Enhancing E-Business Communication with a Hybrid Rule-Based and Extractive-Based Chatbot

Abstract

E-businesses often face challenges related to customer service and communication, leading to increased dissatisfaction among customers and potential damage to the brand. To address these challenges, data-driven and AI-based approaches have emerged, including predictive analytics for optimizing customer interactions and chatbots powered by AI and NLP technologies. This study focuses on developing a hybrid rule-based and extractive-based chatbot for e-business, which can handle both routine and complex inquiries, ensuring quick and accurate responses to improve communication problems. The rule-based QA method used in the chatbot demonstrated high precision and accuracy in providing answers to user queries. The rule-based approach achieved impressive 98% accuracy and 97% precision rates among 1684 queries. The extractive-based approach received positive feedback, with 91% of users rating it as “good” or “excellent” and an average user satisfaction score of 4.38. General user satisfaction was notably high, with an average Likert score of 4.29, and 54% of participants gave the highest score of 5. Communication time was significantly improved, as the chatbot reduced average response times to 41 s, compared to the previous 20-min average for inquiries.

1. Introduction

Electronic business (e-business) has become a cornerstone of modern commerce, enabling companies to streamline operations, reach global markets, and enhance customer engagement through digital platforms. As businesses increasingly shift toward online operations, the need for efficient communication and seamless customer service has grown substantially. E-businesses must address these needs to ensure customer satisfaction and foster brand loyalty [1]. However, maintaining effective communication can be challenging with the high volume of interactions and transactions taking place online. Inefficiencies in handling customer inquiries can lead to dissatisfaction, ultimately impacting a business’s reputation and bottom line [2].

These communication challenges are not limited to e-businesses but also affect other sectors, such as healthcare. Failure to answer inquiries to obtain basic information turns into complaints from patients and their relatives and damages the brand value of the hospital. While 72% of customers share a positive experience with six or more people, only 13% of dissatisfied customers share their experience with 15 or more people [3]. In other words, 100 satisfied customers share their satisfaction with at least eight people, while 100 dissatisfied customers share their dissatisfaction with at least 115 people. In addition, a very small portion of dissatisfied customers (between 4–9%) voice their complaints. In other words, it can be assumed that there is at least 25 times more dissatisfaction than the number of complaints received. As a result, if there are 100 complaints, it can be said that dissatisfaction is shared with 2875 ($115\times 25$) people, considering the presence of people who do not express complaints. A current situation analysis shows that a significant proportion of the complaints are communication-based.

Like in many other sectors, chatbots have emerged as a powerful tool in the e-business sector, offering numerous benefits in enhancing customer service and operational efficiency [4]. They facilitate efficient communication by responding instantly to customer queries, assisting with order processing, offering product recommendations, and resolving issues without human intervention. This capability is valuable in handling high volumes of routine inquiries, thus freeing up human resources for more complex tasks [5]. Chatbots contribute to higher customer satisfaction and better business outcomes by improving the speed and accuracy of customer service.

Integrating chatbots into e-business platforms offers several advantages [6,7]. Firstly, they enhance customer satisfaction by offering immediate and accurate responses to frequently asked questions, such as inquiries about product details, shipping information, and return policies. Secondly, chatbots can reduce operational costs by automating routine tasks, allowing businesses to allocate human resources more effectively. Additionally, chatbots can gather and analyze customer data, providing valuable insights for personalized marketing and improved customer relationship management. Overall, the use of chatbots in e-business not only enhances customer experiences but also drives operational efficiencies and business growth [8].

This study aims to develop a hybrid rule-based and extractive-based chatbot to address communication challenges in e-business. The motivation for using both a rule-based approach and an extractive-based approach in e-business stems from the need to efficiently manage a diverse range of customer inquiries. Rule-based chatbots operate on predefined rules and patterns, making them suitable for handling straightforward inquiries by matching user queries with a predefined database of keywords and responses [9]. They effectively address common and routine questions where answers are well defined [10]. For instance, they can provide information about product availability, order status, return policies, or general support queries, which results in reduced workloads for human customer service representatives, allowing them to focus on more complex tasks. Conversely, extractive-based chatbots use a more data-driven approach. They analyze a vast corpus of text data to understand language patterns and context [11]. Instead of relying solely on predefined rules, they extract relevant information from existing texts or documents. In e-business, these chatbots can respond based on customer reviews, product descriptions, detailed policy documents, or details about payments/cancellations. They excel in handling complex queries that require access to a large amount of information and can provide more contextually relevant responses.

