and INT (β = 0.297***, p < 0.001), and between INFO and INT (β = 0.328***, p < 0.001). Hence, H5a and H5b are supported. In terms of channel-related factors, CT strengthens the effects of ENJ (β = 0.257***, p < 0.001), IRR (β = 0.233**, p < 0.01), and PER (β = 0.387***, p < 0.001) on INT. Thus, H6a and H6d are confirmed.

4. Discussion

Irrespective the popularity of e-learning service delivery in many countries in the context of MOOCs, majority of learners still prefer using traditional channels to request learning materials either by web search or YouTube learning. Understanding the reason why digital divide exists for e-learning services, but not for other types of Internet use, is very important research problem.

Chatbot-supported communication between companies and their clients are becoming a norm of digital communication and service provision. Consequently, surrounding the core themes of “similarity-attraction” theory, the recent study explored the value of chatbot-supported communication in e-government domain and tested the effect of chatbot-learner communication style alignment and similarity on the realization of the value of such communication type.

Through the open-ended online survey, the most important aspects of e-learning service delivery to learners were identified. If properly designed and integrated with the features of personalized content, credibility, and human touch, chatbot-supported communication channel can potential draw the attention of learners in their interaction with MOOCs. In the next stage, quantitative study was conducted to comprehensively address the prevalent aspects of learners’ perceptions toward adoption of chatbot supported e-learning services and the impact of learner-chatbot communication style similarity on those aspects. It was found that communication style similarity between chatbot and learner leads to learners’ perceptions of this communication type being more informative, enjoyable, credible, and irreplaceable. In addition, this type of communication increases the valuableness of the information-seeking task, trust in Internet, personalization, while also makes learners feel anxious about using this technology. Moreover, it does not change the learners’ perceptions on transparency and trust. This is very critical finding that must be addressed by the system designers. Making the interface and content simple and accessible may reduce the anxiety. However, transparency and trust are purely service provider-related problem. If AI functions are fully deployed in designing this type of communication, it can reduce learners’ perceptions on manipulation of information, especially by telling them that chatbot can “learn and improve itself by communicating.” It was also found that task value, informativeness, enjoyment, credibility, irreplaceability, and personalization lead to learners’ intention to use such communication channel.

Chatbot type was added as moderators to see whether or not they strengthen learners’ perceptions toward using chatbot-supported communication. Similar to human communication, people want to interact with computers in their own language [14]. It was postulated that in order to facilitate human-computer interaction, users must be allowed express their interests, wishes, or requests directly and naturally, by speaking, typing, and pointing [50]. It was found that chatbot type being text stream moderates the relationship of task value, informativeness, enjoyment, irreplaceability, and personalization with learners’ intention to use such communication channel. Overall, the results clearly indicate what would be the possible outcomes if communication style between chatbot and learners is aligned well.

4.1 Implications

In theory, this study contributes to the improvement of HCI for learner-chatbot interaction, especially with an understanding on the impact of similarity-attraction theory on learner perceptions and intention to use chatbot-supported e-learning services. The alignment of communication style of chatbots with learners’ needs may lead to sense of “in-group” effect that could ultimately create various benefits for both e-learning service providers and service receivers.

Despite the social and economic potential of AI in public and private sectors, the major concern is the possible negative effects of AI. In economic perspective, it can be explained by the employment effects of AI, as robots are threat to labor market. Special Eurobarometer [51] revealed that 74% of citizens are worried that more jobs will be taken by robots and AI. However, another study in developed countries has revealed that technology replaces low-skill jobs mainly, which enhances productivity [52]. In addition, the real business and public service cases also show that the utilization of AI and especially chatbots contribute to organizations as well as clients. This study identified some patterns of communication between learners and MOOCs providers that can guide designers and decision-makers. In addition, offering chatbot-supported communication channel revealed that learners find it enjoyable and comfortable to engage with, while also efficient in terms of information retrieval time. However, traditional channel is still believed to be effective in being personalized, credible, and has a sense of human touch. One reason can be that they were only provided with relatively simple chatbot without a language style they are happy with. Well-known Chinese e-retailers such as Taobao.com and JD.com use trendy Internet slang, such as “Qin,” meaning “my sweet heart” when they begin conversation with customers, which increases the customer stickiness to e-retailer’s service [23].