The proposed chatbot system in this study leverages a rule-based approach for managing common customer inquiries efficiently and an extractive-based approach for handling more complex questions. By combining these two approaches, the chatbot can provide a comprehensive range of responses, ensuring both quick and accurate answers to routine questions and detailed, contextually relevant information for more complicated inquiries. This dual approach aims to enhance customer communication in e-business, ultimately improving customer satisfaction and operational efficiency.

The remainder of this paper is structured as follows. Section 2 discusses the related works to present the scientific gap. Section 3 explains the proposed methodology. Section 4 gives the evaluation criteria. Section 5 shares the experimental results using the methodology. Finally, Section 6 discusses the results and future directions and concludes this paper.

2. Literature Review

Service quality has become a focal point for e-business establishments, whether public or private, across the globe. Fatima et al. [12] furnished a precise interpretation of service quality, characterizing it as “the extent and orientation of a disparity between consumers’ perceptions and anticipations.” This quality assessment stems from customers’ judgments formed post the service delivery procedure, where they contrast their initial expectations with their actual observations of the rendered services. When expectations surpass perceptions, customers interpret the service quality as poor and unsatisfactory. The critical reasons for dissatisfaction with the e-commerce shopping experience include long waiting times, service delays, non-availability of products/services, and communication problems [13,14]. Artificial intelligence-based tools have the potential to solve these dissatisfaction issues [15,16]. For the long waiting times and delays in service, artificial intelligence (AI) has the ability to optimize care processes, maximize workforce capacity, and reduce waste and costs [17,18]. It can also play a role in customer feedback analysis and personalized marketing, leading to more accurate and efficient business practices for addressing the problem of product availability. [19].

In e-business, the issue of insufficient staff resources often leads to an overwhelming demand for information. Consequently, the challenges in obtaining information exacerbate dissatisfaction and complaints [20,21]. Combining these factors results in a complex issue that hinders the overall quality of e-business services. Due to staff shortages, customers often struggle to access the information they need, leading to increased frustration and dissatisfaction with e-business platforms. This issue underscores the importance of addressing not only the communication barriers but also the underlying staffing challenges to improve the overall customer experience in e-business platforms [22,23].

2.1. Studies on Using AI to Improving Communication

Companies frequently grapple with staffing challenges, which, in turn, lead to an overwhelming demand for information from customers [24]. This heightened demand often results in communication barriers, making it difficult for individuals to obtain the information they need, ultimately leading to increased dissatisfaction and complaints. To address these challenges, various data-based and AI-driven approaches have emerged to help companies improve communication, streamline information dissemination, and ultimately alleviate the strain on their personnel [25,26].

Predictive analytics for communication optimization have been employed to forecast customer communication needs [27]. These models analyze customer data, purchase histories, and other relevant variables to predict the timing and nature of customer inquiries. E-businesses can allocate resources more efficiently to meet customer communication demands, thereby reducing response times and improving satisfaction. While there is extensive research on communication phenomena, predictive analytics in this field is still in its early stages [28,29].

Virtual assistants for communication, powered by AI and natural language processing (NLP), are emerging as valuable tools, especially for communication in e-business environments. These multi-functional AI entities can perform various tasks, including processing orders, personalizing product recommendations, and providing customer support information [30]. They offer customized interactions and can adapt to user preferences, effectively enhancing communication between customers and service providers. Virtual assistants like Apple’s Siri, Amazon’s Alexa, and Google Assistant have the potential to assist users with business-related inquiries and support more natural and interactive communication [31]. Their ability to handle voice commands and remember user preferences adds another dimension to improving communication in e-business settings.

Chatbots are pivotal in improving communication skills across different domains, offering valuable support to learners and individuals seeking to enhance their abilities in various contexts [32]. As AI technology advances, chatbots hold promise for addressing communication challenges and promoting effective interaction in an ever-evolving digital world [33]. While both virtual assistants and chatbots employ AI and NLP technologies for user interaction, they differ in their scope, functionality, and level of personalization. Virtual assistants are more versatile and offer a more comprehensive range of functions, while chatbots are typically designed for specific tasks or industries and may have more limited capabilities. Because of this, Section 2.2 investigates various chatbot studies, specifically in the e-business domain.

2.2. Studies on Chatbots in e-Business

Studies aimed at enhancing communication and increasing satisfaction through chatbot development in e-business have gained prominence. These endeavors recognize the potential of chatbots as powerful tools to strengthen interactions between customers and service providers and improve overall business experiences [32].

Studies on chatbots in e-business have underscored their transformative potential across various domains, from customer service and marketing to sales and user engagement. Research indicates that chatbots enhance customer interaction by providing instant responses, thereby improving customer satisfaction and loyalty [34,35,36]. They streamline transactions by guiding users through purchase processes, resolving queries promptly, and offering personalized recommendations based on user preferences and behavior analysis [33,37].

Some recent studies on AI-based chatbots in e-business are related to technology acceptance. Li and Wang [38] investigated the effect of chatbot language style on customers’ continuance usage intention and attitude toward the brand. Fu et al. [39] analyzed the role of human-like characteristics, such as social presence and empathy, of chatbots for online shopping. Shahzad et al. [40] assessed the effect of chatbot service quality on customers’ brand loyalty. They found that chatbot user experience, trust, and electronic word of mouth mediate the relationship between chatbot service quality and brand loyalty among Chinese luxury fashion brand customers. Ahmed et al. [41] proposed a medical chatbot assistance using machine learning techniques. Patients can send clinical questions from mobile or personal systems. According to the experimental results, the chatbot offers timely first-aid medical information and reduces the necessity for prompt consultations with a physician. Auer et al. [42] conducted a survey to investigate behavioral intentions regarding the use of chatbots for airport consumer inquiries about services. Bhattacharyya [43] conducted a qualitative survey and found fifteen factors affecting the adoption of chatbots developed with AI-driven natural LLM. These chatbots are used for customer–firm interactions. Hanji et al. [44] investigated the adoption of AI-enabled chatbots in tourism and travel services.

Furthermore, some studies have highlighted the efficiency gains achieved by chatbots, reducing operational costs through automation while maintaining service quality [45,46,47]. Chatbots also contribute to data collection and analysis, enabling businesses to glean valuable insights into customer behavior and market trends. By integrating artificial intelligence and natural language processing, chatbots continuously evolve in sophistication and are capable of handling complex inquiries and adapting to diverse user contexts [48,49].

The literature identifies certain challenges, including ensuring chatbot accuracy, effectively handling nuanced user queries, and maintaining user trust and privacy [8].

2.3. Main Contributions

In the current research, several limitations and gaps emphasize the necessity and importance of this study. First, many existing studies on chatbots in e-business focus primarily on either rule-based or extractive-based systems but rarely integrate both approaches. This separation can result in chatbots that are either overly rigid or insufficiently accurate in handling complex inquiries. Furthermore, there is limited research on chatbots’ effectiveness in handling routine and complex customer service interactions within a single framework, as most studies tend to specialize in one area. Additionally, while numerous studies highlight the potential of AI and NLP technologies, there is a lack of comprehensive evaluations that measure both user satisfaction and operational efficiency improvements in a real-world e-business context. This study addresses these gaps by developing and evaluating a hybrid chatbot system that combines rule-based and extractive-based methods, providing a more versatile and effective solution for e-business communication challenges. This paper endeavors to contribute to advancing business communication and customer satisfaction by developing a novel chatbot system. Drawing upon a combination of AI methodologies, including NLP and machine learning, our chatbot aims to facilitate seamless interactions between customers and service providers, provide tailored business information, and reduce administrative burdens. This study aspires to bridge the communication gaps in e-business platforms by addressing the diverse needs of customers and business providers, furthering the body of knowledge in this critical area.

3. Proposed Methodology

The chatbot was developed by integrating NLP, decision trees, and generative pre-trained transformers, as illustrated in Figure 1. The chatbot’s operation involves four main modules: preprocessing, input processing, modeling, and output. In the preprocessing module, a vector database is created.

First, the text of the relevant topic is divided into small pieces, called chunks. Then, each chunk is encoded and stored in a database. At the end of the semantic search, the results are ranked according to their similarities. These results represent the most suitable chunks for generating an answer to the query. In the input processing module, the query received from the user is encoded into numeric vectors through embedding. NLP plays a crucial role in segmenting the query into tokens such as sentences, phrases, and words. This encoded vector is used for the semantic search in the vector database. The modeling module determines the answer generation model. If the query is related to topics for which the answer has already been generated, the decision tree model is preferred; otherwise, the generative pre-trained transformer model is used. According to the chosen model, the output module provides the answer to the user.