5. Conclusion

The current study examined e-learning in Arabic context, with an introduction of chatbot-supported communication channel as well as its comparison with existing channels. Initially, the gaps in government service delivery were addressed and then the potential for chatbot-supported communication between learners and service providers was discussed. In addition, the factors that may significantly affect learners’ channel-choice were assessed to see whether communication style similarity between learners and chatbots would lead them to use this communication channel. Rather than being divider, this study mainly targets the advancement of communication process between learners and MOOC providers by incorporating intelligent, timely, fun, personalized, and efficient features of communication tools. As AI continues to cover wider scope of services, its application in chatbot-supported communication in e-learning setting can be highly beneficial for both learners and service providers. From MOOC providers, it can have a considerable implication for staff and costs [53]. Cross-channel integration (or integration of the beneficial features of all communication channels in one platform) can lead to response consistency, timeliness, access to the same data, and implementation of transaction in a single platform and without conventional ways of learning process.

One of the limitations of this research is that it has been conducted in a single cultural setting. However, extending this study to cross-cultural setting would be highly beneficial for a higher validation of communication-style alignment. As communication style similarity between learners and chatbot is a central focus, it might be different in every country. Moreover, drawing from cultural differences, learners may have different expectations from interacting with chatbots in terms of their functionalities, fun features, service query types and others. In addition, only text-based chatbot has been developed and tested in this study. It could be worthy to add visual features such as avatars, facial elements to chatbot, and test learners’ preferences with the use of Anthropomorphism scale.

Construct/item	Description	Source
Task value
Intrinsic or interest value (IV)		Hagemeier and Murawski [54]
IV1	I enjoy advancing my information base by exploring E-learning chatbot
IV2	The challenge of using E-learning chatbot for service request is exciting
IV3	I liked the challenge of retriving learning materials through chatbot
Attainment value (AV)		Hagemeier and Murawski [54]
AV1	I value the prestige that come with using E-learning chatbot for learning
AV2	I need the E-learning chatbot to fulfill my information retrival potential
AV3	Completing information retrival through E-learning chatbot allowed me to attain high sense of self-worth
Utility value (UV)		Hagemeier and Murawski [54]
UV1	I think chatbot can be an integral part of what I want to do in future in terms of learning service
UV2	Chatbot usage is important because it provides me better funcionality
UV3	I think chatbot can be beneficial because of my need for personalized service
Cost of engaging (CE)		Hagemeier and Murawski [54]
CE1	I worry that spending time for chatbot usage would take me away from other activities
CE2	Prior to using E-learning chatbot, I was concerned that using this channel would not be worth of effort
CE3	I worried that I would waste a lot of time using chatbot for retriving the specific information
Informativeness (INFO)		[23]
INFO1	with the use of E-learning chatbot, I feel well-informed about the service I request
INFO2	with the use of E-learning chatbot, I know how to request the information based on my needs
INFO3	with the use of E-learning chatbot, I can retrieve more information from linked sources
Enjoyment (ENJ)		[23]
ENJ1	I find interacting with E-learning chatbot enjoyable
ENJ2	I find interacting with E-learning chatbot interesting
ENJ3	I find interacting with E-learning chatbot exciting
Transparency (TRNS)		[23]
TRNS1	The advice given by E-learning chatbot is easy to understand
TRNS2	The advice given by E-learning chatbot is not confusing
TRNS3	I understand about what E-learning chatbot talks about
Trust of Internet (TI)		Lallmahomed et al. [55]
TI1	Internet makes me feel comfortable using it to interact with learning services
TI2	I feel assured that legal and technological structures adequately protect me from problems on Internet
TI3	Internet is a robust and safe environment to transact using learning services
Irreplaceabilily (IRR)		Zhang et al. [56, 57]
IRR1	I think chatbots are superior to conventional channels in learning service delivery
IRR2	I think conventional channels cannot complete some functional tasks as chatbots do
IRR3	I think there are some functional differences between chatbots and conventional channels
Personalization (PER)		Kim and Han [58]
PER1	I feel that E-learning chatbot is tailored to me
PER2	I feel that contents in E-learning chatbot are personalized
PER3	I feel that E-learning chatbot is personalized for my usage
Communication style alignment (CS)		[23]
CS1	I feel the language style of E-learning chatbot matches my preferences
CS2	I feel the design of E-learning chatbot matches my communication preferences
CS3	I feel E-learning chatbot communication with me in a style that I like
Intention to use e-learning service (INT)	Lallmahomed et al. [55]
INT1	I intend to use chatbot-supported learning services in the future
INT2	I predict I would use chatbot-supported learning services in the future
INT3	I plan to use chatbot-supported learning services in the future
Chatbot type (CT)		[14]
Text stream	I consider the text-based E-learning chatbot:
	(1) Extremely weird; (2) Highly weird; (3) Moderately weird; (4) Less weird; (5) Not weird at all