A rule-based chatbot is a type of conversational AI system that operates on predefined rules and patterns. Rule-based chatbots follow a set of predetermined instructions to respond to user inputs. These instructions are often created by human developers and are designed to cover specific scenarios or questions. This study develops a rule-based chatbot that utilizes a decision tree to navigate through a sequence of decisions and rules. The decision tree aspect of the chatbot pertains to the logical structure it employs for decision making. This method involves breaking down complex questions into a series of binary choices, leading to specific answers based on predefined rules and conditions.

Algorithm 1 presents the pseudo-code of the rule-based method using a decision tree for the chatbot. When a user interacts with the chatbot, it starts at the root of the decision tree and follows a path determined by the user’s input and the chatbot’s rules. At each decision point, the chatbot evaluates the user input and determines the most appropriate branch to follow. The path followed eventually leads to a node corresponding to a specific response that the chatbot generates.

Algorithm 1: Rule-based method for the chatbot using a decision tree.

The rule-based component of the chatbot is designed to handle routine and predictable customer queries through predefined rules and patterns. This approach ensures quick and precise responses to frequently asked questions (FAQs) and standard procedures. The development process includes several key steps, such as regular expressions, decision trees, and rule management systems. The rule-based system utilizes pattern-matching techniques and regular expressions to identify keywords and phrases within user inputs. This allows the system to match queries with the appropriate responses stored in the chatbot’s database. Decision trees are employed to handle more complex rule-based scenarios. Based on user input, these trees guide the chatbot through a series of decision points, ultimately leading to the correct response. An efficient rule management system is implemented to organize and update the rules dynamically. This system ensures that the chatbot remains up to date with the latest information and can adapt to changing business needs.

The extractive-based component addresses the need for handling more complex and less predictable queries that the rule-based system cannot manage effectively. This approach involves several steps, such as NLP, text classification, and extractive summarization. The chatbot employs various NLP techniques, including tokenization, part-of-speech tagging, named entity recognition, and dependency parsing, to analyze and understand user inputs at a deeper level. Advanced machine learning models, such as Support Vector Machines (SVMs) and Convolutional Neural Networks (CNNs), are used for text classification tasks. These models categorize user queries into predefined classes, enabling the chatbot to determine the appropriate response strategy. Algorithms such as TextRank and BERT-based extractive summarization extract relevant information from longer and more complex inputs. These algorithms identify key sentences and phrases that directly address the user’s query.

The generative pre-trained transformer model, often called “transformers,” represents a revolutionary advancement in NLP and has played a pivotal role in developing competent conversational AI systems, including the extractive-based component. This model is built upon a deep learning architecture known as the transformer architecture, which has proven exceptionally adept at understanding and generating human language. This study uses ChatGPT by OpenAI, based on the GPT-4 (generative pre-trained transformer) architecture. GPT-4 is a type of transformer model that has been pre-trained on a diverse range of internet text and fine-tuned for specific tasks, including natural language understanding and generation. Algorithm 2 presents the pseudo-code of the extractive QA method.

The provided algorithm outlines the extractive question-answering (QA) method for the chatbot. It begins by gathering and preprocessing a corpus of text documents and removing punctuation and stop words. The text is then segmented into chunks and converted into embedding, creating a vector store. When a user question is input, it is also converted into an embedding. The algorithm calculates the similarity between the user’s embedded question and the vectors in the vector store, ranks the similarities, and selects the highest-ranked chunk as the context. Finally, it generates a response using an underlying language model based on the user’s question and the selected context.

Algorithm 2: Extractive QA method for the chatbot.

Imagine a customer service representative with a binder of standard responses for common questions. When a customer asks a question that matches one of the scenarios in the binder, the representative pulls out the pre-written response and delivers it. This method efficiently handles predictable questions but does not adapt to query variations. For an e-business communication chatbot, suppose a user types, “What are your store hours?” The chatbot is programmed with a rule that matches this query to a predefined response: “Our store hours are Monday to Friday, 9 AM to 6 PM.”

This approach is like using a search engine to find specific information within a large set of documents. When you enter a query, the search engine scans through its indexed content and presents you with the most relevant excerpts or documents containing the information you seek. For an e-business communication chatbot, if a user asks, “Can you tell me about the available shipping methods?”, the chatbot uses NLP to understand the query and searches the knowledge base or FAQ section for relevant information. Response: “We offer the following shipping methods: Standard Shipping (5–7 business days), Expedited Shipping (2–3 business days), and Overnight Shipping (1 business day). You can choose your preferred method during checkout.”

4. Evaluation

The evaluation of the chatbot’s performance involved two main aspects: assessing chat performance and measuring user satisfaction. The chatbot’s performance was evaluated from these two perspectives because the generative-based QA method produces new answers, whereas the rule-based QA method uses pre-defined answers.

For the generative-based QA method, an evaluation was conducted through human assessment. Hospital staff members, including administrators and professionals, were asked to compare the responses generated by the model using a Likert scale assessment. They assessed the responses’ accuracy, comprehensibility, and overall quality to determine how well the model performed in providing satisfactory answers. Accuracy cannot be directly calculated for generative-based QA models because these models produce open-ended text responses rather than making discrete, categorical classifications. Therefore, only precision—showing how accurately positive responses were predicted among all responses predicted as positive—was measured.

For the rule-based QA method, a confusion matrix was created to assess the model’s performance by comparing the actual answers with the model’s answers.

Feedback was collected from patients or their relatives to evaluate user satisfaction after deploying the chatbot. A survey was administered to gather information on satisfaction levels, comparing the use of the chatbot with previous communication methods. The survey included questions such as how the users were satisfied with the chatbot (using the Likert scale) and how much time they previously spent on communication on average.

5. Case Study

5.1. Basic Information and Problem Definition

One of the challenges for an e-business organization is managing customer information demand. This request for information can be about products/services, payments/cancellations, or prices. Due to insufficient staffing, only two people are assigned to take calls. Currently, the company receives an average of eight calls per minute, and the average call duration is 2 min. Accordingly, the company has missed 14 calls by only employing two staff members.

A current situation analysis was performed to understand the problem. The 804 comments received in the current situation analysis were evaluated in four categories: “website performance”, “product/service quality”, “customer service (communication)”, and “payment”. Of these, 454 comments were evaluated for information purposes.

Table 1. Results of current situation analysis.

	Positive	Negative	Positive Ratio	Negative Ratio
Website Performance	28	80	26	74
Product/Service Quality	101	125	45	55
Customer Service	0	67	0	100
Payment	45	8	85	15

divides these comments into positive and negative categories. Looking at the ratio of positive comments to total comments, it is evident that there are no positive comments regarding “customer service”.

This chatbot aims to provide answers to customers for questions that cause communication problems within the company. The answers generated do not contain critical information about product/service components, customer privacy, or personnel data.

5.2. Practical Examples

The use case revolves around implementing a hybrid chatbot system in an e-business company that specializes in selling electronics and computer hardware. It aims to enhance its customer service operations by integrating a chatbot capable of handling various customer inquiries. The goal is to improve response times, increase customer satisfaction, and reduce the workloads of human customer service agents.

Figure 2 depicts a practical example of the rule-based QA model. A customer, Defne, is using an e-commerce website’s chatbot to inquire about the return policy and to obtain a recommendation for a laptop that fits her needs. The chatbot uses a hybrid approach combining rule-based and extractive-based systems to handle her queries.

Customer Query 1 about the return policy: “Can you tell me about your return policy?”

The rule-based system detects the keyword “return policy”.

The query is routed to the rule-based component since “return policy” is a common question. The response is generated by the rule-based system using predefined rules.

Chatbot: “Our return policy allows you to return items within 30 days of receipt. Items must be in their original condition with all packaging intact.”

When responding to standard queries, the accuracy is 98%. The response time is 4 s, significantly faster than the previous average of 20 min when handled by human agents.

Figure 3 depicts a practical example of the extractive QA model. Customer Query 2 relates to a recommendation: “Can you recommend a good laptop for graphic design under USD 1500?”

The query involves a complex request requiring an understanding of “laptop for graphic design” and “under USD 1500”. The query is routed to the extractive-based component due to its complexity.

The chatbot tokenizes the query and identifies critical phrases like “laptop for graphic design” and “under USD 1500”, classifies the query as a product recommendation, and uses BERT-based extractive summarization to extract relevant laptop features from a product database.

The chatbot compiles the information using the generative QA model.

Chatbot: “For graphic design under USD 1500, I recommend the Dell XPS 15 with 16 GB RAM, a 512 GB SSD, and an NVIDIA GTX 1650. It offers excellent performance for design tasks and fits your budget.”

The response time is 14 s. The user satisfaction score is 5 out of 5.

5.3. Experimental Results

5.3.1. Chatbot Performance

Performance of rule-based QA method:  A total of 1684 queries were sent to the rule-based QA method.

Table 2. Confusion matrix of the rule-based QA method.

	Predicted Positive	Predicted Negative
Actual Positive	TP = 1102	FP = 32
Actual Negative	FN = 0	TN = 550

presents the confusion matrix of the results. The actual positives represent queries that are in context, while actual negatives are out of context. Independent users asked queries that were classified as actual positives, whereas company experts intentionally asked queries that were classified as actual negatives.

For a user question in the given context that needed to be answered, true positives (TP) occurred when a correct answer was given, and false positives (FP) occurred when an incorrect answer was given. Answering a question that should not be answered (out of context) constituted false negatives (FN), and not answering it constituted true negatives (TN). The rule-based QA method missed answering 32 inquiries out of 1684 that had correct answers. To test the chatbot, 550 out-of-context questions were asked. It identified them as out of context and did not provide answers.

The high accuracy value resulted from the high number of TP and low number of FP, indicating that the chatbot’s answers aligned well with the actual answers. In other words, when the chatbot provided an answer, it tended to be correct, leading to a high overall accuracy rate. The precision of the rule-based QA method was 0.98, computed by $(TP+TN)/(TP+TN+FP+FN)$.

The high number of true positives (TP) and low false positives (FN) contributed to a high precision value, indicating that when the system provided an answer, it was highly likely to be correct. The precision of the rule-based QA method was 0.97, computed by $TP/(TP+FP)$.

Performance of generative-based QA method:  A total of 1803 votes were collected for the generative-based QA method, with respondents assigning scores based on their subjective assessments. Figure 4 breaks down the distribution of these scores, revealing that a significant majority of participants, 52.1%, rated the system as “excellent” (score 5), while 39.0% found it “good” (score 4). Additionally, 4.9% rated it as “average” (score 3), 2.5% as “poor” (score 2), and 1.4% as “very poor” (score 1). The average score, calculated as 4.38, indicates that, on average, respondents considered the system highly effective, with the majority expressing a favorable view by assigning a score of 4 or 5.

5.3.2. User Satisfaction

General Satisfaction: Figure 5 presents the Likert scores and corresponding survey responses, which indicate the level of satisfaction with the chatbot’s performance. It is evident that the majority of respondents rated the chatbot positively. The average score was 4.29, whereas approximately 54% of the participants gave the highest rating of 5, signifying a high degree of satisfaction. Additionally, a substantial number of respondents provided scores of 4, accounting for 31% of the total votes. This suggests that the chatbot was generally well received and met or exceeded the expectations of a significant portion of users. However, it is worth noting that there were also responses in the lower score ranges, indicating room for potential improvement. Overall, the data reflect a positive sentiment regarding the chatbot’s performance, with the majority of users expressing satisfaction.

Improvement in communication time: Before using the chatbot, respondents reported spending significant time on communication-related tasks. The comparison between the time spent previously and the time spent using the chatbot revealed a notable advantage in favor of the chatbot’s efficiency and convenience. For instance, many users reported spending around 20 min on average for communication-related inquiries. In contrast, the average response time per chatbot question was only 41 s. This significant reduction in response time showcases the chatbot’s ability to streamline and expedite communication processes.

Impact on the number of calls: While the chatbot reduced communication time, regarding the impact on staff workload,

Table 3. Statistical analysis of chatbot’s impact on the number of calls.

Group Statistics of Number of Calls
Case	N	Mean	Std. Deviation
Before	500	16.35	3.782
After	500	16.09	3.231
T-test for Equality of Means of Number of Calls
Variances are assumed…	t	df	Sig. (2-tailed)	Mean Difference
Equal	1.151	998	0.250	0.256
Not equal	1.151	974.214	0.250	0.256

shows that there was not a significant change in the number of incoming calls or inquiries. This observation aligns with the chatbot’s specific focus on addressing information-related queries. Since the chatbot primarily handles inquiries for informational purposes, it is unlikely to affect the volume of calls the staff receives directly.

The stability in the number of incoming calls suggests that the chatbot effectively shouldered the responsibility of answering routine and frequently asked questions, allowing staff to attend to more complex or specialized tasks. This outcome indicates a successful integration of the chatbot into the system, as it enhanced efficiency without overwhelming or reducing the need for human staff.

6. Conclusions, Discussion, and Future Directions

This study presents a comprehensive examination of implementing a chatbot to address communication challenges in e-business. The chatbot was designed to provide answers to customer inquiries, focusing on non-sensitive information. This study employed both rule-based and generative-based question-answering methods and evaluated the chatbot’s performance through expert evaluations and user satisfaction surveys.

The findings of this study have significant implications for e-business communication and customer service practices. The developed chatbot system demonstrated high accuracy and user satisfaction rates, indicating its potential to greatly enhance customer interactions. The rule-based QA method showed high accuracy and precision, effectively answering many user queries. The generative-based QA method also received positive evaluations from users, with a majority rating it as “excellent” or “good.” Users reported a substantial reduction in communication time, significantly improving efficiency in obtaining information. The immediate and precise responses provided by the chatbot can dramatically reduce customer wait times and improve overall satisfaction, which is critical for maintaining brand loyalty and positive customer experiences. By automating routine inquiries, businesses can allocate human resources to more complex and value-added tasks, thereby increasing operational efficiency and reducing costs. Furthermore, the results suggest that the chatbot’s introduction led to a streamlined process for information-seeking inquiries while maintaining consistency in the workload of the hospital staff, as there was not a notable change in the number of incoming calls. Instead of replacing human staff, the chatbot complemented their workload by handling routine inquiries, allowing staff to focus on more complex tasks. This integration strategy proved effective in maintaining the quality of service while improving efficiency.

Implementing a chatbot on e-business platforms raises several important points for discussion. This study highlights the significance of human–machine collaboration in e-business settings. The chatbot’s ability to handle routine inquiries frees up staff to concentrate on tasks that require human expertise, such as customer emotions and complex inquiries. This collaboration allows for more efficient resource allocation and enhances overall e-business service quality.

The chatbot’s impact on communication time and user satisfaction emphasizes the importance of customer-centered action. Timely and efficient communication with customers contributes to a positive e-business experience. The chatbot supports customer empowerment and engagement in business management by providing quick answers to common queries.

The successful integration of the chatbot without a significant increase in the number of incoming calls indicates that the solution is scalable. It can potentially be extended to other organizations facing similar challenges, thereby improving communication and reducing workloads across the e-business.

The quantitative results of this study highlight substantial improvements in customer satisfaction and communication efficiency due to the implementation of the hybrid chatbot system. The rule-based chatbot achieved impressive accuracy and precision rates, while the extractive-based chatbot received high user satisfaction scores. The integration of these chatbots led to a notable increase in user satisfaction, with an average Likert score of 4.29 and over half of the participants giving the highest satisfaction rating. Additionally, the average response time to customer inquiries was significantly reduced from 20 min to just 41 s, showcasing the chatbot’s efficiency in handling customer interactions. These outcomes underscore the chatbot system’s effectiveness in enhancing e-business communication by providing quick, accurate, and satisfying responses to customer queries.

While this study demonstrates the chatbot’s effectiveness, it also highlights the need for continuous improvement. Some users provided lower satisfaction scores, indicating areas for refinement. Regular updates, incorporation of user feedback, and expansion of the chatbot’s knowledge base can lead to ongoing enhancements in its performance.

Several areas can be explored in future research to further enhance e-business communication. One potential area is exploring the chatbot’s performance using various language models and embedding techniques. Evaluating how different models impact the chatbot’s effectiveness in responding to queries would provide valuable insights. Expanding the chatbot’s capabilities to include multilingual support can help businesses cater to a global audience more effectively. Another area of interest is the personalization of chatbot interactions based on user data and behavior, which can lead to more tailored and engaging customer experiences. Lastly, conducting longitudinal studies to assess the long-term impact of chatbot implementation on customer satisfaction and business performance would provide valuable insights into the sustained benefits and potential challenges of using AI-driven communication tools in e-business.

The current focus has been on information-seeking queries. It should be noted that the introduction of the chatbot did not lead to a change in the workloads of company personnel due to the excessive volume of calls. Future efforts could aim to increase the adoption of assistive services among customers, potentially reducing call volumes and thereby alleviating staff workload.

Categorizing incoming queries could improve business management and resource allocation. This would allow for preparing reports to aid in planning regulatory and preventive activities.